> For the complete documentation index, see [llms.txt](https://docs.sellermagnet.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.sellermagnet.com/ppc-management-beta/ppc-manager-overview/ppc-ai-optimizer.md).

# PPC AI Optimizer

Review AI-ranked PPC recommendations, wasted-spend diagnostics and optional automation from a single 8-tab workspace, then choose which campaign changes to apply.

{% hint style="info" %}
**Difficulty:** 🔴 Advanced · **Reading time:** \~25 min
{% endhint %}

{% hint style="success" icon="rocket" %}
**Open this page in your dashboard:** [**Go to PPC AI Optimizer \[BETA\] →**](https://dashboard.sellermagnet.com/ppc/ai-optimizer)
{% endhint %}

{% hint style="warning" icon="triangle-exclamation" %}
**BETA:** The PPC Manager is currently in beta. Some features may behave unexpectedly or change based on user feedback. Please report any issues at <info@sellermagnet.com>.
{% endhint %}

## 📋 Overview

The **AI Optimizer** - shown in the dashboard as the **AI Campaign Optimizer** - is SellerMagnet's most advanced PPC feature. Its subtitle describes the job: *review opportunities, diagnose wasted spend, and choose which campaign changes to apply.* It analyzes your campaign data and turns it into ranked recommendations: keyword-level bid adjustments, keyword harvesting, negatives, placement tuning, budget reallocation, cross-campaign coverage reviews, delivery-health findings and anomaly detection.

The page is built around **review the evidence, then choose**. Every recommendation comes with a confidence indicator, a plain-language explanation and an estimated impact, so you can see *why* a change is suggested before you act on it. Treat every estimate as an estimate: price, stock, seasonality, competition and delayed Amazon reports all influence real results, so nothing on this page is a guaranteed saving or a guaranteed uplift.

The whole feature is organized as a **cockpit with eight tabs** - Overview, Recommendations, Wins, Issues, Coverage Audit, Diagnostics, Intelligence and Automation. Each tab opens with a collapsible **"First time here?"** guide (its summary line is the guide title). You can hide all guides at once with **Hide all guides**, and bring them back with the **Restore all page guides** link in the help drawer.

{% hint style="warning" icon="triangle-exclamation" %}
**Apply writes to Amazon directly - it does not go through the Approval Center.** When you click **Apply** on a recommendation card, a confirmation dialog opens with the affected items. Its **Preview (dry-run)** button shows what would be sent to Amazon without changing anything. Confirming the dialog then applies the change to your Amazon Ads account straight away. The [Approval Center](https://docs.sellermagnet.com/ppc-management-beta/ppc-manager-overview/ppc-approval-center.md) holds a different set of items: actions queued by your saved automation mode (for example Manual Approval) and items you queue with a button that says so, such as **Queue negative phrase for approval**. Use **Review queued changes** next to Run Full Analysis to open it.

If **Sandbox mode** is on, an apply becomes a simulation and nothing is sent to Amazon (the Sandbox chip at the top of the page tells you when this is the case).
{% endhint %}

{% hint style="info" icon="circle-info" %}
**What the automation mode means here.** Your saved **automation mode** controls what the optimizer may do on its own. This page names the three modes:

* **Suggestions Only** - nothing is applied on its own. Recommendations appear here and in the Approval Center, and you decide whether to apply each one.
* **Manual Approval** - queues changes in the Approval Center; you batch-apply when ready. A separate opt-in (off by default) lets Manual Approval apply conservative bid changes automatically.
* **Full Autopilot** - eligible changes can apply automatically after the required checks. Review the Activity Log; undo availability depends on the action.

Destructive actions - pausing and negating - always need your review, even in Full Autopilot. The mode can also fall back to Suggestions Only on its own if a kill switch is on, the PPC automation terms are not accepted, a goal guard trips (for example an implausible target), or your settings cannot be read. You set the mode in the governance card on the [Automation tab](#automation-tab) and in PPC Settings.
{% endhint %}

{% hint style="success" icon="shield-halved" %}
**Undo is available for most supported changes.** Within 24 hours, bids, budgets, placements, pause/enable and single negatives can be undone from the [Activity Log](https://docs.sellermagnet.com/ppc-management-beta/ppc-manager-overview/ppc-activity-log.md). Harvested keywords and new campaigns cannot be undone automatically. The Activity Log is the place to check what was recorded and which undo options remain.
{% endhint %}

> **Prerequisites:** Before relying on the AI Optimizer, review the guardrails in the governance card on the [Automation tab](#automation-tab) (Optimization Goal, Target ACoS or ROAS, Max Bid Cap, Min Spend Threshold, and automation mode). They steer which changes are proposed and bound them, but they are not a real-time spending ceiling (see [Where Settings Live](#where-settings-live)).

***

## 🎥 Video Walkthrough

{% hint style="warning" %}
**Video tutorial coming soon.** A full guided walkthrough of this feature will be available here shortly.
{% endhint %}

***

## 🧭 The Cockpit Hero

At the top of the page sits the command hero. It carries the page title and subtitle, a status pill, the **PPC glossary** button, the **Analysis Workspace** controls and a collapsible **Analysis summary**.

* **Status pill** - sits beside the title. It reads **Ready** before a run, **Running** during one, then **Live** when results are in, **No Data** when the analysis returned nothing, or **Error** if it failed.
* **PPC glossary** - opens a searchable popover with about 21 terms used on the page (ACoS, TACoS, ROAS, CTR, CVR, CPC, AOV, Attribution, Harvest, Negate, Thompson Sampling, Marginal ROAS, Confidence floor, Automation mode, Dayparting, Share of Voice and more).
* **Analysis Window** - the lookback the on-demand analysis uses: **Last 7 / 14 / 30 / 60 / 90 days** (default 30). Changing the window re-scores the portfolio.
* **Profile** - scopes the analysis to a single Amazon Ads profile (the default shows all marketplaces).
* **Run Full Analysis** - the workspace stays on a **Ready to Optimize** screen until you press this. A run shows a 3-phase progress panel and "This usually takes 20-60 seconds." It refreshes every module on the page (recommendations, harvesting, negatives, placements, diagnostics, intelligence and the automation cards), so the tabs fill in together. In Fully Automatic mode, this on-demand run honors your automation switches exactly like the scheduled cycle: **"Apply harvests without asking"** and **"Let the autopilot apply bid cuts, even with little data"** (both in [PPC Settings](https://docs.sellermagnet.com/ppc-management-beta/ppc-manager-overview/ppc-settings.md) > Automation) apply on the spot instead of waiting for the next scheduled cycle.
* **Review queued changes** - a button beside Run Full Analysis that opens the [Approval Center](https://docs.sellermagnet.com/ppc-management-beta/ppc-manager-overview/ppc-approval-center.md).

Under the controls, the collapsed **Analysis summary** shows three headline stats that update after each run:

| Stat                 | What it shows                                                                                                 |
| -------------------- | ------------------------------------------------------------------------------------------------------------- |
| **Portfolio Health** | A single 0-100 score for the whole account (detailed under [Portfolio Health score](#portfolio-health-score)) |
| **Open Actions**     | Recommendations plus anomalies awaiting your review                                                           |
| **Projected Upside** | Estimated savings plus growth opportunity from the current recommendations - an estimate, not a guarantee     |

{% hint style="warning" %}
**Mixed currencies.** If your account spans multiple marketplace currencies, a chip above the results explains how the totals were built. When exchange rates are available it reads "Totals span {n} currencies ({list}) and are converted to {base} at the {date} exchange rate - native per-marketplace figures below." When no conversion is available, the totals are flagged as not directly comparable. Either way, the **Show per-marketplace breakdown** link or a single-profile selection gives you the native figures.
{% endhint %}

### Status banners and safety controls

A status line above the hero keeps the page honest about data and safety:

| Item                    | What it tells you                                                                                                                                                                                                                                                                                                                                                      |
| ----------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Sandbox chip**        | Shown when sandbox mode is on: "changes are not sent to Amazon", with a **Go live in Settings** link                                                                                                                                                                                                                                                                   |
| **Stop all automation** | One click to stop automation. The confirmation reads: "Autopilot, bid rules, dayparting, scheduled changes and the budget brake stop sending changes to Amazon right away. Suggestions still reach the Approval Center. You can resume at any time." While a stop is on, a banner says who stopped it and offers **Resume automation** when you are allowed to lift it |
| **Data freshness**      | A freshness chip, plus a banner when your Amazon Ads reports have not refreshed for a while ("Figures on this page can be out of date until the next report sync")                                                                                                                                                                                                     |
| **First sync**          | A banner while your advertising data is first loading, and a failure banner with **Reconnect to retry** and **Contact support** if the first sync failed                                                                                                                                                                                                               |
| **Connect gate**        | If Amazon Ads is not connected, a "Connect Amazon Ads to see your campaigns" card with the connect button                                                                                                                                                                                                                                                              |

### Recent activity and automation

Below the hero, a collapsed **Recent activity and automation** section holds three things:

* **Notices** about how your suggestions are being used. **Suggestions need review** shows your last-30-days apply picture (suggestions applied, surfaced and expired unreviewed) and points you to the Approval Center. When the latest worker notice reports limited suggestion generation, a **Generation limits reported** notice with the same numbers appears near the top of the results instead. Both can be dismissed.
* **The trust strip**, which starts hidden and can be dismissed: "Your saved mode controls automatic execution and approval requirements. Check each pending action's expiry and the Activity Log for recorded changes and available undo options."
* **AI Pulse**, a live panel with today's counters (**Bid moves**, **Harvested**, **Negated**, **Est. monthly savings**) and a **Latest moves** list that links to the Activity Log.

***

## The Eight Tabs

The cockpit splits the work into eight tabs, in this order. Four tabs - Recommendations, Wins, Issues and Coverage Audit - show a count chip when they have items.

| Tab                 | Purpose                                                                                                                          |
| ------------------- | -------------------------------------------------------------------------------------------------------------------------------- |
| **Overview**        | Your daily cockpit - pick your next improvement, work the Smart Recommendations, and open the health, briefing and results block |
| **Recommendations** | The AI's proposed changes, harvesting, negatives, bids, placements and budgets                                                   |
| **Wins**            | Positive campaign observations, recently applied actions and scale opportunities                                                 |
| **Issues**          | Every active issue, severity-ranked - drill in to fix                                                                            |
| **Coverage Audit**  | Cross-campaign review - negative coverage, routing differences and existing exact keywords                                       |
| **Diagnostics**     | Conflict detection, waste control and structural diagnostics                                                                     |
| **Intelligence**    | Pattern mining, momentum, seasonality and share capture                                                                          |
| **Automation**      | Autopilot governance, dayparting, rule preview, benchmarks and profit engineering                                                |

***

### Overview tab

The Overview tab is your daily check-in. Its guide, **Your portfolio at a glance**, reminds you that the KPI strip is computed on the Analysis Window in the hero, so changing the window re-scores everything.

![PPC AI Optimizer - Automated campaign optimization](https://4139654583-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fih2knSvVpBcq5rIeXVjM%2Fuploads%2Fgit-blob-a33c5eec6363fe7474f15d920bfb29ea0c6fd8c2%2Fppc-ai-optimizer.png?alt=media)

#### Choose your next improvement

Four shortcut buttons take you straight to a review list on the Recommendations tab. None of them applies anything:

* **Review bids**
* **Review budgets**
* **Review wasted spend**
* **Find converting keywords**

#### Smart Recommendations

After a run, the **Smart Recommendations** list appears. Use **Refresh** to reload it and the filter chips **All / Critical / High / Auto-Fixable** to narrow it. Cards are ranked by the money at stake and grouped under three headers, each with a count:

* **Cost Savers** - negate, pause and lower-bid moves
* **Revenue Drivers** - harvest, raise-bid and expand moves
* **Cleanup** - everything else

Each card shows:

* An **AUTO-FIX** badge when it is eligible for automatic handling.
* Chips for the estimated monthly impact ("Save" or "Earn" with an amount per month), a confidence percentage, and a category (**Grow sales**, **Save money**, **Improve efficiency**, **Automation**, **Recover profit**).
* Warning chips when something needs a closer look: **Second opinion disagrees**, **Action type underperformed recently**, or **Safety check skipped**.
* An **Evidence and affected items** list and a **Why?** button with the reasoning.
* A snooze button (choose how long to hide the card; it returns when the snooze ends), and **Apply**.
* The footer "Reversible · dry-run preview before commit".

Card states you may see: **Applied** (with **Reset (re-apply)** to unmark it locally), **Apply anyway** when the same action was applied to the same item recently, and a "n already applied" counter with **Clear history**. An item you protected from automation shows a **Protected** badge ("Protected from automation (Settings > Automation) - this suggestion is not applied.").

{% hint style="info" icon="circle-info" %}
**There is no "Apply all" per group.** Each card is applied through its own confirmation and preview, because the groups mix different kinds of action with different safety checks.
{% endhint %}

{% hint style="warning" icon="triangle-exclamation" %}
**Cards with no net profit expected.** When the netted profit impact of a card is zero or negative, it has no one-click Apply. It shows **Review** and the chip "No net profit expected - review before applying" instead.
{% endhint %}

The recommendation types you will meet include:

| Type                              | Typical title pattern                                                                                    |
| --------------------------------- | -------------------------------------------------------------------------------------------------------- |
| Wasted spend                      | "Review {n} keywords with {amount} spent and no orders"                                                  |
| High-ACoS bid cuts                | "Lower bids on {n} high-ACoS keywords to save \~{amount}"                                                |
| Harvest proven terms              | "Harvest {n} proven search terms to boost sales"                                                         |
| AUTO to Manual EXACT              | "Graduate {n} proven AUTO winners to Manual EXACT"                                                       |
| Zero-order negates                | "Review {n} zero-order search terms with {amount} spent"                                                 |
| Watchlist                         | "Watchlist: {n} terms need more evidence" - below the evidence gate, so not counted as projected savings |
| Budget increases                  | "Increase budgets to unlock \~{amount} in additional sales"                                              |
| Converting keywords with headroom | "Review bids on {n} converting keywords with budget headroom"                                            |
| Zero-sales campaigns              | "Review {n} campaigns with zero sales"                                                                   |
| Listing check                     | "{n} keywords get clicks but few orders - check the listing before cutting bids"                         |
| Delivery health                   | A campaign that stopped delivering, collapsing account delivery, or a budget-capped converter            |

Some cards tell you what was held back, for example seasonal search terms kept out of the negate card (a permanent negative would also block them next season), and keywords or campaigns that are too new to judge being kept out of pause and negate cards. A safety check that fell back to a safe default shows as the **Safety check skipped** chip.

#### AI-learned advisory cards

Five Smart Recommendations cards come from patterns the AI has learned from running real advertising accounts, not from a general rule book. All five are advisory: none of them writes to Amazon on its own, and the fix stays a manual step through your existing tools unless noted.

| Card                                                                           | What it flags                                                                                                                                                                                                                                                                                                                                                                                         | What to do                                                                                                                                                                                                                                                                                                                                                   |
| ------------------------------------------------------------------------------ | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| **Product ads stopped serving while a twin SKU of the same product has stock** | An enabled ad is still live, but its impressions over the recent window fell under a small fraction of its usual daily level while another SKU of the same product - usually your own FBA offer - has stock and most likely now holds the featured offer. The ad's status still reads live, so nothing else would tell you. The card shows the before/after impressions-per-day and the twin's stock. | Add the twin SKU shown on the row as a product ad in the same ad group and keep the existing ad - whichever SKU holds the offer then serves. Confidence is medium: an ad can also go quiet a few days before the twin's stock actually arrives.                                                                                                              |
| **Products with a shared word in the title: spend with zero orders**           | A group of your own products that share a word in their titles took a meaningful share of product-ad spend without a single order, even though your conversion rate predicts several orders on those clicks - no single product in the group was big enough on its own to trigger a per-product warning.                                                                                              | Review price, margin and the listing for that group. If it does not pay, pause its product ads in the campaigns that run them. Nothing changes automatically.                                                                                                                                                                                                |
| **New campaigns raised spend without more orders (spend-ramp guard)**          | Three or more campaigns launched in the last 14 days pushed account spend up without a matching rise in orders. The card lists the enabled new campaigns that have not yet proven themselves - newest first.                                                                                                                                                                                          | Cap the daily budget of the newest campaigns, or pause them, and re-open them one at a time once each has enough clicks or its first order. Launching many campaigns at once raises spend before it raises orders.                                                                                                                                           |
| **Marketplaces that need tighter bids than your best one**                     | Compares every marketplace's ACoS against your best-performing one. A marketplace is listed when it runs well above that ACoS, has clicks but no order yet, or is new to advertising.                                                                                                                                                                                                                 | Lower bids there step by step (at most 20% per step) and judge again after two weeks. The Autopilot's own guardrail already backs this up: it drops a bid raise on a cautious marketplace's keyword unless that keyword already has enough orders of its own, so Full Autopilot will not raise bids there on thin evidence even if you never open this card. |
| **Family seasonality: a dated budget-rule proposal**                           | With at least a year of order history, finds a product family's seasonal peak from last year (for example "sold several times its usual weekly units from early to late December") and proposes a Schedule rule - a budget increase from about two weeks before the peak through its end, sized to how strong the peak was.                                                                           | Set the rule up yourself in Amazon's campaign manager. Nothing is scheduled automatically.                                                                                                                                                                                                                                                                   |

{% hint style="info" icon="circle-info" %}
**Where this knowledge also shows up.** The same five patterns, together with two more about generic head terms and new Sponsored Brands/Sponsored Display campaigns needing time to prove themselves, are also given to the AI as background when you ask it questions about your account (the **Ask SellerMagnet** chat in the AI & Planning menu). The chat uses them only to interpret the evidence it reads from your reports - never as evidence on its own.
{% endhint %}

{% hint style="info" icon="lightbulb" %}
**Focused mode.** If your apply rate over the last 30 days is under 5% and your mode is not Suggestions Only, the list shows **Focused mode active**: "You've only applied {rate} of suggestions over the last 30 days. We're now surfacing only our highest-confidence recommendations to respect your time." Otherwise there is no general 80% cut-off - every card carries its own confidence chip, so read it.
{% endhint %}

#### Portfolio health, daily briefing and recent results

The rest of the Overview sits in a collapsed section with this name. Open it to see the following blocks.

**Cockpit highlight cards**

Four cards summarize the account at a glance:

* **Wins this week** - opens the Wins tab
* **Issues this week** - opens the Issues tab
* **Next-best actions** - the top items to work now
* **Automation mode** - shows your saved mode. Its tooltip is explicit: "Shows the configured automation mode. This is not an outcome-based performance score." The card names the mode ("Autopilot is driving", "Semi-Auto (approve)" or "Suggest-only") with a one-line meta. For Manual Approval (shown here as Semi-Auto) the meta reads "Recommendations need approval. Automatic bid changes require a separate opt-in." These are the same three modes described above, worded slightly differently on this card. While no briefing has loaded, the card offers an **Open Governance** button that switches to the [Automation tab](#automation-tab).

**AI Daily Briefing**

A summary, a wins list, a problems list and an **Action Plan**. Some plan items have a **Review candidates** button that takes you to the matching list. Reviewing applies nothing.

**Portfolio Health score**

The cockpit summarizes your account as a single **0-100 score** in a colored ring. It is an **equal-weighted blend of up to five parts**; a part without data is left out:

| Part                     | What it measures                                                                                                     |
| ------------------------ | -------------------------------------------------------------------------------------------------------------------- |
| **Profitability**        | Net ACoS against break-even ACoS from your product costs - or against your target where no product costs are on file |
| **Efficiency**           | The share of spend sitting in campaigns at or under their target ACoS                                                |
| **Waste control**        | How much keyword spend went to keywords with enough clicks and no order                                              |
| **Search-term coverage** | How many converting search terms are already a keyword or a negative                                                 |
| **Delivery**             | Stopped winning campaigns and account-level delivery alerts                                                          |

Click a part to see how it was scored. The ring is **green at 80+**, **blue at 65+**, **amber at 45+** and **red below 45**. A grade (a letter and a word from Exceptional down to Critical) appears with the score. Once at least two daily scores are stored, a small trend chart appears, labelled **Improving**, **Declining** or **Stable** (or **Trend still building**). With too little spend the ring shows "Not enough spend yet to score" - retrying will not change that until more spend accrues.

{% hint style="info" icon="circle-info" %}
**Check your Target ACoS first.** Target ACoS grades the **Efficiency** part. A notice appears when your Target ACoS is 8% or lower and your actual ACoS is more than 1.8 times the target - a strict target flags profitable keywords as too expensive. Profitability uses break-even ACoS when product costs exist, so adding product costs makes the score more meaningful.
{% endhint %}

**AI Analysis Summary**

Alongside the health ring, the summary reports what the latest run found: **Keywords Scored**, **Bid Changes** proposed, **Anomalies**, **Est. Savings**, **Est. Growth**, and the portfolio **Avg Score** (the mean of the per-keyword scores). Est. Savings and Est. Growth are modeled from the recommendations in the run, not measured results.

**Recent AI Actions and the Autopilot Activity Log**

**Recent AI Actions** lists the last changes applied or suggested, linking into the Activity Log. The **Autopilot Activity Log** shows past autopilot runs, with a Refresh button.

**Observed action outcomes**

Formerly called Optimization Impact. It compares dated before-and-after results for actions that have had time to mature. It carries an explicit caveat: "These before-and-after comparisons show association. Price, stock, seasonality and other changes can influence performance; they do not establish the effect of this action alone." You will also see:

* **Results by action type** - a table with **Action**, **Measured win rate** and **Profit impact** columns. Win rate excludes no-change outcomes, and profit impact is a typical (median) change per action type, never a total.
* **Not control-adjusted** / **Too few comparable controls** - labels on comparisons where too few untouched, comparable entities existed to grade a change as an improvement or a decline.
* Notes on chained changes, where several changes on the same item are measured together.
* **Negates that later converted** - how many scored negates went on to convert elsewhere in the account. While that rate is at or above the hold level on a minimum number of scored negates, the autopilot stops applying negates automatically and they wait for your approval.

**Profit results**

The last 8 weeks of ad-attributed profit and total-account profit. It shows **Latest week - ad profit**, **Latest week - total profit** and **Weeks with data**, a legend, and **Autopilot changes in this window** markers (mode and goal changes). It uses rolling 7-day windows and shows association only, not the effect of any single action. Press **Refresh** to load it.

***

### Recommendations tab

The action queue - "Recommendations, harvesting and budget execution." This is where detection turns into execution. Recommendations are ranked by **Expected Impact** (currency saved or earned over the next 30 days). Each card carries its own confidence, so read it before applying; the guide reminds you that some actions cannot be undone.

![PPC Approval Center: review AI suggestions before applying](https://4139654583-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fih2knSvVpBcq5rIeXVjM%2Fuploads%2Fgit-blob-d36088484d6041d51738b7bd9fa2028436d04d87%2Fppc-approval-center.png?alt=media)

{% hint style="info" icon="circle-info" %}
**Applying versus queueing.** Apply, Apply Selected and Apply All open a confirmation first and then write to Amazon (or simulate it in sandbox mode). Items that your automation mode queues, or that you queue deliberately, wait in the [Approval Center](https://docs.sellermagnet.com/ppc-management-beta/ppc-manager-overview/ppc-approval-center.md), shown above.
{% endhint %}

The tab bundles several execution surfaces:

| Surface                          | What it does                                                                                                                                                                           |
| -------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **AI Recommendations**           | The ranked action cards - each shows the change, the projected outcome, the confidence, and the reasoning trail                                                                        |
| **Keyword Harvesting**           | Scan for converting search terms and graduate them into their own exact-match keywords for bid control                                                                                 |
| **Negative Keywords**            | Scan for wasteful search terms (zero conversions or extreme ACoS) and negate them (NEGATIVE\_EXACT / NEGATIVE\_PHRASE)                                                                 |
| **Smart Bid Suggestions**        | Per-keyword bid moves based on Thompson Sampling (a Bayesian explore/exploit policy)                                                                                                   |
| **Profit Forecast (30 Days)**    | A projected profit chart with supporting KPIs                                                                                                                                          |
| **Placement Optimizer**          | Top of Search / Product Pages bid-modifier suggestions from placement-level data                                                                                                       |
| **Budget Allocation**            | Suggested daily-budget shifts toward the highest marginal-ROAS campaigns                                                                                                               |
| **Portfolio Matrix**             | A campaign-level matrix view of your portfolio                                                                                                                                         |
| **Campaign Structure**           | Funnel-structure analysis and recommendations                                                                                                                                          |
| **Anomalies Detected**           | Statistical outliers versus your baseline (spend spikes, CVR drops, impression cliffs)                                                                                                 |
| **Keyword Profitability Scores** | A sortable table of every scored keyword - Score, Grade, Clicks, Orders, Spend, Sales and ACoS. Use the column chooser to show or hide columns and **Export** to download CSV or Excel |

Harvesting and negatives support **Scan → select → Apply Selected**; bid and placement lists support **Apply All**.

**Keyword Harvesting.** Besides candidates, the list tells you what was left out: terms that contain a brand term, terms with characters Amazon does not allow in a keyword (for example commas, slashes or quotes), and duplicate candidates from other campaigns that were merged. A block titled **Converting search terms blocked by your own paused exact keyword** lists terms whose exact keyword you paused earlier, with a **Re-enable keyword** button (the keyword starts spending again at its current bid).

**Negative Keywords.** **Select All** and **Deselect All** help with bulk selection. Terms skipped by a safety check are listed, for example a term that overlaps a brand keyword or would block one of your own active keywords. A **Recurring dead words** block lists words that appear in several search terms and produced no orders in the last 90 days, with the button **Queue negative phrase for approval** (this one does go to the Approval Center). A single-word entry carries a warning: "a negative phrase blocks every future search containing it."

**Smart Bid Suggestions.** Rows carry a High / Medium / Low confidence pill and a Thompson Sampling **Range** (with **pCVR**). When the evidence is thin, a row can read "Hold bid: not enough evidence yet to cut on zero conversions". An **AUTO** badge marks auto-targeting rows, which are managed in the PPC Manager rather than here. A chip can show the campaign's objective (Profit, Growth, Launch, Rank support, Brand defense, Maintain, Liquidate, Stock-constrained or Seasonal): the row is "judged against the objective set on this campaign instead of the account target."

**Profit Forecast (30 Days).** Besides the chart and KPIs it can show a **Likely range** (a band from your own history, not a guarantee) and **Weekly pattern applied** when a recurring weekly pattern was accounted for.

**Placement Optimizer.** Amazon's modifier cap is shown as 0% / 200%, and a suggestion that exceeds it is clamped. A basis chip tells you how the suggestion was sized: **Est. break-even {be}%**, **Target ACoS basis** (no product cost on file), **No margin for ads**, **Break-even not loaded** (no boost is suggested from that card), or **No boost** (your live modifier is left as it is).

**Budget Allocation.** Each campaign shows a state such as **Profit-ranked**, **Budget-limited - raise recommended**, **Lower bids instead** (when a campaign runs out of budget at an ACoS above its target, smaller budgets would only make it run out earlier), **Runs out around {h}:00 local**, **Keep budget**, or **Budget capped · increase on hold**. A budget change that is already waiting in the Approval Center shows as **Raise to {amount} pending approval** (or Cut), with a **Review and apply here** button.

{% hint style="info" icon="circle-info" %}
**Keyword relevance filter.** The same AI learning also filters Amazon's own keyword suggestions wherever SellerMagnet shows them, including the keyword step of the campaign-creation wizard launched from this page and from [Ad Opportunities](https://docs.sellermagnet.com/ppc-management-beta/ppc-manager-overview/ppc-opportunities.md#launching-a-campaign-three-stages). A suggestion is kept only when it has at least two meaningful words, names no other brand from your catalog, and mostly describes your own product (judged against your own product titles, not a fixed word list). The panel tells you how many suggestions were held back ("{n} more suggestions hidden: a single word, another brand, or not matching your product titles"); nothing is silently dropped. The same relevance check holds back a harvested search term from auto-apply when it shares no word with what its ad group advertises - it still appears as a recommendation for your own review, it is just never applied on its own. If your product titles are not on record yet, the filter has nothing to compare against and Amazon's original list passes through unchanged.
{% endhint %}

> **Best Practice:** Sort by Expected Impact and work through the top of the list weekly - recommendations go stale as the account changes. Use **Why?** and the evidence list before applying, and treat low-confidence cards with extra care.

***

### Wins tab

Wins are **observations, not an applied-action ledger**. The tab's guide says it plainly: "These observations describe available campaign evidence. They do not establish that an action was applied or caused a performance improvement." Estimates are not measured savings.

* The hero figure is a **count** of positive observations in the current analysis ("Positive observations in the current analysis. Review the evidence before acting.").
* **Positive campaign observations** lists those observations.
* **Recently applied actions** lists recent changes the optimizer applied, so you can see what was done.
* A panel notes that measured weekly savings are unavailable: a forecast cannot establish historical savings. Review dated action outcomes under **Observed action outcomes** on the Overview tab, and check the Activity Log for recorded changes and available undo options.
* **Scale winners - your next moves** flips to the offensive: opportunities to scale proven winners. The six playbooks are **Raise bid for impression share**, **Increase budget on capped winners**, **Expand match-type on proven keywords**, **Harvest from auto campaigns**, **Lean into top-of-search placement** and **Lock in brand defense**. Each row has **Why?**, **Snooze 7d** and a button. Raising a bid and increasing a budget ask for confirmation in a dialog and then apply directly; the other playbooks open the relevant dialog, editor or page. Statuses include **Applied**, "Simulated · no live change" in sandbox mode, and "Check activity before retrying" when an update could not be confirmed. A row can also show "Changed recently. Wait for attributed sales to settle before increasing again."

{% hint style="info" icon="lightbulb" %}
**Using this with stakeholders.** Present Wins as observations and back any claim about results with **Observed action outcomes** and the Activity Log, which carry dates and the association-only caveat.
{% endhint %}

***

### Issues tab

"Every active issue - drill in to fix." Issues are ranked by severity (**critical → warning → info**) and each row links to the dedicated fix tool in Recommendations or Coverage Audit.

* **Critical** = active spend bleed. **Warning** = inefficiency. **Info** = something to keep an eye on.
* An **Issue activity (last 7 days)** heatmap lets you click a day and type to filter the list; **Show all issue types** clears the filter.
* Rows offer **Fix this** (opens the right tool), **Snooze 7d** (hides the row for 7 days without resolving it) and, where the issue maps to a supported action, an **Apply** button.
* When nothing is open you see **All clear**.
* New findings depend on arriving reports and your saved analysis schedule, and recent conversion data may still change.

{% hint style="info" %}
**Work the queue weekly.** Start with the critical rows, where spend is actively leaking, then move to warnings. Any money figure shown is an estimate, not a measured loss.
{% endhint %}

***

### Coverage Audit tab

"Cross-campaign coverage audit." When the same product is advertised in more than one campaign, single-campaign analysis can miss portfolio-level differences. This tab compares campaigns using the **campaign-to-ASIN map** and is **review-first**: its hero caption reads "Overlap does not by itself establish waste", and its guide notes that differences may be intentional.

| Section                                | What it shows                                                                                                                                                                                     |
| -------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Negative-coverage gaps**             | A search term negated in one campaign but not in a sibling advertising the same product. A negative that suits one campaign may not suit another, so review each destination campaign first       |
| **Cross-campaign routing differences** | A positive keyword in one campaign and a negative in another. This "can support an intentional strategy", for example when one campaign is meant to catch a term and another is meant to block it |
| **Existing exact-keyword coverage**    | A search term you are about to harvest that already matches an exact keyword in another campaign. Compare campaign goals, bids and performance before adding or changing keywords                 |

For negative gaps you can apply a negative across the uncovered siblings of one term, or use **Apply across all uncovered** for everything. The match type is shown **per term-campaign pair**, not as one blanket match type, and the confirmation reads: "Review {n} term-campaign pairs and their displayed negative match types before applying." After applying, the page reports how many negatives were confirmed; if a result cannot be verified it asks you to review the Activity Log before making another change. In sandbox mode the action is a preview only. The routing-difference and existing-keyword sections carry no one-click fix - use the review links where shown to compare campaign search terms and decide for yourself.

{% hint style="info" icon="lightbulb" %}
**How to read this tab.** Negative gaps are the only section with a bulk action, which is why they are easy to act on - but check each pair before you apply. For routing differences and existing exact keywords, compare what each campaign is for before changing either one.
{% endhint %}

***

### Diagnostics tab

"Conflict detection, waste control and structural diagnostics." Diagnostics is **read-only** - it explains *why* a campaign drifted so you can fix the structure before scaling budget. A **System health** strip at the top shows report freshness, sync health and autopilot runs per day.

| Panel                     | What it detects                                                                                                                                                                               |
| ------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Duplicate Keywords**    | The same keyword active in multiple campaigns - bidding against yourself can raise CPC. Overlap may be intentional, so compare targeting, campaign goals and performance before consolidating |
| **ASIN Cannibalization**  | Campaigns where multiple products compete for the same budget                                                                                                                                 |
| **Budget Waste Analyzer** | Forensic waste analysis: zombie keywords (10+ clicks, 0 orders, 30+ days), black holes, diminishing returns, budget-burn campaigns and Pareto violations                                      |
| **Keyword Conflicts**     | Negative keywords blocking profitable terms, match-type cannibalization, and bid disparity across campaigns. Check the campaign, ad group and match type before changing either keyword       |

{% hint style="info" icon="lightbulb" %}
**Diagnose before bidding.** Structural problems limit how much any bid change can help. Fix real duplicates and conflicts first, then watch ACoS over the following weeks. Diagnostics use the reports available when the analysis runs, so report arrival and your saved settings determine how fresh they are.
{% endhint %}

***

### Intelligence tab

"Pattern mining, momentum and share capture." These are strategic, mostly read-monthly modules - they need roughly 28+ days of data to be statistically reliable, so treat earlier runs as directional. A short narrative at the top ("What the AI thinks right now") summarizes the biggest near-term opportunity.

| Module                          | What it tells you                                                                                                                   |
| ------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------- |
| **Search Term N-Gram Analysis** | Breaks search terms into 1-, 2- and 3-word patterns and flags which combinations convert versus waste                               |
| **Performance Trends**          | Period-over-period comparison with a momentum score                                                                                 |
| **Share of Voice**              | Estimated impression share per keyword - the closer to 100%, the less headroom to scale                                             |
| **TACoS Tracker**               | Total ACoS (ad spend over organic + ad sales) - tracks ad dependency and the organic/paid split                                     |
| **ASIN Performance Matrix**     | Cross-campaign, per-ASIN P\&L with a classification of Star / Strong / Average / Underperformer                                     |
| **Keyword Gap Finder**          | Converting search terms not yet targeted as manual keywords                                                                         |
| **Brand vs Generic Spend**      | What share of spend defends your own brand versus hunts new customers (needs brand terms set in PPC Settings → Enterprise Features) |
| **Seasonality Insights**        | Day-of-week patterns, weekly cycles and monthly drift in your daily sales, plus upcoming events and seasonal campaigns (below)      |
| **Day-of-Week Bid Multipliers** | A per-weekday bid multiplier computed from the last 8 weeks of conversion data. Multipliers are capped between 0.5x and 1.5x        |

**Seasonality Insights** also includes:

* **Upcoming Events** - calendar events such as Prime Day or BFCM, matched against your own campaign history.
* **Campaigns we've identified as seasonal** - with a **Schedule for the season** button. Scheduling pauses the campaign now if it is running and re-enables it automatically when the event begins, so budget is not spent before the season starts. You confirm first.
* **Scheduled Campaign Windows** - your upcoming and active schedules, each with a **Cancel** button. Cancelling stops the automatic pause and re-enable; the campaign's current state is not changed.

{% hint style="info" icon="lightbulb" %}
**Read this monthly, not daily.** N-gram patterns answer "what language do customers really use?"; Share of Voice answers "how much headroom is left?"; the TACoS trend answers "is growth healthy or paid-only?" TACoS rising while ACoS falls is healthy; both rising means ads are propping up flat organic sales.
{% endhint %}

***

### Automation tab

"Dayparting, rule engines and profit engineering." This is advanced orchestration. Set sensible limits before turning anything on, and remember that **Stop all automation** (see [Status banners and safety controls](#status-banners-and-safety-controls)) stops automatic changes immediately.

#### Governance: Autopilot mode, goals and guardrails

The **Campaign Autopilot** card, marked with a **GOVERNANCE** badge, sits at the top of the Automation tab under the heading **Autopilot mode, goals and guardrails**. **It is collapsed by default** - open it to see the controls. This is where you configure how the optimizer acts on this page, together with the related settings in PPC Settings. Change them only when you have a clear reason.

The card explains itself: "Autopilot reviews a limited set of recent keyword, search-term and campaign reports on your chosen schedule. Eligible actions depend on your mode, evidence and approval requirements. Caps constrain changes and estimated added exposure; delayed spend reports mean they cannot guarantee a real-time spending ceiling."

| Control                 | Options / meaning                                                                                                                                                             |
| ----------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Mode**                | Suggestions Only · Manual Approval · Full Autopilot (see the legend in the Overview section above). A status badge shows **FULL AUTO**, **SEMI-AUTO** or **SUGGESTIONS ONLY** |
| **Optimization Goal**   | Target ACoS · Target ROAS · Maximize Profit · Maximize Revenue · Maximize Impressions                                                                                         |
| **Target ACoS %**       | Your ACoS target. Shown for every goal except Target ROAS                                                                                                                     |
| **Target ROAS**         | Your ROAS target. Shown only when the goal is Target ROAS                                                                                                                     |
| **Max Bid Cap**         | Bounds proposed base bids                                                                                                                                                     |
| **Min Spend Threshold** | An evidence requirement: the minimum cost a keyword must have accumulated before the optimizer adjusts it. It is not a spending ceiling                                       |

Press **Save Configuration** to store the settings (a confirmation titled "Apply new autopilot settings?" asks you to continue), then **Run Cycle** to run the optimizer once now. What a cycle does depends on the mode: Suggestions Only and Manual Approval queue or surface recommendations, while Full Autopilot can apply eligible changes after the required checks, with destructive actions held for your review. In sandbox mode a cycle is a preview and nothing reaches Amazon.

{% hint style="warning" icon="triangle-exclamation" %}
**Full Autopilot needs the PPC automation terms.** If you save Full Autopilot without having accepted the PPC automation terms, the mode reverts to Suggestions Only and you see "Full Automatic requires accepting the PPC automation terms first - your project stays on Suggestions Only." A Run Cycle without the terms runs in Suggestions Only mode and says so. Accept the terms in PPC Settings.
{% endhint %}

**Run Cycle in Full Autopilot asks you to confirm first** ("Run a live Autopilot cycle now?"), because it can apply eligible changes to live Amazon campaigns and some actions cannot be undone.

**Last cycle status.** Under the card header, a status line summarizes the most recent scheduled cycle: "Last AI cycle {ago} ago - mode {mode} - {total} actions generated, {applied} auto-applied, {queued} queued for approval." It also tells you what was held back, for example:

* brand-term search terms skipped from harvesting;
* converting terms skipped because their cost per click is above what the product margin can carry;
* negative candidates held because they describe the advertised product;
* pause, negate or bid-cut suggestions held because the keyword has been live for too few days;
* seasonal search terms sent to review instead of being negated;
* a note that the cycle saw only part of your spend or orders;
* a settings contradiction such as "Your automation targets contradict each other: Target ACoS {acos}% implies \~{implied}x ROAS but Target ROAS is {roas}x", with an **Open Settings** link;
* in Full Autopilot, a low apply rate: "Fully Automatic, but only {pct}% of suggestions were auto-applied this cycle: {reason}."

**Last Autopilot Run.** After you press Run Cycle and it produces actions, this panel lists them: the number of actions, bids updated, keywords harvested and negatives added, followed by the first few actions.

> **Pro Tip:** Your **Optimization Goal** tells the optimizer what to prioritize when trading off impressions, sales volume, ACoS and profit. Start in **Suggestions Only** for the first 7 days so you can see what is proposed before letting anything apply automatically.

#### Other automation modules

| Module                                          | What it does                                                                                                                                                                                                                                                                                                                                                     |
| ----------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Autopilot run history**                       | A timeline of the optimization cycles the autopilot ran on schedule                                                                                                                                                                                                                                                                                              |
| **Dayparting Intelligence**                     | Timing patterns by hour and weekday. Estimated hourly views are informational, and when no measured hourly data exists the heatmap spreads daily totals across a typical intraday curve, so use it for relative guidance. An AI schedule needs measured hourly data: "Daily reports do not verify hourly bid multipliers. You can still edit a manual schedule." |
| **Top-of-Search Boost Schedule (auto-applied)** | A manual hourly schedule for one campaign. See below                                                                                                                                                                                                                                                                                                             |
| **AI Budget Rules Engine**                      | Custom IF/THEN rules, for example "IF ACoS > 40% for 7 days THEN reduce bid by 20%". **Preview only** - see below                                                                                                                                                                                                                                                |
| **Competitor Benchmark Intelligence**           | A SWOT view comparing your PPC metrics against industry and **Category Benchmarks** (weekly peer percentiles for your Amazon browse categories)                                                                                                                                                                                                                  |
| **Profit Waterfall Analysis**                   | True per-campaign P\&L: Revenue → COGS → FBA Fees → Referral Fees → Ad Spend → Net Profit                                                                                                                                                                                                                                                                        |
| **Search Term Clusters**                        | Groups search terms into semantic clusters so you can scale profitable themes and negate wasteful ones in bulk                                                                                                                                                                                                                                                   |

**Top-of-Search Boost Schedule.** Dayparting here is **not** a set of hourly bid modifiers. "Pick a campaign and boost its Top-of-Search placement for the hours you choose (0-200%). The worker re-applies the active slot's percentage every hour; it does not automatically revert when the slot ends, and it does not touch your bids, Rest of Search or Product Pages. Sandbox mode and the platform safety clamps still apply." In the grid, "Click a cell to boost Top-of-Search (+10%, max 200%); shift-click to reduce it; right-click to clear (0%)." Boosts are shown against an unchanged baseline; there is no "lower" state. A schedule stores one value per hour of day, so if the same hour has different values on different days you are asked to give it a single value before saving. Suggestions built from measured Marketing Stream hours are "not active until you save it".

**AI Budget Rules Engine.** Preview only. Click **Evaluate** to check your rules against live data now. Nothing here changes bids or budgets automatically; for rules that run on their own, use Bid Rules in Settings.

{% hint style="info" icon="circle-info" %}
**Data needs.** Dayparting needs 21+ days of hourly data to be reliable, and the Profit Waterfall uses your COGS / FBA / referral inputs (defaults are category medians, not your real costs).
{% endhint %}

***

## Where Settings Live

The AI Optimizer does not have a single standalone "Configuration" panel. Its controls are split across three places:

| Setting                                                                                                                            | Where it lives                                                                                                      |
| ---------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------- |
| **Analysis Window** (7-90 days) and **Profile** scope                                                                              | The cockpit hero, next to Run Full Analysis                                                                         |
| **Automation mode, Optimization Goal, Target ACoS/ROAS, Max Bid Cap, Min Spend Threshold**                                         | The **governance card** ("Campaign Autopilot") on the **Automation tab**, collapsed by default                      |
| **Analysis schedule, PPC automation terms, brand terms, sandbox mode, Bid Rules, protected items and other account-level options** | The [**PPC Settings**](https://docs.sellermagnet.com/ppc-management-beta/ppc-manager-overview/ppc-settings.md) page |

{% hint style="info" %}
**Max Bid Cap and Min Spend Threshold ship blank.** The value you see in each field is only a placeholder, and a value is saved only if you type one. Max Bid Cap bounds proposed base bids. Min Spend Threshold is an evidence requirement, not a spending ceiling - set it high enough that the optimizer does not act on 2-3 clicks of data. Because spend reports arrive with a delay, neither field can guarantee a real-time spending ceiling.
{% endhint %}

{% hint style="info" icon="circle-info" %}
**Plan differences.** PPC is available as two modules: the **PPC Manager**, which includes everything, and the **PPC AI Optimizer**, which covers the AI layer alone and conflicts with the Manager if you try to hold both. The Optimizer includes the AI recommendation engine, applies and approvals of AI-originated changes, undo, Ad Opportunities and read-only analytics. Manual campaign edits, bid rules, keyword research, planning, brand-defense analysis and harvest/negate on the Search Terms page are Manager features. Where the split applies to your plan and it includes only the Optimizer, Manager-only controls are greyed out and Manager-only pages bring you back here.
{% endhint %}

***

<details>

<summary><strong>⚠️ Common Mistakes to Avoid</strong></summary>

| Mistake                                                   | Why It Hurts                                                                                                          | How to Fix                                                                                                                          |
| --------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------- |
| Never reviewing AI suggestions (leaving hundreds pending) | Suggestions go stale, and a low apply rate switches on Focused mode, which hides all but the highest-confidence cards | Set a daily or weekly cadence; apply, snooze or ignore each suggestion deliberately                                                 |
| Raising bids on a leaky funnel                            | Duplicates, conflicts and zombie keywords limit how much any bid change can help                                      | Review the Diagnostics and Coverage Audit tabs first, then scale                                                                    |
| Ignoring anomaly alerts                                   | A spend surge or CTR drop can waste hundreds in hours                                                                 | Respond to Critical / High anomalies within 24 hours                                                                                |
| Applying everything without reading the reasoning         | Every apply is a real change on Amazon, and thin evidence gives thin results                                          | Use **Why?**, read the confidence and the affected items, and treat low-confidence cards with care                                  |
| Treating estimates as guarantees                          | Projections assume stable competition, price, stock and attribution                                                   | Compare against **Observed action outcomes** and **Profit results**, which show association only                                    |
| Enabling Full Autopilot without setting limits            | Bids and budgets can swing before you notice                                                                          | Set Max Bid Cap and Min Spend Threshold, and read the last cycle status under the Campaign Autopilot header, before switching modes |
| Expecting results in 48 hours                             | The optimizer needs enough history to be meaningful, and attribution takes time to settle                             | Give it time and several cycles before judging; outcomes are only evaluated after an observation period                             |

</details>

## 🔧 Troubleshooting

### AI Not Generating Recommendations

| Symptom                                             | Likely Cause                                                                                            | Solution                                                                                              |
| --------------------------------------------------- | ------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------- |
| "Insufficient data" or empty tabs                   | Not enough campaign data yet (the page notes the AI needs at least 7 days), or no analysis has been run | Press Run Full Analysis; wait for more data to accumulate and make sure sync is running               |
| "Not enough spend yet to score" on Portfolio Health | The window has too little spend to grade                                                                | Widen the Analysis Window or wait for more spend; retrying will not change it                         |
| Zero pending actions                                | Your mode is Suggestions Only and you have not reviewed the list, or the account is genuinely clean     | Open the Recommendations tab and work the list                                                        |
| Fewer cards than expected                           | **Focused mode** is on because your 30-day apply rate is under 5%                                       | Review the cards and apply the ones you want. The setting follows your activity over the last 30 days |
| Recommendations only for some campaigns             | Other campaigns have too few clicks to clear the Min Spend Threshold                                    | Lower the threshold or wait for more traffic                                                          |
| Analysis looks partial                              | A notice says the last analysis saw only part of your spend or orders                                   | Re-run the analysis, and check the data-freshness banners at the top                                  |

### Nothing Applies Automatically

| Symptom                                                           | Likely Cause                                                                           | Solution                                                                                   |
| ----------------------------------------------------------------- | -------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------ |
| Full Autopilot saved but only suggestions appear                  | The PPC automation terms are not accepted, so the mode reverted to Suggestions Only    | Accept the terms in PPC Settings                                                           |
| Nothing is sent to Amazon                                         | Sandbox mode is on (look for the Sandbox chip), or **Stop all automation** is active   | Go live in Settings, or resume automation                                                  |
| Pauses and negatives keep waiting                                 | Destructive actions always need your review, even in Full Autopilot                    | Review them in the Approval Center or on the Recommendations tab                           |
| Mode is Manual Approval but bid changes do not apply on their own | The separate opt-in for automatic bid changes is off by default                        | Enable it in PPC Settings if you want conservative bid changes applied automatically       |
| Mode keeps falling back to Suggestions Only                       | A kill switch, a goal guard (for example an implausible target) or unreadable settings | Check the last cycle status under the Campaign Autopilot header, and your goal and targets |

### Unexpected AI Behavior

| Symptom                                             | Likely Cause                                                                      | Solution                                                                                                                                                            |
| --------------------------------------------------- | --------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| AI suggesting bids above your comfort level         | Max Bid Cap is higher than you want, or you never set one (the field ships blank) | Type a Max Bid Cap in the governance card                                                                                                                           |
| AI suggesting to pause keywords you want to keep    | Keywords sit far above your Target ACoS                                           | Check your Target ACoS reflects your real margins, snooze the card, or protect the item from automation in PPC Settings. Automation never applies pauses on its own |
| Portfolio Health reads low for a profitable account | A very strict Target ACoS drags the Efficiency part down                          | Raise your Target ACoS if it was not intentional, and add product costs so Profitability uses break-even                                                            |
| Projected impact seems too optimistic               | Projections assume stable competition                                             | Treat projections as directional estimates, not guarantees                                                                                                          |
| Savings and upside totals look off                  | Your account spans multiple currencies                                            | Select a single profile, or use the per-marketplace breakdown                                                                                                       |

***

## ❓ FAQ

<details>

<summary><strong>How long does it take for the AI Optimizer to show results?</strong></summary>

The page needs at least 7 days of campaign data to suggest moves, the Intelligence modules want roughly 28+ days, and dayparting wants 21+ days of hourly data. Judge results with **Observed action outcomes**, which only evaluates an action after an observation and attribution period, rather than by what you see in the first days.

</details>

<details>

<summary><strong>Does Apply go through the Approval Center?</strong></summary>

No. Apply on a card opens a confirmation dialog (with an optional **Preview (dry-run)**) and, once you confirm, writes to Amazon - or simulates it in sandbox mode. The Approval Center holds actions that your saved automation mode queues (for example in Manual Approval) and items you queue with a button that says so. See the [PPC Approval Center](https://docs.sellermagnet.com/ppc-management-beta/ppc-manager-overview/ppc-approval-center.md) guide.

</details>

<details>

<summary><strong>Can the AI Optimizer make things worse?</strong></summary>

Any change to bids, budgets or keywords can hurt performance, and an estimate is not a promise. The safeguards are your own review before each apply, the dry-run preview, sandbox mode, the Max Bid Cap and Min Spend Threshold, and the 24-hour undo window for supported changes. In Suggestions Only and Manual Approval modes the optimizer does not change anything on its own (apart from the opt-in conservative bid changes in Manual Approval). Pauses and negates always need your review. Monitor **Observed action outcomes** and the Activity Log regularly.

</details>

<details>

<summary><strong>Does the optimizer adapt to what I apply?</strong></summary>

In one visible way: if your apply rate over the last 30 days falls under 5% and your mode is not Suggestions Only, the Smart Recommendations list switches to **Focused mode** and surfaces only the highest-confidence cards. Snoozing a card hides it for the period you choose. The optimizer also watches negates that later converted and stops applying negates automatically while that rate is at or above its hold level.

</details>

<details>

<summary><strong>How is the "strategy" set now?</strong></summary>

Rather than a separate strategy picker, the cockpit expresses strategy through the **Optimization Goal** in the governance card on the Automation tab (Target ACoS, Target ROAS, Maximize Profit, Maximize Revenue, Maximize Impressions) together with your Target ACoS/ROAS and automation mode. Pick the goal that matches your priority - visibility, efficiency or absolute profit.

</details>

<details>

<summary><strong>Can I undo a change the AI made?</strong></summary>

For many changes, yes. Every applied change is recorded in the [Activity Log](https://docs.sellermagnet.com/ppc-management-beta/ppc-manager-overview/ppc-activity-log.md), and supported changes (bids, budgets, placements, pause/enable and single negatives) can be undone within 24 hours. Harvested keywords and new campaigns cannot be undone automatically. Note that reverting does not undo the market impact - if a bid was raised for 3 days before reverting, the impressions and spend from those days are not refundable.

</details>

<details>

<summary><strong>Why does a card say "Review" instead of "Apply"?</strong></summary>

When the netted profit impact of a recommendation is zero or negative, the card has no one-click Apply. It shows **Review** and the chip "No net profit expected - review before applying", so you read the reasoning before deciding.

</details>

<details>

<summary><strong>How does keyword harvesting work?</strong></summary>

The optimizer looks at search terms in your auto (and broad/phrase) campaigns that convert well. From the Recommendations tab it suggests graduating those terms into their own exact-match keywords, giving you bid control over proven performers. Terms that contain a brand term, or whose cost per click is above what the product margin can carry, are left out. The Coverage Audit tab also flags a harvest candidate that already matches an exact keyword in another campaign, so you can compare the campaigns before adding another.

</details>

<details>

<summary><strong>Is there a limit to how many recommendations I will see?</strong></summary>

There is no fixed quota. Cards are ranked by estimated money at stake, each carries its own confidence chip, and the count chip on the Recommendations tab shows how many there are. The list gets shorter when Focused mode is on or when evidence is thin (watchlist items are kept out of projected savings).

</details>

<details>

<summary><strong>Which PPC plan do I need?</strong></summary>

Where your plan separates them, the **PPC Manager** includes everything, and the **PPC AI Optimizer** covers the AI layer on its own. An Optimizer-only plan keeps the recommendation engine, AI-originated applies and approvals, undo, Ad Opportunities and read-only analytics, and sends you back to this page if you open a Manager-only page.

</details>

***

## Use Case Scenarios

### Scenario 1: Managing 500 Keywords Across 3 EU Marketplaces

You sell kitchen accessories on Amazon DE, FR and IT with 500 active keywords across 12 campaigns.

**Challenge:** Reviewing bids for 500 keywords across 3 markets by hand is a full-time job.

**Solution:**

{% stepper %}
{% step %}

### Set the goal

Open the governance card on the **Automation tab** and set **Optimization Goal** to Target ACoS (a solid all-round default) with a realistic Target ACoS.
{% endstep %}

{% step %}

### 14-day window

Set the **Analysis Window** to 14 days and run a full analysis per **Profile**.
{% endstep %}

{% step %}

### Full Autopilot with limits

Switch to **Full Autopilot** only after typing a sensible **Max Bid Cap** and **Min Spend Threshold** (both ship blank) and accepting the PPC automation terms.
{% endstep %}

{% step %}

### Watch the results

Check **Observed action outcomes**, **Profit results** and the Activity Log weekly to see whether ACoS is moving the right way. Remember they show association, not proof.
{% endstep %}

{% step %}

### Coverage Audit

Open the **Coverage Audit** to review negative-coverage gaps and routing differences between campaigns, one pair at a time.
{% endstep %}
{% endstepper %}

**Expected Outcome:** Routine bid management across the three markets runs within your limits, and you spend your review time on the pauses and negates that always need your review instead of adjusting bids by hand.

### Scenario 2: Launching a New Product with Zero History

You are launching in the US with one auto campaign and one manual campaign of 30 seed keywords.

**Challenge:** No historical data means the optimizer has little to analyze. You need traffic fast to build keyword data and organic ranking.

**Solution:**

{% stepper %}
{% step %}

### Goal for visibility

Set the **Optimization Goal** to Maximize Impressions (or Maximize Revenue) with a higher Target ACoS - you are investing in visibility, not profit yet.
{% endstep %}

{% step %}

### 7-day window

Set the **Analysis Window** to 7 days to react quickly to early data.
{% endstep %}

{% step %}

### Suggestions Only

Keep **Suggestions Only** for the first two weeks so you see every suggestion while data is thin.
{% endstep %}

{% step %}

### Harvest early

Watch the Recommendations tab's **Keyword Harvesting** to graduate converting auto-campaign terms into exact match, reviewing each candidate first.
{% endstep %}
{% endstepper %}

**Expected Outcome:** After a few weeks you have a reviewed list of harvest candidates from your auto campaign and a clearer view of which seed keywords are not converting. You can then switch the goal toward Target ACoS for the next phase.

### Scenario 3: Agency Onboarding a New Client

You are onboarding a client running Amazon PPC for a year with no optimization tooling. Their ACoS is 45% against a 25% target.

**Challenge:** Accumulated inefficiency - wasteful keywords, outdated bids, no negative strategy. You need to show progress within 30 days.

**Solution:**

{% stepper %}
{% step %}

### Sync data

Connect the client's account and let the data sync.
{% endstep %}

{% step %}

### Goal for efficiency

Set the **Optimization Goal** to Target ACoS (or Maximize Profit) - the priority is cutting waste.
{% endstep %}

{% step %}

### Manual Approval

Use **Manual Approval** so changes the autopilot proposes queue in the Approval Center for your sign-off. Changes you apply yourself from a card still go live after your confirmation.
{% endstep %}

{% step %}

### Review structure first

Work the **Diagnostics** and **Coverage Audit** tabs, then the Recommendations tab's negatives and bid decreases, reviewing each item.
{% endstep %}

{% step %}

### Report with evidence

Present **Observed action outcomes**, **Profit results** and the Activity Log to the client weekly, with the association-only caveat they carry.
{% endstep %}
{% endstepper %}

**Expected Outcome:** The client gets a dated trail of what was changed and how performance moved afterwards. Results depend on the account, so set expectations from the evidence rather than from a promised ACoS.

***

## ➡️ What's Next?

{% content-ref url="/pages/xUaJFWbcCTCya95L64xQ" %}
[PPC Approval Center](https://docs.sellermagnet.com/ppc-management-beta/ppc-manager-overview/ppc-approval-center.md)
{% endcontent-ref %}

{% content-ref url="/pages/0tkou4fwmuJsWMEro64g" %}
[PPC AI connections (MCP server)](https://docs.sellermagnet.com/ppc-management-beta/ppc-manager-overview/ppc-mcp.md)
{% endcontent-ref %}

{% content-ref url="/pages/UR1mVi1OEupeXaCZA7zK" %}
[PPC Activity Log](https://docs.sellermagnet.com/ppc-management-beta/ppc-manager-overview/ppc-activity-log.md)
{% endcontent-ref %}

{% content-ref url="/pages/gVXllc12Kbkb0xuLJfxw" %}
[PPC Settings](https://docs.sellermagnet.com/ppc-management-beta/ppc-manager-overview/ppc-settings.md)
{% endcontent-ref %}

{% content-ref url="/pages/pQTsyjXdDCQAMg0Eueka" %}
[PPC Analytics](https://docs.sellermagnet.com/ppc-management-beta/ppc-manager-overview/ppc-analytics.md)
{% endcontent-ref %}
