Auto-campaign search terms with >2 orders + ROAS > target that have not yet been promoted to manual targeting. Promote weekly.
At a glance
List of search terms found by Auto Targeting campaigns that converted well (≥2 orders + ROAS > target) and have NOT yet been promoted to manual keyword targeting. The structured-harvest workflow, auto-campaigns discover converting queries; manual campaigns exploit them at tighter bids. Promoting weekly is the canonical Amazon Ads optimisation rhythm. Industry rule: a healthy account harvests 5-15 candidates per week.
| The formula | Filter Auto-campaign search terms over the last 7 days where: (a) attributedConversions14d >= 2, (b) ROAS > target_ROAS (default 3×, configurable), (c) the term is NOT already a manual keyword in any campaign in the same account. The unpromoted-status check is the key, prevents re-flagging already-harvested terms. |
| Reports API endpoint | POST /reporting/reports with reportTypeId=spSearchTerm filtered to Auto campaigns. Plus GET /sp/keywords to enumerate already-promoted manual keywords (for the de-duplication). |
| What “promote” means | Add the search term as an exact-match or phrase-match keyword to a manual Sponsored Products campaign, then reduce the auto campaign’s bid on that term (or add it as a campaign-negative on the auto campaign so the auto stops competing with the manual). |
| ACOS vs ROAS framing | The card filters on ROAS > 3× by default (= ACOS < 33%). Threshold is account-configurable; some accounts use 4× (ACOS < 25%) for tighter discipline. |
| Attribution model | Last-click within Amazon ecosystem, 14-day click window. Same as ACOS / ROAS calculations. |
| Brand vs non-brand keyword scope | The card includes both. Branded harvests are usually unwanted, promoting a branded query you already win organically just costs you more per click. Filter branded out using Branded Harvest Candidates. The non-branded harvest is the high-value list (Non-Branded Harvest Candidates). |
| Sponsored Products vs Brands vs Display vs DSP | The harvest workflow is specific to Sponsored Products (SP). SB, SD, DSP don’t have the same auto-discovery pattern; their targeting is set up-front. |
| Currency | Account currency only. |
| Amazon-only attribution gap | The candidate list is generated from on-Amazon click + conversion data. Not affected by the cross-channel attribution gap. |
| Time window | 7D. Weekly rhythm matches the recommended harvest cadence. |
| Alert trigger | >0 candidates (weekly review). Any non-zero count is the signal that a harvest sweep is overdue. |
| Roles | owner, marketing, agency |
Calculation
Calculated automatically from your Amazon Ads data. See the At a glance summary above for what the metric tracks and the worked example below for a typical reading.Worked example
A US home-goods seller, weekly harvest review on 12 Apr 26 covering the 7 days from 06 Apr to 12 Apr. The card surfaces 8 candidates.| Search term (from Auto campaign) | Orders (7D) | Spend | Sales | ROAS | Branded? | Action |
|---|---|---|---|---|---|---|
| ”cooling memory foam pillow” | 6 | $48 | $282 | 5.9× | non-branded | Promote, exact match. |
| ”king size cooling pillow” | 4 | $36 | $176 | 4.9× | non-branded | Promote, exact match. |
| ”hot sleeper pillow” | 5 | $42 | $215 | 5.1× | non-branded | Promote, phrase match. |
| ”bamboo pillow case set” | 3 | $28 | $112 | 4.0× | non-branded | Promote, exact match. |
| ”breathable mattress topper” | 3 | $35 | $135 | 3.9× | non-branded | Promote, phrase match. |
| ”[brand-name] memory foam pillow” | 4 | $12 | $176 | 14.7× | branded | Skip, already winning organically. |
| ”down alternative pillow” | 2 | $24 | $76 | 3.2× | non-branded | Promote, exact match (just over threshold). |
| “pillow for side sleepers” | 2 | $19 | $58 | 3.1× | non-branded | Promote, exact match (marginal). |
- The 7 non-branded candidates take ~15 minutes to promote. Add each as an exact or phrase keyword to a manual SP campaign with a starting bid 10-20% below the auto-campaign’s average CPC. Then add the term as a campaign-level negative on the auto campaign so the auto stops competing.
- The branded candidate is intentionally skipped. Promoting “[brand-name] memory foam pillow” costs more per click than letting it convert organically. Branded harvests are a trap, they inflate ACOS / ROAS visibly while not adding incremental revenue.
- Weekly cadence matters because the harvest window is short. A converting auto-term not promoted within 14 days starts to lose its statistical edge as the auto campaign’s bid optimises elsewhere. The 7-day rhythm catches them while the signal is fresh.
- The 3.0× ROAS threshold is generous. Some accounts tighten to 4×; others extend to 2.5× to be more aggressive about harvesting. The right threshold equals your category’s
target_ROAS(= 1 / target_ACOS). - Promoting + auto-pause is the full pattern, just promoting the keyword doubles your spend on the term (auto + manual both bid). Adding the auto-side negative is the second half.
- 5-15 candidates per week = healthy auto-campaign discovery cadence.
- 0 candidates: either the auto campaigns are too restrictive (loosen targeting) or the team is harvesting daily already.
- 30+ candidates: backlog of unharvested wins. A one-time bulk sweep is needed; resume weekly after.
- Same terms reappearing weekly: someone is promoting them but not adding the auto-side negative, auto and manual are competing.
- Mostly branded candidates: the auto campaigns are over-indexed on branded discovery; tighten audience signals.
Sibling cards merchants should reference together
| Card | Why pair it with Harvest Queue |
|---|---|
| Branded Harvest Candidates | The split, branded harvests are usually NOT what you want to promote. |
| Non-Branded Harvest Candidates | The high-value half of the queue. |
| Search Terms Ready to Promote | An adjacent view of the same workflow. |
| Harvest Velocity Trend | Weekly cadence health, are you keeping up with the discovery rate? |
| Oldest Unpromoted Harvest Term | The longest-pending candidate, an SLA check on the harvest workflow. |
| Promotion Conversion Lift | Does promoting actually improve ROAS, or are auto and manual identical? |
| Search Terms Report | The full search-term universe (not just harvest candidates). |
| Negative-Keyword Candidates | The opposite end of the workbench, terms to negate, not promote. |
Reconciling against the vendor’s own dashboard
Where to look in Amazon Ads Console: Amazon Ads Console > Campaign Manager > [Auto Campaign] > Search Terms tab, the Search Terms tab on any Auto campaign shows discovered queries with conversion data. Sort by Orders descending and filter to ROAS > 3 to manually replicate this card. Amazon Ads Console > Recommendations. Amazon’s “Add as keyword” suggestions overlap heavily with this card but use Amazon’s own threshold (typically more permissive). Useful as a cross-check. Amazon Ads Console > Reports > Sponsored Products > Search Term Report, the canonical search-term-level report. Filter to Auto campaigns + Orders >= 2 + ACOS < target. Why our number may legitimately differ from Amazon’s “Recommendations”:| Reason | Direction of divergence | Why it happens |
|---|---|---|
| Threshold definition. Amazon’s recommendations use a proprietary scoring model; this card uses explicit thresholds (≥2 orders, ROAS > target). | Amazon may suggest more candidates (lower-bar scoring) or different ones. | Amazon’s model includes click-volume and CTR signals; this card uses outcome-only signals. |
| De-duplication scope. This card excludes terms already manual-keyworded anywhere in the account; Amazon’s UI may suggest adding to a different campaign even if already promoted. | This card surfaces fewer candidates. | Intentional, avoids re-promoting. |
| Branded filter. This card surfaces both branded and non-branded; the branded list is shown but flagged as skip-recommended. | Same. | Both views are available. |
| Timezone. Both use PT (Pacific). | None. | Amazon’s reporting backend is in Seattle. |
| Report-generation latency (1-3 hours). | Today’s queue is provisional for ~24h. | Reports batch-build hourly. |
| Card | Expected relationship | What causes legitimate divergence |
|---|---|---|
amazon_sp.amzn_sp_total_sales | A successful harvest typically lifts attributed revenue without raising spend (manual bids are tighter than auto). | None expected. |
google_ads.gads_search_terms | Cross-platform analogue. Promoting auto-discovered terms is a Google-Ads PMax pattern too. | Independent platforms; no direct relationship. |
shopify.total_revenue | No relationship. | N/A. |