Keyword Harvesting

Keyword harvesting is the workflow by which an Amazon advertiser converts raw shopper search queries — surfaced by broad-match and automatic-targeting campaigns — into deliberate, math-bid exact-match keywords in their own dedicated campaigns. It is not a one-time setup task. It is a continuous weekly (or daily) loop that defines the difference between an Amazon PPC account that compounds and one that stagnates.

The economics are simple: a search term bid on at the broad level is paying a discovery premium for the privilege of finding it. Once it's identified as a winner, that same search term can be re-bid at the exact level with a math-derived bid tuned to its specific CVR — typically capturing the same conversion at 30–60% lower ACOS.

The full harvest loop

Auto campaign  →  Broad-match manual  →  Exact-match manual
     ↓                    ↓                       ↓
  search term         search term            negate harvested
  report mines        report mines           terms in upstream
  for winners         for winners            campaign

Step by step:

1. Run discovery campaigns. An Auto campaign (no keyword input, Amazon decides) and a Broad-match manual campaign (your seed keywords, broad match) are the two discovery engines. Both deliberately accept a higher ACOS than the rest of the account because their job is to find keywords, not to convert efficiently.

2. Pull the Search Term Report. Weekly minimum; daily during launches and peak events. Filter to terms with: ≥3 orders, ACOS below a defined "harvest threshold" (typically your Target ACOS × 1.2), and CVR above the campaign's rolling average.

3. Promote winners to exact-match. Add each qualifying search term as an exact-match keyword in a dedicated manual campaign (often a single "Harvested Exact" campaign per SKU or per theme). Set the bid via ASP × CVR_observed × Target ACOS.

4. Negative-exact the upstream. Add the harvested term as a Negative Exact keyword in the source Auto and Broad campaigns. Without this step, the broad-match parent keeps bidding on the term, paying its discovery premium in perpetuity.

5. Monitor 14-day post-harvest. A harvested keyword that bombs in exact-match (CVR collapses, ACOS doubles) usually means the original conversions were noise from a small sample. Pause and re-test rather than chasing the bid down.

Why this loop matters

• Bid precision. A broad-match bid is necessarily averaged across hundreds of search terms with wildly different CVRs. An exact-match bid is tuned to one term. The math gets sharper by an order of magnitude.
• ACOS reduction. A typical account harvesting consistently sees its blended ACOS decline 15–25% over 3–6 months at constant or growing spend.
• Defensive moat. Once a term lives in your exact-match campaign at the optimal bid, competitors can only out-bid you by destroying their own ACOS. The keyword becomes a defensible margin pool.
• Organic flywheel. Harvested exact-match terms typically also rank organically (the inverse correlation between paid CVR and organic rank is one of Amazon's better-documented patterns). The harvest loop quietly funds the organic ranking layer.

Cadence and thresholds

A defensible default for accounts under 1,000 clicks/day:

• Pull cadence. Weekly Search Term Report on Mondays for the trailing 7 days.
• Significance threshold. ≥3 orders OR ≥30 clicks. Below either, ignore.
• Promotion threshold. ACOS ≤ Target ACOS × 1.2 AND CVR ≥ campaign rolling average.
• Negation threshold. Search terms with ≥15 clicks and 0 orders → Negative Exact in all campaigns.

For accounts running Amazon Marketing Stream, this loop accelerates: hourly data means a launch SKU can have its first harvest cycle complete within 48 hours rather than 14 days.

Common mistakes

• Harvesting without negating the upstream. The broad-match parent keeps spending on the term you just promoted. You're now paying twice for the same conversion.
• Harvesting on too little data. Promoting a term with 1 order in 5 clicks is noise-chasing. Most "harvest didn't work" complaints trace back to under-powered samples.
• No dedicated harvested-exact campaign. Mixing harvested terms into existing seed-keyword campaigns destroys the ability to bid each term at its specific CVR. Keep harvested terms in their own campaign structure.
• Treating harvest as a setup task, not a loop. A one-time harvest at campaign launch captures maybe 20% of the value. The remaining 80% comes from weekly cadence over 6–12 months.