Amazon PPC Bid Optimization: The Math of a Correct Bid

Most bid management is vibes: nudge it up when sales feel slow, panic it down when ACOS spikes. But a keyword's correct bid is a number you can calculate. This guide shows the math and the routine.

The maximum profitable bid

Each click is worth its expected revenue times the share you're willing to give to ads:

Max bid = conversion rate × average order value × target ACOS

Example: product sells for $30, the keyword converts at 10%, your target ACOS is 25%. Expected revenue per click = $3.00. Max profitable bid = $3.00 × 0.25 = $0.75.

Bid above $0.75 and the keyword will run above target even when everything works. Bid far below and you stop winning auctions. The actual CPC you pay (second-price auction) is usually below your bid, which is your margin of safety.

Where the inputs come from — and where they lie

Conversion rate is the fragile input. Three traps:

1. Small samples. 12 clicks and 1 order is not "8.3% CVR" — the honest range at that sample size runs from ~1% to ~30%. Use the keyword's 90-day data, and where it's thin, fall back to the ad group or campaign CVR.
2. Attribution lag. Amazon credits orders up to 7 days after the click, so the last week of data always looks worse than reality. Never compute CVR from windows that include the most recent 2–3 days.
3. Blending eras. A keyword whose listing got a price cut converts differently now than it did last quarter. Weight recent mature data over ancient history — a 65/35 blend of 30-day and 90-day windows works well.

Sizing the change

Once the math says a bid is wrong, move decisively but not violently:

• Changes under 5% are noise — skip them.
• 10–25% for routine corrections; 30–50% when the evidence is strong (lots of clicks, consistent windows).
• After a change, give the keyword a cooldown of 5–7 days. Bids alter traffic mix and placement, and attribution needs time to settle. Re-changing daily creates oscillation, not optimization.

Special cases that override the math

• Brand keywords: defensive value exceeds their measured ACOS; be conservative.
• New keywords (<14 days): no real data yet — leave them alone.
• Launch mode: deliberately bidding above max-profitable buys velocity and organic rank — fine, as long as it's a decision with an end date, not a leak.

The weekly routine

1. Recompute each keyword's CVR on mature windows.
2. Compare current bid to max-profitable bid.
3. Adjust the worst offenders in both directions (cap it at the top 20–50 changes; the long tail can wait).
4. Respect cooldowns.
5. Log changes and review what last week's changes actually did.

That's a few hours weekly by hand — or it's Machine AI's nightly bid pass: every keyword judged against your target ACOS on blended 30/90-day windows, sanity-checked by rules, delivered for one-click approval with 7-day undo. Start with the free audit.