How Amazon PPC Actually Works

The mechanics, in the order they actually happen

Every Amazon PPC ad runs through the same sequence, whether it's a $2 keyword or a $200 one. A shopper searches. Amazon holds an instant auction among every advertiser targeting that term who still has budget left. Amazon ranks the eligible bids not by price alone but by bid × relevance × predicted conversion rate — a listing with strong click and conversion history can beat a higher bid from a weaker listing.

The winner's ad shows. If nobody clicks, nobody pays — that's the pay-per-click part. If someone does click, you're charged roughly one cent more than the next-highest qualifying bid, capped at whatever you set as your own maximum. You rarely pay your full bid unless the auction is genuinely tight.

This is the part most explanations stop at. The auction is one event. A campaign is thousands of these events running continuously, and the bid that wins on Monday can lose on Friday because a competitor raised theirs, your listing's conversion rate dropped, or something else changed what's competing for that placement.

A worked example: from bid to ACOS

Numbers make this concrete. Say you bid $1.20 on a keyword. Over a week you get 1,000 impressions at a 3% click-through rate — 30 clicks. The next-highest qualifying bidder was at $0.94, so Amazon charges you $0.95 per click, not your full $1.20.

• Spend: 30 clicks × $0.95 = $28.50
• Conversion rate: 10% → 3 orders
• Average order value: $30 → $90 in ad-attributed revenue
• ACOS (ad spend ÷ ad revenue): $28.50 ÷ $90 = 31.7%

That 31.7% is only good or bad relative to your margin. If the product carries a 45% gross margin before advertising, a 31.7% ACOS leaves room to profit. If it carries a 20% margin, the same campaign is losing money on every sale it drives — the auction did exactly what it was supposed to do; the product's economics just didn't support that bid.

Automatic vs manual: the loop most sellers abandon

Amazon gives you two targeting modes. Automatic campaigns let Amazon match your ad to search terms it judges relevant — useful for discovery, weak on control. Manual campaigns let you pick the exact keywords and bids — strong on control, useless without data to point at first.

The standard sequence: run automatic first, pull the search term report once it's had enough traffic to mean something, then move anything that converted into a manual campaign where you control the bid precisely, adding it as a negative in the automatic campaign so the two stop competing for the same click. Anything that spent without converting becomes a negative keyword everywhere.

The honest mistake — one we've made — is harvesting too early. Pulling a search term into manual after two clicks and one lucky sale isn't a pattern, it's noise, and a manual campaign built on noise burns budget on a keyword that was never actually good. The fix isn't to stop harvesting. It's to wait for enough clicks that the conversion rate you're reacting to is real, not a coin flip that happened to land right.

When your ACOS looks wrong, check this before touching a bid

Work through these in order before you change anything:

• Attribution window — Amazon credits sales up to seven days (fourteen for some categories) after the click. A campaign that looks bad today can look fine once the window closes.
• New launch or new keyword — a keyword with a small handful of clicks doesn't have a reliable conversion rate yet. High ACOS here is a sample-size problem, not a targeting problem.
• Search term report — pull it and look for terms with clicks but zero orders. That's spend with no return, and it's fixable today with a negative keyword.
• Placement, not just bid — the same keyword costs differently on top-of-search versus product pages versus the rest of search. A bad overall ACOS sometimes hides one placement that's fine and one that's burning the budget.

If none of that explains it, the bid genuinely is too high for what the product can afford, and the fix is a lower bid or a pause — not a bigger budget hoping it evens out. Whatever the change, it should carry three things before it runs: the evidence behind it, a measurement plan, and a rollback trigger. Without the third one, you don't find out you were wrong until a month of spend has already gone.

Four ad types, four different auctions

The mechanics above describe Sponsored Products, the most common Amazon ad type and the one that decides whether the rest of this is worth learning. The other three run on the same pay-per-click principle but auction different things — see the table below.

Sponsored Brands auctions a keyword too, but the ad sits above search results with logo and lifestyle imagery, so what gets judged includes creative quality, not just the product listing. Sponsored Display targets an audience or a competing product page rather than a search term, and can follow shoppers off Amazon. DSP is a different system entirely — programmatic, audience-based, and bidding in real time across and beyond Amazon's own inventory, mostly at spend levels smaller sellers won't hit.

Where this gets tedious enough to hand off

Everything above is the same whether you run it yourself, in software, or hand it to someone else — what changes is who's checking the search term report every week and whether the bid change was made with evidence or a hunch. Dr. PPC reads a brand's ad account, proposes each change with the evidence behind it, a measurement plan, and a rollback trigger, and lets the client set how much runs automatically versus waits for a click. It's $300 a month plus 3% of ad spend, capped, month-to-month, with the first 30 days free, and Orbit — the analytics and tracking software underneath it — is included at no extra charge. None of that changes how the auction works. It changes how often someone's actually watching it.