Optimizing PPC: the weekly cycle that lowers your ACOS

In the ad console (reports → search terms) you see which REAL queries triggered your ads — with clicks, spend, orders and ACOS per term. That's the difference between keyword (what you booked) and search term (what the customer typed). Auto and broad campaigns keep producing new material here — the weekly routine is essentially the evaluation of this report.

Step 1: harvest winners

• Rule: a search term with at least 2–3 orders and an ACOS below target gets promoted as an exact keyword into the performance campaign.
• Then set the same term as a negative exact in auto/broad — otherwise your campaigns compete against each other and bid up your own click price.

Step 2: stop money burners

• Rule: a term with many clicks and zero orders gets negatived. Rule of thumb: once click costs exceed one unit's profit (at €7 unit profit and €0.70 clicks, that's ~10–15 clicks without a sale), enough is proven.
• Briefly check relevance first: is the term actually fitting and just your listing weak for it? Then it's a listing issue, not a bid issue.
• Also negative thematic mismatches as negative phrase (your product is manual, “electric” searches keep coming? → “electric” negative phrase).

Step 3: adjust bids by formula

• Adjust in steps of at most ±20–30 %, not jumps — the algorithm needs stability.
• No decision without data: a keyword with 4 clicks has no ACOS, it has luck. Let it run or pause it deliberately — but don't “optimize” it blind.

Step 4: steer placements

The placement report splits performance by top-of-search / rest-of-search / product pages. If top-of-search converts clearly better (typical!), raise the placement modifier there instead of the base bid for all placements — a more precise lever, same effect.

Step 5: budget control

• Campaigns that regularly exhaust their budget AND run profitably: raise the budget — money is lying on the street there.
• Campaigns permanently above target ACOS despite optimization: lower or pause, and find the cause (relevance? listing? price?).

• Launch phase (months 1–2): ACOS above break-even is normal — you're buying data and rank. Watch the TREND, not the daily value.
• Consolidation (months 3–6): the harvesting routine shifts revenue into efficient exact keywords; ACOS sinks toward target.
• Maturity: attention moves from ACOS to TACOS: as organic sales grow, individual campaigns may look “expensive” as long as the whole system gets more profitable — the ranking perspective follows in lesson 3.

Your campaign's average ACOS is a blended number — and blended numbers hide the decisive question: what did the additional clicks deliver that your last bid raise bought? That is exactly what incremental ACOS measures: additional ad spend ÷ additional revenue, compared between two states — not within one report.

Higher bids work twice: they make the clicks you were already getting more expensive on average (in the auction you usually pay less than your maximum bid), and they buy additional clicks in weaker positions that convert worse. Model calculation for the AURELO spice grinder set (assumptions; one week, €7 unit profit before ads, break-even ACOS 28 %):

Both columns look healthy — 16 % and 23 % sit clearly below the 28 % break-even. The difference column still exposes the raise: €70 extra cost for €150 extra revenue is a 47 % incremental ACOS. Translated into profit: the 6 additional orders bring 6 × €7 = €42 profit before ads but cost €70 extra — weekly profit falls from €60 (20 × 7 − 80) to €32 (26 × 7 − 150). More revenue, less profit, and the average ACOS never showed it.

• Decision rule: every raise of bid, budget or placement modifier is judged by its incremental ACOS. If it sits above break-even ACOS, the last step loses money — dial it back.
• The announced exception: deliberately scaling above it is legitimate when you are buying rank — but time-boxed, with a fixed date for the return (lesson 3).
• Stay measurable: change only one control per campaign per week and compare 14-day windows before/after (attribution lag). Turn three dials at once and the delta can be attributed to nothing.
• Edge case: if conversion rises in parallel (new reviews, a coupon), the before/after comparison flatters the raise. When in doubt, extend the window or cross-check against an untouched control keyword.