Three days a week, in the same order.
Most advertising accounts are not badly managed. They are inconsistently managed — reviewed thoroughly in month one, glanced at in month four. Here is a routine that survives a busy week, with the thresholds that make each decision automatic.
· Advertising
Ask someone why their advertising drifted and you will rarely hear that they did not know what to do. You will hear that they got busy. The work is not difficult; it is easy to postpone, and postponing it has no immediate consequence, which is exactly what makes it accumulate.
A routine fixes that better than expertise does, for one reason: it removes the decision about whether to do the work. What follows is three sessions a week, always in the same order, with numeric thresholds so that most decisions are made before you open the file.
Monday — search terms
Pull the last seven days of search-term data for every automatic and broad campaign. Those are your discovery layer: each week they pick up queries you never chose, a few of which convert and most of which do not. This is the session that pays for itself.
Sort by spend descending, not by orders. You are looking for where the money went before you look at what it produced — on most accounts the top twenty or thirty rows account for the majority of spend, so that is where an hour is best spent.
Then tag every row you touch with one of four decisions. Tagging matters because it turns a scroll into a list you can act on in bulk:
| What you see | What it means | Action |
|---|---|---|
| 2+ orders in 7 days, not in an exact campaign | Proven converter you are not bidding on deliberately | Harvest into exact match |
| 1 order in 7 days, not in phrase or exact | Promising, not proven | Harvest into phrase as a test |
| 20+ clicks, zero orders | Had a fair run, declined to convert | Negative exact in the source campaign |
| Same term in a broad AND an exact campaign | Your campaigns are competing | Negative exact in the broad or auto one |
| 1 order, under 15 clicks | Not enough data | Leave it, check next week |
Adjust the click threshold to your price point
Twenty clicks is a reasonable default and a poor universal rule. Conversion takes longer at higher prices, because the decision is bigger. A ₹500 impulse item and a ₹6,000 considered purchase do not deserve the same patience.
- Low-priced items: around 15 clicks with no orders is already a signal.
- Mid-priced items: 20 clicks is the usual working threshold.
- Higher-priced items: wait for 30 or so, because a term that looks dead at 12 clicks often is not.
The mathematically honest version is the reciprocal of your conversion rate — the click count at which you would normally expect one order — with a margin on top. Negating below that threshold removes terms that were about to work.
The cannibalisation check people skip
This is the step most often left out, because it needs a second export. Pull the search-term data for your exact campaigns as well, then look for search terms appearing in both.
When one does, your campaigns are bidding on the same query. Amazon serves one ad per seller per search, so if the broad campaign wins the impression, your exact campaign — the one with your highest bid and best historical conversion — does not get it. Nothing about that shows up as a bad number anywhere. It simply costs more.
The fix is a negative exact in the broad or automatic campaign, which forces the query through the campaign you actually optimised for it. Keeping a simple matrix of which exact keywords have been negated where is worth the ten minutes it takes to start.
Wednesday — bids
Bid decisions need a longer window than search-term decisions. Seven days carries too much noise — one strong weekend or one bad Tuesday will move a keyword enough to trigger a change you would not make with more data. Fourteen days is the minimum worth acting on.
| Keyword state | Action |
|---|---|
| 20+ clicks, ACOS above target, at least 1 order | Reduce bid 10–20% |
| 20+ clicks, ACOS above target, zero orders | Reduce 30%, or pause if it repeats next week |
| 20+ clicks, ACOS at or below target, 3+ orders | Increase bid 10–15% |
| Under 20 clicks | Leave it alone entirely |
| Zero impressions in 7 days | Raise bid 20% once, then wait a week |
That fourth row is the one that takes discipline. A keyword with eleven clicks and no orders looks like it needs action and does not have enough data to justify any. Touching it is worse than leaving it, because you will not be able to attribute what happens next.
The budget check that belongs here
While you have the campaign data open, put spend next to daily budget. Any campaign consistently exhausting its budget before midday is turning away orders during the hours people are actually shopping.
But do not simply raise the budget. First check whether waste inside that campaign is eating it — a campaign capping out because a handful of non-converting terms are consuming it does not need more money, it needs Monday's session applied to it. Raise the budget once you know the money is going somewhere useful.
The mirror image is just as common and almost always misdiagnosed: a campaign spending a third of its budget is not budget-constrained at all. It is losing auctions. More budget will change nothing, and raising it is the most frequent response to a campaign that is underperforming.
Make changes in bulk, with a reason column
Clicking through campaigns individually is slow and leaves no record. Download the bulk file, make the changes in a spreadsheet with a column recording why, and upload it. The reason column is not bureaucracy — it is the thing that lets you understand in October what you were thinking in August.
Friday — the health check
Fifteen minutes, and none of it is about individual keywords. This session exists to catch the account-level drift that keyword work cannot see.
- TACoS against last week — total ad spend over total revenue, organic included. Two consecutive weeks of increase is worth investigating rather than waiting for month-end to notice.
- Organic rank on your handful of genuinely important keywords. A drop of more than a few positions warrants checking whether the ranking campaign actually spent, whether it was budget-capped, and whether listing conversion moved.
- A one-line log: date, TACoS, anything flagged, anything changed. This is worthless for eight weeks and then becomes the only way to reconstruct what happened.
The TACoS check is the one that earns its place. Every campaign can show an acceptable ACOS while advertising steadily becomes a larger share of everything you sell, and ACOS is structurally incapable of showing you that.
Monthly, on top
Once a month, do two things the weekly cycle is too short to support. First, an n-gram pass over thirty days of converting search terms — breaking them into word and phrase fragments to find intent patterns no single term reveals. If a word appears across five or more converting terms and has no campaign of its own, that is your next campaign.
Second, budget allocation across objectives rather than campaigns. Weekly work moves money within a structure; monthly work asks whether the structure is still where you want the money.
What makes this survive a busy month
Three sessions, roughly two hours a week for a single brand, and the thresholds mean most rows decide themselves. That is the entire point — a routine you can follow on a bad week is worth more than a thorough process you abandon in month three.
If you skip a week, resume at Monday rather than trying to catch up on everything at once. The search-term session is the one that compounds when neglected; the others tolerate a gap.
The Monday session, in about ten minutes.
The Weekly PPC module takes the exports this routine calls for and returns the harvest candidates, the negatives and the cannibalisation flags already sorted. It parses in your browser, so the file never reaches us.
Frequently asked questions.
How often should I optimise Amazon PPC campaigns?
Search terms and negatives weekly, because automatic and broad campaigns pick up new queries continuously and waste rebuilds after every pass. Bids fortnightly at most, since a bid decision needs at least fourteen days of data to be worth making. Structure and budget allocation monthly. The cadence matters less than its consistency — a simple routine followed every week beats a thorough one abandoned after two.
How many clicks before adding a negative keyword?
Around twenty with no orders as a working default, adjusted for price. Cheaper items convert faster, so fifteen is often enough; higher-priced items deserve thirty or so, because a considered purchase takes longer and a term that looks dead at twelve clicks frequently is not. The principle underneath is the reciprocal of your conversion rate plus a margin — negate below that and you remove terms that were about to work.
How many orders before promoting a search term to exact match?
Two in a week is a reasonable bar for exact match; one is worth a phrase-match test rather than a promotion. A single order is as likely to be luck as signal, and exact match means committing your highest bid to it. Whatever you promote, remember to negate it in the campaign that discovered it, or the two will compete for the same query.
Should I change bids weekly?
Only on keywords with enough data, which in practice means fortnightly for most of them. Seven days carries too much noise — a strong weekend can push a keyword over a threshold it does not really cross. The keywords with under twenty clicks should simply be left alone, which is harder discipline than it sounds because they are the ones that look like they need attention.
What should I check if TACoS is rising?
Whether revenue is rising with it. Rising TACoS on rising revenue is buying growth, which is fine when deliberate. Rising TACoS on flat revenue means you are paying more for the same sales, and the usual causes are a competitor entering the auction, a listing losing rank or the Buy Box, or bids drifting up without a decision. Two consecutive weeks of increase is the point to investigate rather than wait.
- Where Amazon ad spend actually goes
Monday's session in depth — the thresholds, and the word-level view that finds what term-level review misses.
- One account, eight jobs
Why the Wednesday budget check keeps finding the same two problems, and what sits underneath them.
- The Amazon Search Term Report
The export Monday runs on, with the exact path and the columns worth understanding.