The full advertising stack: formats, audiences, attribution
- You know the real job of every ad format — and the job it cannot do.
- You can tell audience targeting from keyword targeting and use both where they belong.
- You understand attribution windows and why the sum of your reports never adds up.
- You know the multiplier chain of placement uplift and bidding strategy.
Most accounts run one format for years and wonder why ACOS rises with every growth push. The reason is rarely poor optimisation — it is a mix-up of jobs: a format built to HARVEST demand is suddenly asked to CREATE it. This lesson sorts the toolbox. It assumes Growth L1 (structure) and L2 (the weekly cycle).
1The formats and their real jobs
| Format | Job | Strength | What it CANNOT do |
|---|---|---|---|
| Sponsored Products | Harvest demand | People who search intend to buy. Best cost-to-purchase ratio, right at the result. | Create demand. Where nobody searches, there is nothing to harvest. |
| Sponsored Brands | Show brand and range | Prominent space above the results, several products or video, a path into the Brand Store. | Fine-grained single-product control. Measurement is blurrier than with SP. |
| Sponsored Display | Re-engagement and displacement | Places you on other detail pages and off the search results; audiences instead of search terms. | No substitute for search presence. Without a convincing detail page it evaporates. |
| DSP | Reach and audiences at scale | Also off Amazon; fine-grained audiences; built for brand building. | Nothing for small budgets. Not assessable without clean measurement. |
The order in this table is also the order of adoption. A format joins only when the previous one is exhausted — and “exhausted” means: you reliably reach your key search terms, and further bid increases cost more than they return. Start Sponsored Display while search terms in Sponsored Products are still unserved and you are buying expensive reach instead of cheap sales.
Sponsored Products is the market stall next to the till: whoever walks past is already shopping. Sponsored Brands is the big sign above the market square. Sponsored Display is the flyer handed to someone who was just standing at a competitor's stall. And DSP is the billboard on the bypass. All four can work — but if you only have one budget, you do not hang it on the bypass.
2Audiences instead of search terms: the second way of thinking
Sponsored Products thinks in queries: someone types something, you appear. Sponsored Display and DSP think in audiences: someone DID something, you appear. That is a completely different lever, and it demands different expectations.
- Re-engagement (remarketing): people who viewed your detail page and did not buy. Almost always the first and best audience — they already know you.
- Product and category targeting: ads on specific competitor detail pages or within categories, refinable by price range, star rating and delivery type. That refinement is the actual trick: without it you also buy placements on products that are cheaper and better rated than yours.
- Interest and in-market audiences: people active in a category. Broad, cheaper per contact, but far from the purchase.
- Exclusion is part of the audience: exclude your own recent buyers if your product is not a repeat purchase. Otherwise you pay to advertise to people whose shelf is already full.
The AURELO set runs a Sponsored Display campaign on competitor detail pages in the spice grinder category. Without refinement the ad also appears on €12.99 products rated 4.7 — hardly anyone there clicks a €24.99 set, and whoever does compares unfavourably. With the refinements “price above €19” and “rating below 4.3”, what remains is exactly the space where the AURELO set is the better buy: pricier than the cheap ones, better rated than the expensive ones. Same campaign, same bid — only the space is different, and with it the conversion rate.
3Attribution: why the sum of your reports never adds up
Every format attributes sales that happen within an attribution window after its ad. Those windows differ in length: 7 days are reported for Sponsored Products, 14 days for Sponsored Brands (evidence level: practice — the authoritative values sit in the ad console, behind the login). For view-based delivery Amazon changed the model on 1 January 2026 and reports a comparison figure on the old basis alongside it (evidence level: practice; anyone who saw a Sponsored Display collapse back then saw a measurement correction, not a demand collapse).
Three consequences follow that almost every report gets wrong:
- The same order can be counted several times. Someone who clicks a Sponsored Brands ad and later a Sponsored Products ad can appear in both reports. The sum across formats is therefore larger than your actual advertised revenue.
- Longer windows look better. A format with 14 days automatically appears stronger than one with 7 — without a single extra sale.
- Sales cross period boundaries. A click at month end pays into the following month. Month-on-month comparisons with tight margins are systematically skewed.
The one number free of all this is TACOS: ad spend divided by TOTAL revenue. No double counting, because the denominator is real revenue — and no window question, because it attributes nothing (Growth L11).
Adding up advertised revenue across all formats and comparing it to total revenue — then being amazed or delighted by a 90 % “ad share”. That sum does not exist: it contains the same orders several times. Formats are judged individually or not at all; comparison happens via TACOS and contribution margin.
4The multiplier chain: what your bid really is
In Sponsored Products the bid can be raised per placement — top of search, product pages and rest of search, each between 0 and 900 % (evidence level: practice). Two quirks that are routinely overlooked:
- Up only, campaign level only. You cannot devalue a weak placement, only uplift the others. To genuinely avoid a placement you need a separate campaign.
- Multipliers stack with the bidding strategy. A 100 % top-of-search uplift plus the “dynamic, up and down” strategy can mean four times your base bid at the peak. Turn both up at once and you pay prices you never set.
| Base bid | Placement uplift | Bidding strategy | Maximum possible |
|---|---|---|---|
| €0.60 | none | fixed | €0.60 |
| €0.60 | +100 % | fixed | €1.20 |
| €0.60 | +100 % | dynamic, up and down | up to €2.40 |
| €0.60 | +300 % | dynamic, up and down | up to €4.80 |
Hence the professionals' rule: turn one screw at a time. Either base bid OR placement uplift OR strategy — then measure a full week before the next one.
On the AURELO set, ACOS jumps from 22 to 38 % in a week with no bid changed. The change log holds the cause: two weeks earlier the top-of-search uplift was set to 250 %, and in the same week the strategy switched to “dynamic, up and down”. Individually each change would have been defensible; together they tripled the effective bid on the most expensive space. The strategy was reverted first — it was the more recent change and the harder one to steer.
5Amazon Attribution and external channels in one paragraph
Amazon Attribution supplies measurement tags for traffic from outside — search engines, social networks, newsletters — and reports clicks, detail page views and purchases for them. Without those tags external traffic is blind (Growth L8). Two pro notes: the frequently cited rebate on sales from external traffic (Brand Referral Bonus) is a US store programme and not available in Germany (evidence level: practice, several 2026 industry sources) — budgeting for it means budgeting wrong. And: attribution measures assignment, not additionality; the difference is covered in the expert chapter of Growth L8.
- One job noted per format — no format without a job in the account.
- New format started only once the previous one is demonstrably exhausted.
- On audience campaigns: refinements by price, rating and delivery type applied.
- Your own recent buyers excluded where the product is not a repeat purchase.
- Advertised revenue NOT summed across formats; steering via TACOS.
- Placement uplift and bidding strategy never changed in the same week.
- External traffic only with attribution tags, and no bonus programme in the plan.
6Expert insight: the branded keyword test — are you paying for sales you would get anyway?
The most expensive habit in mature accounts is advertising on your own brand name. It looks superb in every report — low ACOS, high conversion — and is suspect for exactly that reason: whoever searches your brand was coming to you anyway. The question is not whether the campaign shows good numbers but whether it delivers additional sales.
Three reasons justify brand advertising, and all three can be checked rather than claimed:
- Displacement: competitors advertise on your brand name. Checkable by searching yourself in a clean browser window — is there a competitor's ad above your result?
- Range: the brand searcher should see more than the one product they know. Sponsored Brands with a store destination is built for that.
- New customers: some brand searchers know you from a recommendation and buy for the first time. The metric for that is “new-to-brand” and it is in the reports.
The test that settles it is uncomfortable but short: pause the campaign and watch the totals. Not the campaign numbers — those obviously go to zero. Watch the product's total revenue:
| Observation after pausing | Interpretation | Consequence |
|---|---|---|
| Total revenue stays roughly the same | The ad replaced organic sales rather than adding to them | Leave it off, or reduce to a minimal displacement bid |
| Total revenue drops noticeably | The ad was additional — likely displacement or range effect | Switch it back on, the budget justifies itself |
| Revenue drops but new-to-brand share was low | You are paying for existing customers who would find you anyway | Rebuild as store campaigns with a range effect |
Two rules make the test reliable. First: at least two full weeks, and never across a promotion, a single weekend or a month boundary — otherwise you measure seasonality, not effect. Second: change nothing else at the same time. A new price in the same week makes the result worthless, and you repeat the test in three months with the same open questions.
The same logic applies to any campaign with suspiciously good numbers: a very low ACOS is not proof of good advertising, it is often a sign of harvesting a field that was already yours. The extreme case is the automatic campaign that serves almost exclusively on your own ASIN and brand name — technically flawless, economically a loop. You spot those in the search term report, not in the headline metric.
Pausing a well-running campaign also removes its ranking contribution — advertised sales count like any other sales. So test ONE campaign, never the whole account, and never in peak season. A test that costs you ranking was more expensive than the insight is worth.
The pro track assumes the other three and repeats nothing from them. It opens once you have completed all three in full — every lesson quiz at 80 % or better, and every final quiz passed.
Not logged in? Your progress only counts with a free account.
Check yourself
6 quick questions — one at a time, instant feedback. With a free account your progress is saved.
Frequently asked
When does DSP make sense for a mid-sized seller?
Practically only when two conditions coincide: search space is exhausted, and you can judge effect without click attribution — that is, via total revenue and TACOS rather than campaign ACOS. Without that you buy reach you cannot assess and abandon the experiment after three months with nothing learned.
Should Sponsored Brands point at the Brand Store or a product list?
The store, if the range is your strength and the searcher does not yet know which product they want. A product list, if one clear bestseller carries the decision. That cannot be settled by feel, only by a clean swap: one destination per period, same campaign, same budget.
How often should I change audiences in Sponsored Display?
Less often than it feels right. Audiences need far longer than search terms to accumulate enough data for a verdict — four weeks is a lower bound, not a target. Change them weekly and you collect many small samples and learn nothing from any of them.
The free Listing Check scores any ASIN from 0 to 100 in a minute — the fastest way to see whether the mechanics from this lesson actually hold on your own listing.
Go deeper: the complete guide to optimizing Amazon listings →
Everything in this academy comes from day-to-day selling practice — the same playbook behind Listimo, the tool that turns product photos into complete Amazon listings.