Your dashboard shows a strange combination: ACOS rose from 20% to 25%, while TACOS fell from 10% to 8%. One teammate says the ads are getting worse. Another says the product is building organic momentum.
Neither conclusion is ready yet.
Suppose ad spend increased from $1,000 to $1,200, attributed sales slipped from $5,000 to $4,800, and total sales climbed from $10,000 to $15,000. ACOS worsened because spend grew while attributed sales fell. TACOS improved because total sales grew much faster than spend. The two ratios are not contradicting each other; they are answering different questions with different denominators.
That arithmetic tells you what moved. It does not tell you whether ads created incremental demand, a price increase lifted revenue, stock availability changed, a promotion shifted the product mix, or reporting rules moved sales between periods.
The practical answer to “ACOS vs. TACOS” is therefore not to choose one metric. It is to match their scopes, reopen the raw amounts, test several explanations, and only then make a budget decision.
One Spend Number, Two Different Denominators
Let:
- S = the ad spend included in the analysis
- A = the sales credited to those ads under a named attribution rule
- T = the total sales for a named product set and revenue definition
Then:
ACOS = S ÷ A × 100
TACOS = S ÷ T × 100
Amazon’s own ACOS guide defines ACOS as ad spend divided by ad revenue and warns that there is no single “good” ACOS for every business. Targets depend on economics and objectives, not on a universal percentage (Amazon Ads, “What is advertising cost of sales?”).
For this guide, TACOS is an explicit calculation, not an assumed universal field: selected ad spend divided by the total-sales denominator you name. That wording matters because tools can use different totals. Perpetua’s provider documentation, for example, lists versions based on ordered revenue, shipped revenue, and shipped product cost of goods sold (Perpetua, “Sales & Traffic — Metrics Index”). Two dashboards can therefore display a value called “Total ACoS” while measuring different economic bases.
With comparable, positive inputs and the same spend numerator, this identity holds:
TACOS ÷ ACOS = A ÷ T
It is useful as an arithmetic check. It is not a causal formula. A ÷ T is the credited share under the selected definitions, not proof that those sales would disappear without advertising.
The four trend patterns can all occur legitimately:
| Trend | Arithmetic condition | What it establishes |
|---|---|---|
| ACOS up, TACOS down | Attributed sales grow more slowly than spend; total sales grow faster than spend | Ad-attributed efficiency weakened while spend became a smaller share of total sales |
| ACOS down, TACOS up | Attributed sales grow faster than spend; total sales grow more slowly than spend | Ad-attributed efficiency improved while spend became a larger share of total sales |
| Both up | Spend grows faster than both sales measures | Advertising burden increased against both denominators |
| Both down | Spend grows more slowly than both sales measures | Advertising burden decreased against both denominators |
None of those rows names a cause. “Launch,” “organic flywheel,” “brand decay,” and “efficient scaling” are hypotheses that need additional evidence.
Match the Reports Before You Interpret the Trend
A neat formula can still compare incompatible populations. Before discussing performance, fill in a scope card for both periods.
| Scope decision | What must be explicit |
|---|---|
| Marketplace and currency | One marketplace, currency, and exchange-rate treatment |
| Advertiser and ad products | Seller or vendor; Sponsored Products, Sponsored Brands, Sponsored Display, DSP, or a stated combination |
| Product boundary | Child ASIN, parent, SKU, fixed portfolio, brand, or account |
| Spend numerator | Which products and channels contribute spend; whether fees or taxes are excluded |
| Attributed-sales field | Exact report field, click/view basis, lookback, purchased-product scope, and reporting date basis |
| Total-sales field | Ordered, shipped, or net sales; product boundary; tax, shipping, discounts, refunds, and cancellations |
| Period | Start and end dates, time zone, equal-length or normalized comparison |
| Data maturity | Extraction timestamp, report version, and whether both periods were re-exported under the same method |
The total-sales side is not a generic number. Amazon’s Selling Partner API documentation says the seller Sales and Traffic Business Report covers the seller’s catalog and can aggregate by date and ASIN, with PARENT, CHILD, and SKU options. It also notes that weekly and monthly requests can expand the supplied date boundaries to full calendar periods (Amazon Selling Partner API, “Analytics Reports”). A child-level ad numerator divided by a parent-level retail denominator may be useful for one management question and invalid for another.
Returns create another alignment problem. Amazon’s separate returns reports are organized by return date and include order date, return-request date, ASIN, return quantity, order amount, and refunded amount (Amazon Selling Partner API, “Returns Reports”). That does not establish how every Amazon Ads sales field is adjusted. It does establish why “total sales” and “attributed sales” should not be assumed to have identical timing and adjustment rules.
Do not borrow an attribution window from a different Amazon product. The official guide for Amazon Attribution—the measurement product for non-Amazon marketing—uses a 14-day, last-touch click model (Amazon Ads, “Your complete guide to Amazon Attribution”). That is not, by itself, documentation of the default for every Sponsored Products, Sponsored Brands, Sponsored Display, seller, or vendor report. Save the exact field and attribution basis used in your export instead of writing “Amazon uses 14 days” in an analysis note.
Report versions also matter. Amazon announced unified reporting on June 8, 2026, with standardized metrics, dimensions, and attribution methodology across ad products, and said the legacy Sponsored Ads and DSP report pages would be sunset by December 31, 2026 (Amazon Ads, unified reporting announcement). When a comparison crosses a reporting migration, re-export both periods through the same surface where possible and retain the field names.
Finally, handle zeroes honestly. If attributed sales are zero, ACOS is not available, not 0%. If total sales are zero, TACOS is not available. Missing data stays missing. A ratio above 100% can be mathematically valid. A negative adjusted denominator is a reconciliation event, not a normal performance trend.
The accompanying ACOS/TACOS diagnostic worksheet flags scope and denominator issues alongside its calculations. Reconcile any warnings before interpreting the trend. If you export the worksheet as CSV, retain the time zone and adjustment basis separately; those two fields are not included in its export.
Download the eight-row worked-case CSV to inspect the inputs behind all four examples.
Four Products, Four Different Investigations
The following are fictional worked examples, not Hookin or client results. They use a US seller account, USD, Sponsored Products only, and two equal 28-day periods: May 4–31 and June 1–28, 2026. Cases 1–3 use one child ASIN; Case 4 uses a fixed two-child family. Both periods use the same illustrative seven-day click field, extracted on September 8, 2026. That lookback is a teaching assumption, not a claim about a universal Sponsored Products default.
Revenue excludes tax and shipping. The examples assume no cross-period credited orders, no outside-product halo, no refunds, and no duplicate sales. Contribution means total sales − variable costs − ad spend; it excludes fixed overhead, financing, and tax.
| Case | Period | Spend S | Attributed sales A | Total sales T | ACOS | TACOS | Contribution after ads |
|---|---|---|---|---|---|---|---|
| 1. Water bottle | P1 | $1,000 | $5,000 | $10,000 | 20% | 10% | $1,500 |
| P2 | $1,200 | $4,800 | $15,000 | 25% | 8% | $2,550 | |
| 2. Desk lamp | P1 | $1,200 | $4,000 | $12,000 | 30% | 10% | $2,400 |
| P2 | $900 | $4,500 | $7,500 | 20% | 12% | $1,350 | |
| 3. Storage set | P1 | $800 | $4,000 | $8,000 | 20% | 10% | $1,600 |
| P2 | $1,600 | $5,000 | $10,000 | 32% | 16% | $1,400 | |
| 4. Packing-cube family | P1 | $1,500 | $5,000 | $10,000 | 30% | 15% | $1,750 |
| P2 | $900 | $4,500 | $7,500 | 20% | 12% | $975 |
Case 1: ACOS rises while TACOS falls
The water bottle appears to fit a familiar story: advertising efficiency weakened, but the product’s broader business grew. The raw trend is real. The usual explanation is not yet verified.
Clicks fell from 1,000 to 800 while spend rose from $1,000 to $1,200, so CPC increased from $1.00 to $1.50. Attributed purchases fell from 200 to 160, but purchase conversion rate remained exactly 20%. Attributed average order value rose from $25 to $30.
ACOS = CPC ÷ (purchase CVR × attributed AOV)
P1: $1.00 ÷ (20% × $25) = 20%
P2: $1.50 ÷ (20% × $30) = 25%
For this record, “conversion rate collapsed” is false. Higher click cost outpaced the higher order value.
Total units rose from 400 to 500 while realized price rose from $25 to $30. The $5,000 revenue increase can be decomposed into $2,000 from applying the $5 price change to the original 400 units and $3,000 from 100 additional units at the new price. That is an accounting decomposition—not proof that price caused the volume change or that ads caused either component.
At a 25% pre-ad contribution margin, contribution after ads rises from $1,500 to $2,550. The business outcome improved under these cost assumptions, even though ACOS worsened. The next check is paid query and placement CPC, alongside total units, realized price, category query volume, and the ASIN’s share of that demand. Falling TACOS alone does not justify automatic scaling.
Case 2: ACOS falls while TACOS rises
The desk lamp’s ACOS improves from 30% to 20%, yet TACOS rises from 10% to 12% because total sales fall much faster than spend.
The price stays at $50. Paid purchases rise from 80 to 90, clicks fall from 1,200 to 900, CPC stays at $1, and paid purchase conversion improves from 6.67% to 10%. The advertising slice looks better.
But the product is sellable and eligible for only 420 of 672 hours in P2, versus all 672 hours in P1. Total units per available hour are identical:
240 ÷ 672 = 150 ÷ 420 = 0.357143 units per available hour
The missing 252 hours multiplied by that rate equals 90 units, or the observed $4,500 revenue gap. This is an arithmetic consistency check, not a causal lost-sales estimate. Demand varies by hour, and stockouts can be caused by demand as well as cause missed opportunity.
Amazon’s Sponsored Products page says the ads are cost-per-click, appear only when advertised items are in stock, and require product eligibility for the Featured Offer (Amazon Ads, “Sponsored Products”). That makes availability a necessary diagnostic, not proof that availability caused this account’s decline.
At a 30% pre-ad margin, contribution after ads falls from $2,400 to $1,350 despite the lower ACOS. The next check is hour-level sellability and eligibility, matched weekday and time-of-day demand, sessions, and branded versus non-branded target mix. Cutting efficient traffic solely because TACOS rose may attack the surviving part of the business rather than the missing retail opportunity.
Case 3: both ACOS and TACOS rise
The storage set looks broadly worse: ACOS moves from 20% to 32%, TACOS from 10% to 16%, and contribution after ads slips from $1,600 to $1,400. Aggregation hides where the change occurred.
Existing targets are unchanged in both periods: $800 spend, $4,000 attributed sales, 20% ACOS. A new target block added in P2 spends another $800 and receives $1,000 in attributed sales, or 80% ACOS. Combined, P2 becomes $1,600 ÷ $5,000 = 32%.
That localizes the ratio change to the new block. It does not prove the block is a bad launch investment, nor does the word “launch” guarantee later payback.
With a 30% pre-ad contribution margin, the extra $800 of spend requires at least $800 ÷ 30% = $2,666.67 in incremental net revenue to break even on variable contribution. The block’s $1,000 in credited sales is not an incrementality estimate. Nor is the product’s $2,000 total-sales increase automatically caused by the new targeting.
The next check is the new targets as their own matured cohort: queries, placements, CPC, conversion, purpose, cap, and a credible counterfactual for incremental revenue. Do not pause the unchanged 20% ACOS targets because the aggregate worsened. Do not fund the 80% block indefinitely because it has a lifecycle label.
Case 4: both ratios fall while no child improves
The packing-cube family is the most dangerous “good news” case. Blended ACOS improves from 30% to 20% and TACOS from 15% to 12%, yet total sales fall and contribution after ads drops from $1,750 to $975.
| Period and child | Spend | Attributed sales | Total sales | ACOS | TACOS | Pre-ad margin | Contribution after ads |
|---|---|---|---|---|---|---|---|
| P1, low-margin child | $375 | $2,500 | $3,750 | 15% | 10% | 20% | $375 |
| P1, high-margin child | $1,125 | $2,500 | $6,250 | 45% | 18% | 40% | $1,375 |
| P2, low-margin child | $562.50 | $3,750 | $5,625 | 15% | 10% | 20% | $562.50 |
| P2, high-margin child | $337.50 | $750 | $1,875 | 45% | 18% | 40% | $412.50 |
Neither child’s ACOS nor TACOS changes. The family ratios improve because revenue shifts toward the low-ACOS, low-margin child. Its share of family revenue rises from 37.5% to 75%. The blended pre-ad margin falls from 32.5% to 25%, and the high-margin child loses most of its volume.
The next check is child-level stock, eligibility, price, promotion, traffic, and unit demand. “Both ratios fell” does not establish that every product improved—or that the family can be safely scaled.
Why Total Sales Minus Attributed Sales Is Not Organic Growth
Subtracting A from T can produce a useful residual only after the populations match. Calling that residual “organic sales” makes a stronger claim than the calculation supports.
First, date bases can disagree. Imagine one $100 order generated after a May 31 click but placed on June 2. If the attributed report assigns it to click date while the retail report assigns it to order date, May shows A = $100, T = $0, and June shows A = $0, T = $100. The residual is negative in May and positive in June even though only one purchase occurred.
Second, product boundaries can disagree. If attributed sales include $400 of other purchased products, then A = $1,200 can exceed T = $1,000 for the advertised child ASIN. The resulting −$200 is not “negative organic sales.” It is a reconciliation alarm.
Third, a falling TACOS does not even guarantee that residual dollars increased. Compare these two periods:
| Period | Spend | Attributed sales | Total sales | ACOS | TACOS | T − A |
|---|---|---|---|---|---|---|
| P1 | $100 | $500 | $1,000 | 20% | 10% | $500 |
| P2 | $50 | $250 | $600 | 20% | 8.33% | $350 |
TACOS falls and the residual share rises, but residual dollars fall from $500 to $350.
Even a perfectly aligned T − A is only unattributed residual sales under the selected model. It can include direct navigation, organic search, promotions, email, external traffic, repeat purchases, and sales that advertising influenced but the model did not credit. Attribution assigns credit after an outcome; incrementality asks what would have happened without the ad.
That is not a semantic distinction. An Amazon-affiliated paper on multi-touch attribution explicitly separates observational credit assignment from causal effect measurement and treats randomized holdouts as the ground-truth route to incrementality (Lewis et al., “Amazon Ads Multi-Touch Attribution,” 2025). Standard ACOS and TACOS trends do not supply that counterfactual.
Set a Margin-Based Limit, Not a Universal Target
A ratio target becomes a business target only after revenue and costs share the same basis.
Consider a fictional period with $10,000 in gross ordered revenue, $1,000 in refunds, $9,000 in net revenue, $6,300 in variable costs, and $1,000 in ad spend.
| Measure | Calculation | Result |
|---|---|---|
| Dashboard TACOS on gross ordered revenue | $1,000 ÷ $10,000 | 10% |
| Ad burden on net revenue | $1,000 ÷ $9,000 | 11.11% |
| Pre-ad contribution margin | ($9,000 − $6,300) ÷ $9,000 | 30% |
| Contribution after ads | $9,000 − $6,300 − $1,000 | $1,700 |
Subtracting 10% TACOS from a 30% net-revenue margin to declare a 20% result would mix denominators. The corrected contribution here is 18.89% of net revenue before fixed overhead and tax.
Amazon’s ACOS guide links break-even ACOS to profit margin, which is a useful starting principle (Amazon Ads). In practice, the margin must match the revenue basis and include the variable costs relevant to the decision. A credited-sale break-even check still does not prove that every credited sale was incremental.
You can also solve for a conditional spend ceiling. If this product has $500 of fixed costs assigned to the period and the business wants a result equal to 10% of net revenue, then:
Maximum spend = net revenue × pre-ad margin − fixed costs − target result
$9,000 × 30% − $500 − ($9,000 × 10%) = $1,300
That equals 14.44% of net revenue. It is not a universal “good TACOS.” Change the refund treatment, variable costs, fixed-cost allocation, or target result and the ceiling changes.
Use the same logic at the margin. In Case 3, the extra $800 spend needs $2,666.67 of incremental net revenue at a 30% margin. Average account ACOS cannot answer whether that additional block cleared the hurdle.
Choose the Next Check Before the Next Budget Move
A useful ACOS-versus-TACOS review ends with a discriminating check, not a lifecycle label.
- Freeze the scope. Record marketplace, advertiser type, ad products, product set, field names, attribution basis, total-sales basis, currency, period, time zone, report version, and extraction timestamp.
- Re-export comparable periods. Use the same method and save both raw amounts. If the same historical period changes between extracts, investigate data maturity or report methodology before inventing a new business story.
- Decompose the ratios. Compare the growth of spend, attributed sales, and total sales. Then inspect CPC, conversion rate, attributed order value, price, units, and product mix where relevant.
- Test the leading alternatives. Check stock and eligibility, promotions, price, search demand and ASIN share, target and placement mix, and product-level contribution.
- Apply economics last. Use net revenue, real variable costs, fixed-cost assumptions, and a stated target. Separate average credited return from marginal incremental return.
| Observed pattern | Best next check in the worked case | Decision still unsupported |
|---|---|---|
| ACOS up, TACOS down | CPC and placement mix; units and realized price; demand versus ASIN share | Ads created organic growth; scale automatically |
| ACOS down, TACOS up | Hour-level sellability and eligibility; matched demand periods | Organic demand collapsed; cut efficient traffic automatically |
| Both up | Old versus new target cohorts; marginal contribution hurdle | A launch will inevitably pay back later |
| Both down | Child-level mix, margin, stock, price, and contribution | Every product improved; scale the family |
ACOS tells you how much selected ad spend sits against selected attributed sales. TACOS tells you how much that spend sits against a named total-sales base. Their divergence is valuable because it shows where to investigate—but only after the reports are comparable.
Read the ratios as coordinates, not conclusions. The story begins with the raw amounts. It earns a business explanation only when the next piece of evidence rules competing stories in or out.
Sources
- “What is advertising cost of sales (ACOS)? Here’s everything you should know” — Amazon Ads
- “Streamline campaign analysis with unified reporting, now generally available” — Amazon Ads, June 8, 2026
- “Analytics Reports” — Amazon Selling Partner API
- “Returns Reports” — Amazon Selling Partner API
- “Sponsored Products” — Amazon Ads
- “Your complete guide to Amazon Attribution” — Amazon Ads
- “Sales & Traffic — Metrics Index” — Perpetua, March 9, 2026
- “Amazon Ads Multi-Touch Attribution” — Lewis et al., arXiv, August 11, 2025
