ACoS 与 TACoS 有何区别，2026 年 Amazon 广告该看哪个指标

ACoS versus TACoS comes down to one word in the denominator. ACoS measures advertising efficiency: ad spend divided by ad-attributed sales. TACoS measures advertising impact: ad spend divided by total sales, organic included. A campaign can post a beautiful ACoS while the business underneath it stalls — which is why TACoS, not ACoS, should lead your dashboard in 2026. For more than a decade, Amazon advertising tools were built around ACoS. It was the north star for how sellers measured campaigns, agencies reported progress, and software companies designed dashboards. That consensus is breaking down — and with the average account running roughly 32–34% ACoS against $1.18–$1.22 CPCs, the blind spots are no longer affordable. The shift is from a reporting paradigm to a diagnostic one. Reporting tools tell you what happened. Diagnostic tools tell you why it happened and what to do about it. For most of Amazon's history, that interpretation layer was the agency's job — sellers paid a typical $2,500–$5,000 monthly retainer for someone to read the data and recommend actions. AI has closed enough of that gap that the diagnostic layer now ships inside seller software, and increasingly inside Amazon's own console. Here are the two headline metrics side by side, the decision framework that connects them — the TACoS Quadrant — and the four diagnostic capabilities that separate the new generation of tools from the dashboards most sellers still use. ## ACoS vs TACoS: The Difference in One Table ACoS divides ad spend by ad-attributed revenue and answers a narrow question: is this campaign profitable in isolation? TACoS divides the same spend by total revenue — organic included — and answers the question that actually predicts business health: is this advertising creating growth, or just shifting where sales get attributed? The math shows why the distinction matters. A campaign with 50% ACoS and 12% TACoS is dramatically more valuable than a campaign with 20% ACoS and 22% TACoS, even though the first looks worse on a standard dashboard. The reason is the halo effect: when ads drive organic ranking improvements, those organic sales never appear in the ACoS calculation. A campaign that costs $500 to generate $1,000 in attributed sales (50% ACoS) but lifts organic sales by $5,000 in the same period actually produced $6,000 of revenue for $500 of spend — an 8.3% TACoS. The healthy number is invisible in standard reporting. Agency guidance has converged on the same point — Canopy Management's guide describes TACoS as a business health metric, not just a campaign efficiency metric. That is the difference between optimizing a campaign and optimizing a business, and it is why the rest of this post treats TACoS as the headline number. ## The TACoS Quadrant: Scale, Fix, Invest, or Pull Back The TACoS Quadrant is the framework I use to turn the two headline metrics into a decision. Read your ACoS against your break-even, read your TACoS trend over 8–12 weeks, and the account lands in one of four states — each with one correct move. It ends the meeting where one person cites ACoS and another cites revenue and nobody decides anything. Break-even ACoS is simply your pre-ad contribution margin: at a 35% margin, ACoS above 35% loses money on the attributed sale. The quadrant's power is the combination. A 25% ACoS holding a 12% TACoS is a healthy flywheel that deserves more budget. The same 25% ACoS against a TACoS drifting from 18% to 30% means organic is decaying underneath the ads — and more spend makes that worse, not better. ## The Cannibalization Tax Most Sellers Pay Without Seeing It Cannibalization is paying for clicks on keywords where you already rank high organically. The product appears twice on the results page — once as a paid ad on top, once in the organic results below — the ad soaks up the click, and you buy traffic you already owned. It is the single most expensive blind spot in ACoS-only reporting, because the wasted spend still converts and still looks efficient. Across the 57 accounts I managed at Worldfront, the directional numbers I planned with: at organic rank #1, roughly 85 cents of every dollar spent on that keyword bought traffic we already owned; at rank #2, roughly 65 cents; at rank #3, roughly 45 cents; below the top three, the effect fell off quickly. Treat those as planning heuristics, not physics. The honest way to get your own numbers is a holdout test: 1. Pull the Search Query Performance report in Brand Analytics and list every keyword where you rank top-3 organically and still spend meaningfully. 2. Pause or sharply cap bids on that list for two full weeks — long enough to smooth daily noise, short enough to limit rank risk. 3. Compare total sales (ad plus organic) for those ASINs against the prior two weeks. Total sales, not ad-attributed sales, is the whole point. 4. If total sales dropped less than about 10%, the spend was mostly cannibalization — redeploy it to keywords where you rank below the fold. 5. If total sales dropped 30% or more, the ads were genuinely incremental — restore them and re-test quarterly. 6. Check new-to-brand rate as the supporting signal: NTB below about 20% on a campaign means you are mostly re-converting customers you already own. The reason most sellers never see any of this: the Ads Console does not show organic rank alongside ad performance. The two data sources live in different reports with different update cadences, and connecting them manually through Search Query Performance is tedious enough that almost nobody repeats it. Tools that join both streams automatically — m19 built its Top of Search Rankings Optimizer specifically around this overlap, and it is the pattern across the new generation — make the tax visible by default. In my experience the recoverable waste is typically 1–3% of total revenue. The defensive case for some overlap is real — keeping a competitor out of your branded top-of-search is worth paying for when they are actually bidding. But the old default of bidding on everything that converts has aged badly. The better default is bid where you do not already win, and pay for defense only where there is a live threat. ## Search-Term Harvesting Needs Context Rules Can't Provide Threshold harvesting — promote any search term with three or more orders, ACoS under target, and 14+ days of data into a manual exact-match campaign — works for the cleanest cases and fails on the edges where the interesting decisions live. It promotes false positives inflated by promotions, and it misses durable early candidates that have not yet cleared the order threshold. The false positive: a term hits the thresholds during a Lightning Deal or coupon push, gets promoted to exact-match at peak bid just as the promotion ends, and the campaign bleeds on a keyword that no longer converts at its two-week rate. The missed candidate: a term with a strong conversion rate but young data gets harvested late, after more attentive advertisers have bid it up, and you pay a higher CPC for arriving last. The fix is contextual evaluation, and it is now an AI job. Beyond the threshold check, the questions that matter: Is the conversion pattern stable or seasonal? Does the term match the advertised ASIN's actual use case, or is it converting on related-product spillover? Is the competitive landscape on that term sustainable for exact-match bidding? What destination match type actually fits this candidate? A scored harvest recommendation that weighs seasonality, ASIN fit, and conversion stability — and explains its reasoning so you can override it — produces meaningfully different decisions than a threshold check alone. The underlying shift is from rules to judgment. Rules are fine for the clear-cut middle and dangerous on the margins, and the margins are where the money is. ## Metrics Without Narratives Are Operationally Useless When ACoS jumps eight points, the chart is not the answer — the cause is. Anomaly narration synthesizes the signals around a metric move (competitor entries, listing suppressions, inventory cover, bid changes), ranks the likely causes, and recommends a response. It replaces the ninety-minute Monday hunt through five reports that every advertiser knows too well. The data to answer the why has always been technically available. Search-term reports show the new terms. The Account Health Dashboard shows suppression events. Inventory reports show days of cover. The change log shows recent bid moves. The problem is synthesis: no human pieces together five reports across three interfaces every time a metric moves. The output of the new diagnostic layer reads like an analyst's note instead of a chart: The output has to stay verifiable and overridable — every claim referencing specific data, every recommendation reversible. Sellers do not have to accept the AI's interpretation. They just no longer have to do the diagnostic legwork before they can evaluate it. ## The Diagnostic Four: Grading Your Current Toolset Four questions grade any Amazon advertising tool in 2026. Does the dashboard lead with TACoS or only show ACoS? Can it show where you pay for clicks you already win organically? When it recommends harvesting a term, can it explain why the candidate is durable rather than seasonal? When a metric moves, does it name probable causes — or just chart the movement? • TACoS-first reporting — the headline metric is ad spend against total revenue, with ACoS demoted to a campaign-tuning stat. • PPC/organic overlap detection — organic rank joined to ad spend by default, so cannibalization is visible without a manual SQP cross-reference. • Contextual harvesting — recommendations scored on seasonality, ASIN fit, and competitive sustainability, with reasoning you can override. • Anomaly narration — metric moves explained with ranked probable causes and a suggested response, not just an alert. The era of treating Amazon advertising as a metric-display problem is ending. The era of treating it as a diagnostic problem has begun — and Amazon shipping its own free analyst layer just moved the deadline up for every tool, and every seller, still reporting ACoS alone.