← All issues

The Impact Report · Issue 1 · October 2026

What the agentic era means for non‑endemic brands

A monthly note on what The Impact Lab is learning about measurement for brands that convert off Amazon.

What unBoxed and Ad Week mean for non‑endemic brands

Amazon spent unBoxed handing the decisions to agents. The ads console and the DSP are now one platform, Amazon Ads Agent. Full Funnel Campaigns and DVA+ let Amazon's models choose channel, creative, audience and pacing, and AMC is getting a tool that writes the SQL for you.

For a brand that sells on Amazon, that's a gift: the agent optimizes toward the sale, and Amazon sees the sale. For a non‑endemic brand, the outcome that matters happens off Amazon. A loan application is submitted, a first time deposit in a sportsbook account is made, someone walks into a dealership to take a test drive, and Amazon never sees it. An agent can only optimize toward what it can see, so without that outcome it optimizes toward the closest proxy: clicks, site visits, sign ups that never finish. It will get very good at finding people who start. The agent has no way to tell whether those people ever apply, deposit or show up.

Closing the gap: three ways to send the outcome back to Amazon

  • Amazon Ad Tag. A tag on the brand’s site that fires on key pages, such as a completed application. It’s the fastest to set up, but it only sees what happens on the website, not the funding or approval that comes later.
  • Conversions API. A server to server feed of conversion events from the brand’s own systems. This is how the real outcome can reach Amazon even when it happens days later and away from the site, and it’s the signal campaigns can be set to optimize toward.
  • AMC data uploads. Hashed first party records loaded into AMC, such as customers, outcomes and value. Uploads feed measurement and audience building rather than live bidding, so they answer “did it work” and “who should we find more of.”

In practice: the tag gets you started, the conversions API teaches the agent the real outcome, and uploads give measurement the full picture. In our experience the first two take days, not months.

The new AMC SQL generator: good news, with a catch

Amazon is releasing a tool that turns a plain English question into a validated AMC query, with general availability expected in 2027. That matters. Writing AMC SQL has been the main reason most brands and agencies never got past provisioning an instance. When anyone on the team can ask a question, AMC stops being a specialist tool.

The catch is that the tool writes whatever question you ask, and for non‑endemic brands the easy question is usually the wrong one. “Which audiences drove applications” will get a fast, clean answer. The question worth asking is closer to “which audiences drove funded loans from new customers, within the three weeks it takes a loan to fund.” Getting that right depends on things the tool can't infer on its own:

  • Where the real outcome lives. A funded loan or bound policy arrives through the conversions API or an AMC upload, not through Amazon’s purchase data. The query has to point at the right event.
  • The right window. Lookback windows built for retail purchases can miss outcomes that take weeks to land.
  • Who to leave out. Existing customers have to be excluded with the brand’s own customer list, or renewals get counted as new business.
  • Privacy thresholds. AMC suppresses small counts, so narrow questions can come back blank. Knowing how to restructure a question to get past that is a skill.

A query that runs isn't the same as an answer that's right. The bottleneck moves from writing SQL to knowing what to ask and whether the answer makes sense.

Two more to watch:

Publisher Insights in AMC. Now in beta, it adds show, genre and content signals. For non‑endemic advertisers on live sports and Prime Video, it's the first real way to ask which content reaches customers who complete the journey.

Apple's tracking block. We heard a lot about Apple's iOS update during Ad Week. The update started cutting Safari off from several cross site tracking and identity companies, and because every iPhone browser runs on Apple's engine, the block reaches all of them. If ID based targeting and measurement keep weakening, authenticated signals like AMC and a brand's own conversion data should become more valuable, not less.

Highlights by Vertical

What the work looks like in practice. Brands are not named, and some use cases combine more than one engagement.

Financial Services

Measuring a seasonal business before the season starts

The situation. A financial brand earns most of its year in a few months. Its Amazon budget for the season gets set in the fall, before any of this year’s results exist, so the case for Amazon has to be made with last season’s data or not at all.

What we’re doing, in three parts:

  1. 01A look back at last season in AMC. Which Amazon audiences and formats reached people who went on to convert, how long it took from first exposure to conversion, and the frequency at which more impressions stopped adding conversions. That puts a baseline into the fall budget conversation instead of a guess.
  2. 02Getting the brand’s own outcomes flowing before the peak. The completed conversion goes through the conversions API and is loaded into AMC ahead of the season, so the season is measured on real customers from day one rather than reconstructed afterward.
  3. 03In season reads every one to two weeks. Which audiences and formats are producing completed conversions now, so budget can move while there’s still season left to move it in.

Why it matters. Without this, the fall budget decision is a judgment call, and the brand finds out whether Amazon worked after the season is over and the money is spent.

Financial Services

Streaming spend measured on reach, not on customers

The situation. A lender or fintech buys Prime Video and live sports, often as part of an upfront commitment, and has no AMC instance. The read on that spend comes from DSP reports: reach, impressions, completion rate. They show who saw the ad, not whether any of them became customers.

What we do: the 10 day QuickStart:

  1. 01Provision AMC and take stock. Confirm which signals are already there: exposure across Prime Video, live sports and the DSP, plus site activity if the Amazon Ad Tag is live.
  2. 02Answer one question. Do people who saw the streaming ads go on to complete the journey at a higher rate than comparable people who didn’t? Where completion happens off Amazon, start with the furthest step Amazon can see, usually a completed application, and name the gap plainly.
  3. 03Read it out for the budget holder. One answer and one recommendation: keep, cut, or shift between live sports, Prime Video and the rest of the DSP.
  4. 04Pick the next data to bring in. Usually the completed outcome through the conversions API, which turns the next read from completed applications into funded accounts.

Why it matters. Streaming is often the largest line in a financial brand’s Amazon spend and the hardest to tie to customers. A read on who it brings in gives the brand something to defend it with when budgets get set.

Insurance

Finding the frequency where more impressions stop working

The situation. An insurer runs Amazon media across streaming and display with frequency caps set by habit or platform default. Set them too low and households never see the ad enough to act. Set them too high and the brand pays for impressions that add nothing.

What we do:

  1. 01Group exposed households by frequency. In AMC, sort households by how many impressions they saw over the campaign window.
  2. 02Measure the outcome at each level. Quote starts at each frequency band, and bound policies once that outcome is flowing through the conversions API.
  3. 03Find where the curve flattens. The point after which more impressions stop adding quote starts is the frequency worth capping at.
  4. 04Move the savings to new households. Impressions cut from the flat part of the curve go to reaching households that haven’t seen the ad yet.

Why it matters. Impressions past the point where the curve flattens are spend with no return. Moving them to new households buys more reach without a bigger budget.

Sports, Entertainment and Gaming

Measuring customer value, not just new customers

The situation. Advertisers in streaming, sports and gaming usually judge Amazon on acquisition: new subscribers, sign ups, first deposits. But the customers acquired vary widely in value. Some cancel after a free trial or play once and leave.

What we do, as ongoing support alongside the brand’s analysts:

  1. 01Bring the value signal into AMC. Revenue, activity or retention per customer, loaded as hashed records alongside the acquisition data.
  2. 02Compare activity, not just acquisition. Measure how customers exposed to Amazon media, including live sports and the brand’s own content, behave over time compared with customers who weren’t.
  3. 03Keep the heaviest users out of the average. A handful of the brand’s highest value existing customers can skew every result, so they are excluded from the comparison.
  4. 04Rank channels by the value of who they bring. Formats and audiences get judged on what their customers are worth, not just how many they deliver.

Why it matters. Acquisition reporting rewards the cheapest customers, not the best ones. When value leads the read, budget moves toward the channels bringing customers worth keeping, and that is the number finance trusts.

Agencies

Building an AMC practice inside the agency

The situation. A number of independent agencies want to offer AMC to their clients as their own capability, not resell a vendor’s. They have analysts and engineers but nobody who has run AMC at scale, and leadership needs to see the size of the opportunity before funding it.

What we do, in two phases:

  1. 01Executive education. A map of where AMC creates value across each agency’s client base, by vertical and use case, with a sense of how big each opportunity is, so leadership can decide where to start.
  2. 02Enablement through a crawl phase. We run the first client analyses alongside each agency’s team, hand over the query library and readout templates, and help define the senior AMC lead they should hire. Some agencies also want us working under their name for existing clients in the meantime.

Why it matters. An agency that rents AMC capability competes on price with every other reseller. One that owns it can put AMC into every pitch and every renewal.

Agencies

One brand, several agencies, data split across instances

The situation. A large brand works with more than one agency on Amazon, and each agency runs its own AMC instance for its share of the media. Each agency sees its own campaigns. Nobody sees the full path, so overlap between agencies’ media is invisible and each one reports the same customers as its own.

What we’re doing:

  1. 01Map what exists. Which instances there are, who owns each, and which campaigns and data sit in each one.
  2. 02Design a brand owned instance. One instance that holds the full picture, with each agency keeping its own view for day to day work.
  3. 03Agree how it’s governed. Who can run what, and how the brand’s own data flows in once rather than separately to every agency.

Why it matters. When every agency measures its own slice, the results they report can add up to more than the brand got, and nobody can see it. A brand owned instance is also the one that survives an agency change.

Automotive

Joining manufacturer, regional and dealer media

The situation. Car buying runs across three tiers of advertisers: the manufacturer’s national media, regional dealer groups, and individual dealers. Each buys its own media and none sees what the others ran, so when a car sells, nobody knows which exposure contributed.

What we’re designing:

  1. 01An aggregation layer across the tiers. One view of each tier’s Amazon media, without any tier seeing data it shouldn’t.
  2. 02A shared outcome signal. Lead and sales data that manufacturers and dealers often already collect, sent through the conversions API.
  3. 03A faster start where the signal already exists. For automotive advertisers whose conversions API is already live, a QuickStart that begins with the outcome in place.

Why it matters. Today each tier optimizes on its own, and the manufacturer cannot tell whether its national spend is setting up the dealer’s local sale.

Nonprofit

Finding the audiences that bring in first time donors

The situation. Nonprofits run paid media to acquire donors, but much of the response can come from people who already give. When reporting blends the two, a campaign can look healthy while bringing in very few new donors.

How we approach it:

  1. 01Separate new donors from existing ones. Load the donor file into AMC as hashed records so first time donors can be told apart from people already giving.
  2. 02Measure what reached first time donors. Which Amazon audiences and formats reached people who went on to make a first gift.
  3. 03Shift and suppress. Move spend toward the audiences that bring in new donors, and keep existing donors out of acquisition media.

Why it matters. Existing donors can be reached cheaply through email. Paid media earns its place by finding the donors a nonprofit doesn’t have yet.

What didn't work

Leading with a data upload. When the first conversation is about getting first party data into AMC, it stalls in privacy and compliance review. We now lead with a question AMC can answer on Amazon signals alone, and bring the data in second.

When to call us

If you work with a non‑endemic advertiser who would spend more on Amazon but can't connect that spend to results, call us. We get AMC provisioned, answer one business question in about ten days, and deliver a readout the brand team can take into a budget conversation.

One page briefs are available for the 10 day QuickStart, a streaming TV sample readout, the conversions API, and agency partnerships.

Start the conversationGet the next issue by email

Next month: what a first AMC readout should contain when the reader is the CMO, not the analyst.