Path to Conversion: Reading the Report Properly

Where this data actually lives

Sponsored Products, Sponsored Brands, Sponsored Display and DSP each report their own attributed conversions in isolation, using their own attribution windows and their own definition of a converting touch. None of them natively shows you the sequence of ad exposures a single shopper experienced across all four before buying. That sequence — the actual path — only exists as a queryable object inside Amazon Marketing Cloud, where pseudonymized impression and click events across every Amazon ad product can be joined back to the same shopper and ordered by timestamp.

That's worth stating plainly because a lot of confusion in this category comes from people looking for a "path to conversion" button in the standard Sponsored Ads or DSP console and not finding one. It isn't missing — it was never there. It has to be built as a query.

What a properly-built path actually contains

A well-built AMC path-to-conversion query returns, per purchase, the ordered sequence of ad touches that preceded it within your chosen lookback window: which ad product, which campaign, what format, and the timestamp of each. Aggregated across many purchases, this lets you answer genuinely useful questions — what share of purchases had more than one touch, what the most common two-step and three-step sequences look like, and which channel combinations show up disproportionately in converting paths versus the account's overall traffic mix.

It's worth naming what the query returns that a single-channel report can't: overlap. A DSP display touch and a Sponsored Brands click on the same path, from the same shopper, will show up as two separate "wins" in two separate native dashboards — path data is the only place that overlap becomes visible as one sequence rather than two competing claims on the same sale.

The mistake almost everyone makes reading it

The most common misread is treating path frequency as proof of importance — assuming that because a channel appears in 70% of converting paths, it caused 70% of those sales. Appearing in a path only means the shopper was exposed; it says nothing about whether that exposure changed the outcome. A channel with wide reach and heavy frequency capping — Alexa device inventory in our own DSP data delivers 34.9% of all impressions for just 3.0% of spend — will naturally show up in a large share of converting paths simply because it reaches so many people, independent of how much actual influence any single touch carried.

Path-to-conversion data answers "what happened before the sale." It does not answer "what caused the sale." Confusing the two is the same underlying error that makes last-click attribution misleading, just wearing a more sophisticated-looking report.

A worked reading of a real pattern

Say an AMC path query returns this: of 1,000 purchases in a month, 640 had only one attributed touch, 280 had two touches, and 80 had three or more. Of the two-touch paths, the most common sequence — appearing in 190 of the 280 — was a DSP display impression followed by a branded search click. That's a genuinely useful finding: it tells you two-thirds of your multi-touch converters are following a specific, repeatable pattern, which is worth checking creative and audience alignment against. It does not tell you that removing the DSP display touch would have cost you those 190 sales — that's a separate, causal question, and answering it requires a holdout, not a bigger path query.

The common mistake, including ours

We've presented a path-frequency finding — "DSP appears in X% of converting paths" — to a client as evidence the DSP spend was working, in a context where the honest framing needed a second sentence about what the number could and couldn't prove. It wasn't false; DSP genuinely did appear in that share of paths. But without the causation caveat attached, the number implicitly did more persuasive work than it had earned. We now pair every path-frequency finding with either an incrementality result or an explicit note that one hasn't been run yet, rather than letting the path data stand in for a claim it can't support alone.

Setting up a path query that gives you something reliable

Two settings matter more than most people check. First, the lookback window: set it to genuinely cover your category's typical path length, not a generic default, or you'll systematically truncate real early touches out of every path you build — Amazon Marketing Cloud's extended traffic lookback now supports considerably longer windows than any single ad product's native reporting. Second, decide up front how you'll handle deduplication when the same shopper has near-simultaneous touches across two placements — a display impression and a video impression served within seconds of each other on the same page load can register as two distinct touches when they were really one exposure, and an unhandled duplicate inflates whatever channel served both formats.

How to use path data well

Use path-to-conversion queries for what they're actually good at: spotting common sequences worth testing, checking whether channels are structurally reaching the same shoppers redundantly, and identifying candidates for an incrementality test — a channel that shows up in a large share of paths but whose real contribution is unclear is exactly the channel worth holding out next. Don't use path frequency alone to justify a budget decision; pair it with a causal test before treating any pattern in the sequence data as settled.