The Audience Your Data Partner Calls In-Market Was Scored Against Past Purchase Signals, Not Against Current Purchase Intent

When a data provider labels an audience segment as in-market for a category, the label sounds like a present-tense description. Someone is in-market. They are looking. They are ready. The label implies a moment, not a history.

In practice, most in-market audiences are built from a backward-looking model. The underlying signal is a pattern of past purchases, browsing sessions, or search queries that the provider collected over some trailing window, often 30 to 90 days, sometimes longer. A scoring model then ranks identifiers by how closely their historical behavior matches what buyers in that category tended to do before they converted in prior periods. The score describes similarity to a past behavioral fingerprint, not an observation of active shopping happening today.

That distinction matters for how you evaluate these segments before you activate them.

What the model is actually predicting

An in-market score is essentially a propensity score. The model says: identifiers that looked like this in the past tended to convert within some window. It does not say: this person opened a browser tab this morning and is comparing prices.

The model is trained on historical converters. Those converters had certain traits, browsing rhythms, and category touchpoints before they bought. The model learns those patterns and applies them to new identifiers. When the data provider surfaces an in-market segment to you, they are giving you identifiers whose recent behavior resembles the pre-conversion behavior of past buyers.

That is a reasonable and useful thing to model. The problem arises when buyers treat the resulting segment as if it were a feed of real-time intent signals rather than a population filtered by historical pattern matching.

Where the gap shows up in campaign performance

Consider a hypothetical example. A buyer activates an in-market auto segment expecting to reach people currently cross-shopping vehicles. Some portion of that segment will indeed be in an active shopping window. But another portion will include people who bought a vehicle three months ago and whose historical behavior still scores them as in-market because the trailing window the model used extends back further than their purchase date. They are in the segment because they looked like buyers recently. They are not in the segment because anyone verified they are still shopping.

The practical effect is that your campaign can serve impressions to people who have already resolved their need. They converted for someone else, or they converted for you on a prior flight, and the segment has not flushed them out yet. This is not a data quality failure in the traditional sense. The data is accurate. The model worked as designed. The issue is a mismatch between what the label implies and what the methodology delivers.

Questions worth asking your data provider before you activate

Most providers will answer direct methodology questions if you ask them clearly. Before you treat an in-market label as a live-intent indicator, it is worth understanding a few things.

First, ask what the lookback window is for the signals that feed the model. A 30-day window and a 90-day window produce very different populations. A longer window captures more people but includes more who have already moved past the purchase decision.

Second, ask whether the model refreshes the audience on a rolling basis and how often. Some segments are rebuilt weekly. Others are built less frequently and can sit for weeks before the identifiers inside them are re-evaluated. If the segment was last scored two weeks ago and the lookback window is 60 days, some identifiers inside it are being held on the basis of behavior from more than two months prior.

Third, ask whether there is any suppression logic for recent converters. Some providers attempt to remove identifiers that show post-purchase signals from in-market audiences, but this requires the provider to have a signal about the purchase event itself. If their data sources do not include that signal for your category, converters stay in the pool.

None of these questions are adversarial. They are the same questions you would ask about any modeling methodology. The fact that an audience has a confident-sounding name does not mean the naming convention maps precisely to the measurement question your campaign is trying to answer.

How to adjust your activation approach

If real-time or near-real-time intent is what your campaign needs, in-market segments built on historical purchase pattern models are a reasonable baseline but probably not sufficient on their own. You might consider layering them with signals that have shorter latency, such as contextual targeting environments where the page content itself signals active research behavior, or CRM suppression to remove known recent converters from the pool.

You can also use the historical nature of in-market segments productively. If your category has a long consideration cycle, a 60-day lookback may still be highly relevant because buyers in your category genuinely spend months in evaluation. The lag built into the model is a problem for categories with short decision windows and much less of a problem for categories where the journey from initial interest to purchase takes a quarter or longer.

Matching the model's lookback window to your category's typical purchase cycle is a simple calibration step that most buyers skip. If your average buyer takes 45 days from first search to purchase and the model window is 90 days, you are likely getting a reasonable signal for people still in the funnel. If your average buyer decides in a week and the model window is 90 days, the mismatch is structural.

What this means for measurement

When you measure performance against in-market segments, keep the modeling methodology in mind when interpreting results. If you see higher conversion rates from an in-market segment relative to a broader audience, some of that lift may reflect genuine propensity. Some of it may reflect the fact that historical purchase signals correlate with demographic and behavioral characteristics that convert well regardless of current intent. Without a holdout group and a clean incrementality test, you cannot separate the two.

Treating in-market segment performance as proof of intent-based targeting working is a measurement framing that the data may not fully support. The more careful framing is that the historical behavioral profile this segment uses to qualify identifiers is associated with better conversion rates in your category. That is worth knowing and worth using. It is also a narrower claim than the label suggests.