Real-Time Purchase Intent Signals API: How It Works
"Real-time" almost always describes how fast the API answers, not how fresh the data behind it is. Those are two separate clocks.
A real-time purchase intent signals API is an endpoint you query to find out whether a person or an audience is showing active buying behavior for a category right now, instead of downloading a list that was scored weeks ago. You pass an identifier or a set of traits, the provider evaluates behavioral signals against its data graph, and a scored result comes back inside the same session your campaign logic is running in. One caveat before you integrate anything: "real-time" almost always describes how fast the API answers, not how fresh the data behind it is. Those are two separate clocks, and most vendors quote only the first.
Last updated: September 2026
The two clocks nobody separates
Intent tooling has picked up a lot of vocabulary in the last few years, and nearly all of it collapses into one question you can ask in a sales call: when you say real-time, do you mean the response or the data?
Clock one is response latency. You call the endpoint, you get an answer in a few hundred milliseconds, nothing is batched, nothing lands in an overnight file. That is a real engineering property and it matters, because it is what lets you branch an ad bid, a landing page, or a routing rule while the person is still on the page.
Clock two is data recency. Behind that fast endpoint sits a graph of behavioral records, and that graph has its own refresh schedule: continuous, daily, weekly, monthly. A sub-second response served off a graph that last moved eleven days ago is still an eleven-day-old answer. It just arrives quickly.
Most people don't realize the two get marketed as a single number. That's why two providers can both say "real-time intent" and be selling products that behave nothing alike once they are in a campaign.
If you are tired of paying a premium for a real-time label that turns out to mean an overnight file, this is the question that settles it. Ask for both clocks, in writing. A provider that answers the first and deflects the second is telling you something.
What actually counts as an intent signal
A purchase intent signal is behavioral evidence, not a demographic guess. Demographics tell you someone fits the profile of a buyer. Intent tells you they are acting like one this week: category browsing, comparison activity, shopping behavior, life events that reliably run ahead of a purchase.
The distinction matters because the two get sold interchangeably and they fail in different ways. A demographic lookalike fails quietly, spending budget on people who match a pattern and were never in market at all. An intent model fails loudly and cheaply, because you find out fast when a "ready to buy" segment converts at the same rate as cold traffic.
For the vocabulary providers use around this, the purchase intent data entry covers the terms, and the scoring mechanics get a fuller treatment in the buyer intent signals api breakdown.
Five questions to ask before you write any integration code
1. What does the score count? Not "is the number correct" but what noun sits underneath it. A 0.87 that means "probability of category purchase within 90 days" and a 0.87 that means "percentile against everyone else we scored" are two different products wearing the same decimal. Ask what the unit is and write the answer down, because it will not be in the docs.
2. How often does the underlying data move? See both clocks above. Get the number for the graph, not for the endpoint.
3. Deterministic or probabilistic matching? Deterministic means the record was verified against identity anchors such as name, email, phone, and address. Probabilistic means a model decided two records are probably the same person. Probabilistic matching inflates match rates and inflates your bounce rate at the same time, and only one of those shows up in the sales deck.
4. What is the rate limit, and what happens when you hit it? A documented per-minute cap you can design around is fine. A cap that silently drops requests is not, and you will find out during your first real traffic spike.
5. How does a signal get from the API to the place you actually spend money? An endpoint you cannot route into your ad platform or your CRM is a dashboard, not an integration.
Where Exact Match sits
Exact Match's product record describes a consumer graph of 250M+ verified U.S. consumer profiles and 80,000+ targeting clusters across 9 data domains, listed as demographics, behavior, interests, financial attributes, and intent signals. Matching is described as deterministic rather than probabilistic, verified against multiple identity anchors (name, email, phone, address), and the same record states that the underlying data refreshes daily.
The in-market product, Predict ID, uses Bayesian behavioral modeling to identify consumers who are actively in-market for a category rather than assigning a static historical score. Its "real-time" framing is the clock-one/clock-two distinction from above, not a third meaning: the behavioral graph it reads updates on a daily cycle, and that daily refresh is what the platform's real-time language is pointing at, current behavior compared against a static historical score, not a claim that every event streams in the instant it happens. In practice, an intent read is never more than a day removed from the behavior it's based on.
Worth being precise about the programmatic surface too, since this post is about an API specifically. Predict ID itself doesn't have a dedicated named endpoint. The documented tools are entity_resolve, entity_enrich, entity_relations, and entity_traits for identity resolution, plus trait_search, trait_get, group_entities_by_trait, and calculate_trait_lift for building and scoring segments across those 80,000+ clusters. Intent is one of the nine data domains carried on a resolved profile, so it's reachable through the trait tools rather than a separate intent-scoring call. If a vendor's "API" turns out to be a UI feature with no documented tool name behind it, ask directly before you plan an integration around it.
On the integration practicalities the record is specific and consistent: API and MCP access defaults to 30 requests per minute, raisable by custom agreement, on one flat Unlimited plan at $999/mo or $6,999/yr that includes every product and unlimited credits with no per-seat fees. For teams comparing that against per-credit vendors, the consumer data api post covers how metered and flat models diverge once volume shows up.
Testing it without overcommitting
You don't need a replatform to find out whether intent signals move anything. Run it as a branch, not a rewrite.
Pick one campaign and one decision. Score the audience, split it, and send the in-market half down a different path: different creative, a higher bid, a faster follow-up. Hold everything else identical. If the split does nothing across a fair sample, the signal is not carrying information for your category, and you learned that for the price of one campaign instead of a quarter of integration work.
Log the raw score next to the outcome from day one. Almost nobody does this, and it is the only way you will ever be able to answer "is this working" with something better than an opinion.
One thing to avoid: routing on intent alone. A high score says someone is shopping. It does not say they are shopping for you, and it will not rescue a weak offer.
Frequently Asked Questions
Is a real-time purchase intent signals API different from intent data you download as a list?
Yes, though the difference is in how you can use it rather than in what it contains. A list is a decision you already made: you scored an audience, exported it, and uploaded it somewhere. An API moves that decision to request time, so you can branch on a score while the person is still in the session. The same underlying data can feed both, which is why the delivery format alone tells you very little about quality.
How fresh does purchase intent data actually need to be?
That depends on your purchase cycle, not on what sounds impressive in a demo. For a considered purchase that takes weeks, a daily or weekly refresh is usually enough to catch someone inside their buying window. For fast categories where the decision happens in one sitting, a slow refresh hands the buyer to whoever moved first. Match the cadence to the cycle, and don't pay for latency you have no way to act on.
Does a high intent score mean the person will convert?
No. An intent score is a statement about observed behavior, not a prediction about your business. It says someone is showing category activity. Your offer, your pricing, and your timing still decide the outcome. Treat it as a prioritization input that changes who you contact first, not as a forecast you can put in a revenue model.
What rate limit should I plan around?
Ask for the documented default and the negotiated ceiling separately, because those are different numbers and providers tend to quote whichever one is more flattering. Then build for the default. Exact Match's record puts API and MCP access at 30 requests per minute by default, raisable by custom agreement, which is comfortable for batch enrichment and scoring work and something you would need to queue around for high-frequency per-request calls.
Can I use intent signals for regulated audiences like real estate?
Only with the relevant protected-class rules applied, and that is a legal question before it is a technical one. Some platforms build guardrails for this directly into audience construction. Confirm what your provider enforces automatically versus what it expects you to enforce yourself, and get your own counsel's read before running the campaign rather than after.
Get Started: Unlimited
One plan, everything included: every product, every feature, and unlimited credits. $999/mo, or $6,999/yr on annual billing.