Affinity and Propensity Scoring

An affinity score is a per-contact number — typically 0 to 100 — that estimates how likely a contact is to buy a specific product, attend a specific event, or engage with a specific category, artist, team, or theme. Simon AI builds affinity scores from the data already in your warehouse and saves them as AI-generated datasets, so you can threshold them into segments and activate them in flows and journeys like any other field.

This is how marketing teams move from personalizing for a handful of hand-picked products or events to personalizing for hundreds or thousands: instead of a person deciding who should hear about each one, the agent scores every contact against each target and you pick the threshold.

This capability is part of AI Chat.

🟦

Core AI · Included with your Simon platform.

What an affinity model is

An affinity model is the scoring logic behind an affinity score. For a given target (say, an upcoming concert, a product line, or a category), it combines several kinds of evidence about each contact into a single ranked score:

The agent normalizes the result so scores spread across the range rather than piling up at 0 or 100, which is what makes thresholds useful for segmentation.

What signals feed it

The agent uses whatever relevant tables it can see in your connected schema. The more of these you have connected, the stronger the model:

• Purchase and order history — line items, categories, brands, venues, price tiers, order dates
• Web and app behavior — product and event page views, searches, favorites, wishlists, tracked items, carts
• Campaign engagement — email and SMS opens and clicks, which reveal what a contact responds to
• Catalog attributes — the product, event, or content metadata that lets the agent reason about similarity (genre, category, tags, performer, venue)
• Customer-provided third-party signals — data you already load into your warehouse from partners or your own models, such as streaming or listening preferences, enrichment attributes, or the output of an in-house propensity model. The agent can use these as inputs alongside first-party behavior.

If a signal isn't in your warehouse, the agent can't use it. Connect the table through Schema Builder and the agent will find it.

How to build one

Open AI Studio > Chats and describe the target and the outcome you care about.

Example prompts

• "Score every contact 0–100 on their likelihood to buy tickets to Artist at Venue on Date. Weight direct artist engagement highest, then genre affinity, then general event behavior."
• "Build an affinity score for our running-shoe category using purchase history, browse behavior, and email clicks."
• "Create a dataset with a purchase-likelihood score for each product line, one field per line."
• "Rank contacts by affinity for outdoor gear and tell me how many are above 70."

The agent then:

1. Explores your schema to find the relevant purchase, behavior, engagement, and catalog tables
2. Proposes the scoring logic, including the tiers and weights it plans to use
3. Shows you the SQL, an explanation, and sample values for review
4. On approval, saves the score as an AI-generated dataset
5. Lets you set a refresh schedule — one-time, or recurring so scores stay current as behavior changes

📘

You review before anything is saved

Nothing is created until you approve the proposal. Ask the agent to change a weight, add a signal, or exclude recent purchasers, and it revises the logic before saving.

Using the score

Once saved, an affinity score is a normal field:

• Segment on it — from the field's details page in Schema Builder, click Create segment with this field, or ask the chat for "contacts with a score above 70"
• Tier it — build "very likely" and "somewhat likely" segments at different thresholds and treat them differently
• Activate it — use the segment in a flow or journey, or sync it to your engagement platform as an audience
• Personalize with it — reference the score or a derived label in content

Combine affinity scores with AI Moments to reach the contacts most likely to act when a real-world trigger fires, and with Projects to keep the scores, segments, and content for a campaign together.

Scaling to many targets

The pattern above is per target. To cover a full catalog or event calendar, ask the agent for a dataset with one score per target, or run the same prompt for each new target as it comes up. Because each score carries a refresh schedule, scores for ongoing targets stay current without anyone re-running them.

How this differs from other Simon AI features

Availability

Affinity and propensity scoring is part of AI Chat and AI-generated datasets, available to all AI-enabled organizations.