Behavior Model.
It learns how your members actually behave — not how a segment says they should.
Segments describe members from the outside. The model learns them from their own actions.
A compact behavioral model, trained per brand on your members' own event stream — visits, purchases, points, opens, clicks, dozens of governed event types. From that sequence it learns each member's behavioral fingerprint and keeps a live read on what tends to come next: purchase or lapse, and when. One model per brand. Never pooled.
One member's event stream
Fingerprint
who behaves like whom
What's next
purchase vs. lapse odds
When
due window · engaged hour
one model per brand · trained only on your members · never pooled
Why a model
Attributes tell you who someone is. Behavior tells you what they'll do.
Rules go stale
“Bought 3× in 90 days” was a good rule the day you wrote it. Members drift; the rule doesn't. A learned model updates with every event instead of waiting for someone to re-tune thresholds.
Averages hide the person
A segment average says members like this buy every 40 days. This member buys every 12 — or 80. Per-member predictions replace one-size-fits-the-segment guesses.
Lookalikes need more than demographics
“Same age, same city” finds people who look similar. A behavioral fingerprint finds people who act similar — the same rhythm of visits, categories and responses. That's what predicts value.
What it produces
Three live reads on every member.
Behavioral fingerprint
A compact representation of how this member behaves, learned from their whole event history. It's what lets the platform answer “who else behaves like your best members?” — the engine behind lookalike audiences and behavioral similarity.
Next-event probabilities
For each member: how likely the next event is a purchase, and how likely they're lapsing instead — refreshed as new behavior lands, surfaced on the member profile and available to segments and the agents that decide.
Per-member timing
Each member's own purchase rhythm — when the next purchase is due, with an honest window — and the hour of day they actually open and click, feeding send-time optimization.
How it's built
Small on purpose.
This is not a giant foundation model that has read the whole internet. It's a compact behavioral model — deliberately small, so that every brand can afford its own: trained on your data alone, retrained as your base evolves, and cheap enough to re-verify constantly instead of once a quarter.
Small also means accountable. Training is deterministic — the same data produces the same model, bit for bit — and every model is versioned, so any score on any member can be traced to the exact model that produced it. When an auditor, a regulator or your own team asks “why this number?”, there's an answer.
And when the model is wrong — every model sometimes is — being small and per-brand means the blast radius is one brand's predictions, caught by continuous re-verification, with glass-box math standing behind it. Not a shared model quietly drifting for everyone at once.
| Pooled “big model” | Flash Behavior Model | |
|---|---|---|
| Learns from | Everyone's customers, pooled | Your members' events — yours alone |
| Reproducible | Practically untraceable | Deterministic & versioned — same data, same model |
| Retrains | On the vendor's schedule | As your base evolves — cheap enough to keep current |
| When it's wrong | Drifts silently, for everyone | Caught by re-verification; falls back to glass-box math |
The honesty gate
The model has to earn its job — on your data.
Every brand's model is tested against a transparent baseline on that brand's own held-out data — behavior the model never saw in training. Only if it genuinely predicts better does it go live. Until then, the platform runs on glass-box math — it never fakes a model it doesn't have.
And the test never stops: the model keeps being re-verified as your base evolves. If it stops earning its place, downstream features fall back automatically. No silent degradation, no black-box drift.
What it learns from
A governed catalog of behavior — not everything it could see.
The model trains on a curated catalog of behavioral events. Every event type is explicitly admitted — and some are barred on purpose, because a world model that learns the wrong things confidently is worse than no model at all.
Commerce
Loyalty
Engagement
Store
Barred from training — by policy
No feedback loops
Anything the model itself writes back — its own tags, its own scores — is barred from its training data. A model that learns from its own output ends up agreeing with itself; this one can't.
Behavior, never content
It learns that a member clicked — never what the message said, never free text, never anything a member wrote. The model knows rhythms, not words.
Restriction means restriction
Members who exercise their right to restrict processing are excluded from training end-to-end — enforced in the pipeline, not in a policy document.
Privacy, machine-enforced
A model of your members that respects your members.
Yours alone
One model per brand, trained only on that brand's members. Your model never learns from anyone else's base — or teaches theirs.
Forgets on request
When a member is erased, their data leaves the model too — purging and retraining are part of the deletion machinery itself, not a manual afterthought.
Governed events only
The model trains on a governed catalog of behavioral events — not on free text, not on message content, and never on data a member asked you to restrict.
The other half of the world model runs this one forward — Digital Twin →
In the product
You'll meet it in four places.
There's no “model console” to babysit. The model's reads surface inside the tools your team already works in:
Member profile
Each member's page carries the live read: next-purchase odds, churn risk, expected timing — beside the facts they're computed from.
Members →Segment builder
Model outputs are targeting fields: build audiences on purchase probability, churn risk or predicted timing the same way you'd filter on tier.
Segments →Lookalike audiences
Seed with proven members; the fingerprint finds the nearest neighbors in behavior — the audience expands from evidence, not demographics.
Ad Audiences →Agent briefs
When SoClaw weighs a win-back, the brief it reasons over already contains the model's read: risk, value at stake, timing.
SoClaw →One event, one cycle of the loop — no exports, no overnight batch, no manual re-scoring.
The model's reads surface on the member profile, in segments, and in the agents' reasoning.
Related reading
Keep exploring the pages most related to this one.
Intelligent Decision
A next-best-action: who, which offer, which channel, when.
Read more CapabilityDashboard
Real-time scan-to-redemption data as KPI trends, funnel diagnostics, and a rule + LLM next-best-action panel — plus retention curves by signup cohort and a 14-day revenue outlook that's honest about being a projection, not a commitment.
Read more CapabilitySoTag for Slack
The Slack agent that hands governed, audited numbers to Claude in-thread: daily anomaly scan + human-approved coupon-send. Early access.
Read more CapabilitySoClaw — Hire your AI employees
Hire AI employees for your loyalty program — autonomous agents that each hold a real job: a Loyalty Manager that wins back at-risk customers, an Order Manager that clears the receipt and order review queue, a Risk Manager that protects points from fraud, and a Marketing Manager that turns a goal into a ready campaign. Each works one case at a time within limits you set and logs every decision; you manage them like a team with goals, budgets, tasks and feedback. Off by default, capped, instantly pausable. Early access.
Read more CapabilityAI Governance & Approvals
Every AI action runs on a per-action authorization gear — automatic, human review, or forbidden — with an approvals console, a complete decision log, one-click undo, and automatic demotion to human review when an action starts misbehaving.
Read more CapabilityBusiness Ontology
The concept behind the governed AI stack: a world model of your business — objects, properties and relationships plus governed actions — as a metadata layer mapped over the data you already have, so AI knows what a customer is, what has happened, and what it may do about it.
Read more