AI Agents · Predictive Intelligence
Predictions you can read, check, and act on.
Who's likely to leave and what they're worth. When each member is due to buy again. Who'll click, which products go together, and what the next two weeks look like. Every score is computed from your own data with the math visible— and when the data can't support an honest answer, the platform shows nothing rather than a guess.
What it answers
Six questions every retention team asks — answered from your own data.
Who is worth saving — and who changes their mind?
Every member profile shows churn risk and predicted 12-month value, computed from that member's own purchase pattern.
The win-back agent works its list by revenue at risk — and as your program's own control-group history builds, it learns where a nudge actually changes the outcome, so a big spender who was coming back anyway stops outranking a member who genuinely needs the outreach.
SoClaw win-back →When is each member due to buy again?
Each member's own buying rhythm, projected forward: an expected next-purchase date with a window that reflects how regular they actually are — and overdue shown as overdue.
One segment filter — "days until expected next purchase, between −14 and 3" — is a ready-made replenishment-reminder audience.
Segments →Who will respond — and when should you send?
A click likelihood for every member you actually email, learned from your own sends — build high-responder audiences, or find members where email isn't the right channel.
And the schedule step reads when this audience actually opens and clicks, recommending the hour they're most responsive — one click applies it, or automated sends can apply it hands-free. If the audience lacks history, it says so and keeps your time.
Campaigns →Which products go together?
Product pairing corrects for popularity so genuinely related products surface — not just bestsellers riding along with everything.
Cross-sell suggestions quote their evidence: "buyers of X go on to buy Y in 23% of orders — 3.1× more often than average" (illustrative), with the raw order counts stored so anyone can recompute them.
Recommendations →Who else looks like your best members?
Pick a winning segment or a handful of members as a seed, and Flash finds the members who most resemble them — inside your own base, never uploaded to an ad platform.
Every match explains itself: a "Why they're similar" summary shows exactly what it's built on — shared traits, buying rhythm, spend tier, favorite category, churn outlook — and skips any factor it has no data for.
Segments →What do the next months look like?
A 14-day revenue outlook from your own trend and weekly rhythm — a dashed line that is labelled what it is: a projection, not a commitment.
Retention curves by monthly signup cohort answer the question that matters most: are newer members staying longer than older ones did — with the definition of "churned" printed on the chart, and a short list of what retention correlates with in your own history, labelled correlation, not causation.
Dashboard →Why you can trust it
The glass-box contract.
Most predictive marketing asks you to trust a score you can't inspect. Ours makes the opposite deal: every number can be traced, and no number appears without the history to back it.
Your data only
Every model is refit nightly on your own program — your purchases, your sends, your experiments. No pooled cross-customer models, no industry averages dressed as predictions.
The math is visible
Wherever a score is used, its basis is shown: profile tiles state what they were computed from, win-back tasks carry the arithmetic behind their rank, cross-sell rules keep the raw order counts so anyone can recompute them.
Silence over guesses
Too few purchases → no next-purchase date. Barely emailed → no click likelihood. A young program → no forecast. Every prediction has a stated minimum of real history, and below it the platform shows nothing rather than something invented.
It never acts alone
Predictions rank, sort and suggest. Sends, coupons and points still pass the same approvals, caps, frequency rules and consent checks as always — a score can't spend a cent by itself.
On the member profile
Every number explains itself.
Churn risk
72%
from her purchase recency & frequency
Expected next purchase
Jul 18 · ±4 days
her own rhythm: about every 32 days
Click likelihood
63
from your last 90 days of sends
Illustrative figures. A member without enough history shows no tiles at all — never an invented number.
Where you meet it
Not a separate tool — the layer under the tools you already use.
There's nothing to configure and no model to manage. Scores refresh nightly and show up where the work happens.
Member profile
Churn risk, predicted 12-month value, expected next purchase, click likelihood — with recommended actions that quote the numbers.
Learn more →Segments
Predictive audience conditions: click likelihood and days-until-expected-purchase (negative = overdue).
Learn more →Win-back worklist
Ranked by where outreach changes the most revenue; every task shows why it ranked where it did.
Learn more →Recommendations & email
Popularity-corrected product pairing and auditable cross-sell across email, site and app.
Learn more →Dashboards
Revenue outlook and retention-by-cohort curves, fine print printed on the chart.
Learn more →Points protection
A statistical sentinel flags earning patterns far outside your program's own normal — freeze for review only, never auto-punish.
Learn more →The complete list
Thirteen predictions. One contract.
Every prediction shipping today, and where it lives — each under the same glass-box contract above.
Predictions propose. Experiments prove.
A prediction is a hypothesis until real members respond. The same platform runs honest A/B tests and keeps untouched control groups — and on recurring campaigns it can shift traffic toward what's working while the test runs, every shift logged. What the models suggest gets measured; only proven lift gets reported.
See Experiments & A/B →Straight answers
The questions a careful buyer should ask.
Is this black-box AI?
No. Every prediction is a piece of transparent statistics computed from your own program's data — and everywhere a score is used, the inputs behind it are shown: a profile tile states what it was computed from, a win-back task shows the arithmetic behind its rank, a cross-sell suggestion keeps the raw order counts. If a number can't be explained, we don't ship it.
What happens when there isn't enough data?
Nothing is shown — deliberately. A member with three purchases doesn't get an invented next-purchase date; a member you've barely emailed doesn't get a click likelihood; a program with under a month of sales doesn't get a revenue outlook. Every prediction has a stated minimum of real history behind it, and below that bar the honest answer is silence, not a guess.
Can predictions send messages or spend money on their own?
No. Predictions rank worklists, sort audiences and suggest actions. Anything that actually touches a member — a send, a coupon, a points grant — still goes through the same approvals, budget caps, frequency rules and consent checks as everything else on the platform.
Where do the scores come from?
From your own members' behavior — purchases, engagement with your sends, and your program's own experiments — refit nightly. There are no pooled cross-customer models and no industry benchmarks dressed up as predictions: your scores describe your business.
The dashboard's forward-looking views print their fine print on the chart — a projection, not a commitment.
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 — AI Win-Back
An autonomous AI teammate that wins back high-value, at-risk customers one at a time — a personalized offer or a deliberate hold — and proves the real incremental lift against an untouched control group. Its worklist is ranked by where outreach changes the outcome — revenue at risk × each member's measured persuadability from the program's own control group. 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