Fourteen proprietary models re-read your store every night. The intelligence layer reads all of them at once, collides the findings, and hands you a short ranked list of actions worth money this week — not fourteen dashboards to check.
Every proprietary model in LiftSignals, named in plain language with the signal it emits underneath. This list is definitive — if it isn't here, we don't ship it. Full field-guide entries live on Customer signals.
Fourteen models produce thousands of scores a night. Left raw, that's homework. The intelligence layer runs the same four moves a great analyst would — every night, without being asked.
Every fresh score from all fourteen signals, plus what changed since yesterday — the overnight diff is where the news lives.
High churn risk and high lifetime value is an urgent save; high propensity and full-price tolerance is a coupon you shouldn't send. Signals mean more together than alone.
Each candidate action gets sized in dollars — audience × likelihood × value — and sorted. The list leads with the biggest number, not the newest feature.
Frequency caps, cooldowns, and confidence floors filter the list. An action that would burn your list — or that the models aren't sure of — doesn't make the cut. Silence is a recommendation.
Growth plans get ten of these a week; Starter gets five. Each one is a finding, its evidence, and a one-click action — never a chart to interpret.
Plain language, one sentence. Which customers, what's happening, how urgent.
The exact signals that fired — with scores, so you can interrogate it, not just trust it.
Sized in dollars, not percentages — so ranking against the other nine is honest.
One click to a governed audience push — caps, cooldowns, and the holdout already applied.
Predictions you can't interrogate are just confident guesses. Everything the layer says comes with its accuracy, its drivers, and its training data on display.
Every model's performance — AUC, precision, forecast error — is visible in Model Studio and updated with every retrain. Churn detection typically lands 87–94% precision across stores. If a model has a bad month, you see it before we mention it.
Every model trains per workspace, on your order history alone — never pooled across merchants, never used to help a competitor. The connection is read-only, and GDPR export and deletion are self-service.
Models need roughly 500 customers and 90 days of order history to score reliably. Below that, scores are marked provisional — directional guidance, clearly labeled, never dressed up as certainty.
Insights that push audiences carry an automatic holdout, so the expected value gets checked against a control group when the window closes. The layer that makes the promise is graded on it — that's the whole point.
The know-your-customers set — churn, RFM, reactivation, returns, discounts, CLV, forecasting, geography — with five ranked insights each week.
Adds the predictive five — propensity, next product, market basket, CAC & ROAS, next-best-action — and doubles the weekly list. Where the layer earns its keep.
Adds media mix modeling, runs the full layer in every client workspace, and feeds the white-label reports clients actually read.
A finding, its evidence, and a one-click action — generated overnight by the synthesis layer reading all fourteen signals together. Each one names the customers involved, shows the scores that fired, sizes the opportunity in dollars, and ends in a governed audience push. It's the analyst's memo, not the analyst's spreadsheet.
Because a hundred recommendations is a dashboard with extra steps. The layer ranks everything it finds and ships only what clears the expected-value and confidence bars. Scarcity is the feature — it's what makes #1 mean something.
Yes — every score shows its drivers ("days since last order," "category shift"), and Model Studio exposes each model's accuracy, thresholds, and training history. If you can't interrogate a prediction, you shouldn't act on it — that goes for ours too.
The models want roughly 500 customers and 90 days of orders for reliable scores. With less, LiftSignals still runs — but marks scores provisional and keeps them out of high-stakes insights until confidence clears the bar. You'll never get a bold recommendation built on thin evidence.
Start your free trial — the full layer trains on your store overnight and your first ranked list lands the next morning. 30 days free, no credit card required.