Customer signals — everything the models know, in plain language
Every customer, scored on every signal, every night — who's leaving, who's buying, what they're worth, and what to do about it. Any answer becomes a campaign audience in one click, and every campaign comes back with proof.
No jargon on the surface — each signal is a question a merchant would ask, with the model name underneath for the curious. The core signals ship on Starter; the predictive set — propensity, next purchase, market basket — arrives with Growth.
A gradient-boosted ensemble scores every customer 0–100 — with the drivers visible, like "days since last order." High churn × high value is your revenue at risk, found nightly.
Classic RFM segments — Champions, Loyal, At-risk — recomputed continuously. The overnight downgrade ("fell out of Champions last night") is one of the highest-signal alerts in the product.
Scores everyone who's gone quiet on the odds they can actually be recovered — so win-back budget goes to the recoverable, not sprayed across everyone who left.
Return propensity, 0–100. Trim return-heavy segments out of promotions and see the true-margin impact of return behavior — because revenue that comes back isn't revenue.
The cheap wins. Concentrate spend on people about to buy — or exclude them and stop paying for orders that were coming anyway. Both are one click.
Sequence analysis predicts each customer's most likely next product, with ranked alternatives. Win-backs stop being generic: "come back for the thing you're due for."
Association rules mined from your orders power bundles and cross-sell — the "customers who bought this also bought" you actually own.
Full-price buyers, value seekers, bargain hunters — classified from real order history. Stop sending 20%-off codes to people who never needed them, and catch Champions slipping into discount dependence.
The decision layer. Every score fuses into one recommended action per customer with a confidence level — and "don't contact" is a real verdict, not a missing one. More on that below.
Probabilistic lifetime value with VIP → low tiers and a dollar figure. It travels with every export, so Google and Meta bid on customer value, not just clicks.
Time-series revenue forecasting with its quality metrics on display. When actuals drift from forecast, that gap becomes a recommendation — not a surprise at month-end.
Acquisition cost by channel and period, from your synced ad spend joined to real acquisition data — the denominator your ROAS screenshots have been missing.
ZIP-level performance joined with ZIP-level ad spend — plus lookalike ZIPs: untapped territories that behave like your best ones. A prospecting map grounded in your own data.
The media mix model estimates each channel's true contribution and lets you simulate budget shifts before spending a dollar — Meta to Google to TikTok, modeled first, moved second.
Cross any two signals and the sharpest lists appear — high churn × high value, full-price × due-to-buy. Every cell of that matrix is a button.
See how one click works →Every tool you've ever paid wants you to send more — more blasts, more codes, more "touchpoints." LiftSignals fuses every signal into one verdict per customer, and no-contact is a first-class verdict.
It's the tool that tells you to send less when that's what protects your margin and your list. That's what makes the rest of its advice believable.
Every signal ships with its accuracy on display and its drivers open for inspection — the transparency you'd demand from an analyst, from software.
AUC, MAPE, forecast error — visible in Model Studio for every model, updated with every retrain. If a model is having a bad month, you'll see it before we tell you.
See what moves each score — "days since last order" driving churn, cadence decay, discount patterns. Explainable enough to defend in a meeting.
Change your mind about a threshold, re-run a model, re-score the base. Raw data is never touched — recompute forever, safely.
Models train only on your store's first-party purchase history — never pooled across merchants, never sold, isolated at the database level. GDPR-ready, with self-service export and deletion built in.
Connect a store — or a spreadsheet — and see your own signals tonight. Push one audience, and the proof follows three weeks later. Free for 30 days.