Skip to content

Core Concepts

Predictions

The four prediction types Cadence generates for each customer — replenishment, lapse prevention, cross-sell, and subscription upsell.

Predictions are the core output of Cadence. For each customer in your store, Cadence analyzes their order history and identifies whether they’re in one of four revenue moments — each representing a specific opportunity to drive a purchase.

How predictions work

Cadence looks at each customer’s individual purchase history — what they bought, how often they reorder, and how long they’ve been inactive — to determine which prediction (if any) applies to them right now.

Predictions are per-customer and per-product. Two customers who both buy the same serum may have completely different prediction statuses based on their individual reorder patterns.

Predictions are recalculated daily as part of Cadence’s automatic cycle. A customer who was “due to replenish” yesterday may become “about to lapse” if they still haven’t ordered, or their prediction may clear if they purchase.

The four prediction types

Replenishment

When it fires: A customer is due to reorder a product they’ve purchased before, based on their personal buying rhythm.

Example: Sarah buys your 30-day supply of vitamins every 28 days. Cadence identifies she’s due to reorder around day 26–28 and triggers a replenishment event.

What the customer receives: An email or SMS with the product they need to reorder and a one-tap ConvertCart link pre-filled with that product.

Best for: Consumable products with predictable repurchase cycles — supplements, skincare, coffee, pet food.

Lapse prevention

When it fires: A customer hasn’t reordered within their expected window and is at risk of churning permanently.

Example: Sarah usually reorders every 28 days but it’s now been 45 days. Cadence flags her as about to lapse and triggers a win-back event.

What the customer receives: A re-engagement message — often with an optional discount — encouraging them to come back before they switch to a competitor or forget about the product.

Best for: Recovering customers who’ve gone quiet. Often the highest-ROI prediction type because these customers already know and trust your brand.

Cross-sell

When it fires: A customer is ready to discover a complementary product based on their purchase patterns and your catalog.

Example: Sarah has been buying your face serum for three months. Cadence identifies she’s primed to try your complementary moisturizer and triggers a cross-sell event.

What the customer receives: A message introducing the complementary product with a ConvertCart link pre-filled with the recommended item.

Best for: Expanding wallet share from existing customers. Works best when you have a clear product catalog with natural pairings.

Subscription upsell

When it fires: A repeat buyer who purchases one-time is primed to convert to a subscribe-and-save plan.

Example: Sarah has reordered her vitamins three times on a consistent schedule. Cadence identifies she’s a strong subscription candidate and triggers a subscription upsell event.

What the customer receives: A message presenting the subscribe-and-save option for their regular product, with a ConvertCart link that includes the subscription option.

Requirements: A subscription platform must be connected (Recharge, Skio, Loop, or Shopify Subscriptions). Customers who already subscribe to the product are excluded automatically.

Best for: Brands with subscribe-and-save programs who want to grow subscription revenue from existing one-time buyers.

Prediction priority

When a customer has multiple active predictions, Cadence prioritizes them in this order:

  1. Subscription upsell
  2. Replenishment
  3. Lapse prevention
  4. Cross-sell

This priority applies to Nudges and LiveBlock (which show one prediction per customer). For event delivery, all active predictions are sent to your marketing platform — so a customer can receive messages for multiple prediction types through different flows.

Prediction lifecycle

Customer orders → Cadence learns rhythm → Prediction generated
                                              ↓
                                    Event sent to marketing platform
                                              ↓
                                    Your flow sends message
                                              ↓
                          Customer purchases (or doesn't)
                                              ↓
                          Prediction clears or updates next day

A prediction typically remains active for several days (the exact duration is managed by Cadence). If the customer purchases through ConvertCart, the prediction clears. If they don’t, the prediction may escalate (e.g., replenishment → lapse prevention).

What you can’t configure

Prediction timing, thresholds, and eligibility rules are managed by the Cadence team and optimized per store. As a store owner, you don’t set replenishment windows, lapse thresholds, or cross-sell confidence levels — Cadence handles this based on your store’s data.

You can influence predictions through:

  • Product exclusions — exclude products that shouldn’t trigger predictions (gift cards, samples)
  • Pinned products — prioritize specific products in cross-sell recommendations
  • Connected subscription platform — required for subscription upsell predictions

Where to go next