Strain intelligence built from what your customers bought.

A customer buys. They come back and buy again. That pair — what followed what — is the signal, and it is a purchase, not an opinion. The cohort is anyone with at least 3 cannabis purchases in 90 days, so a single visit never moves it.

Those consecutive pairs get tested against a 250-strain terpene reference — did buyers stay within a dominant terpene, or cross it? The answer is a lift number, and the lift decides what the shop is told:

• ≥ 1.5×strong predictor — terpene is doing real work here — worth building recommendations on.
• 1.1–1.5×weak signal — hold. Something is there, but not enough to steer a buyer on.
• < 1.1×decoration — terpene is not predicting repeat purchase in your shop. The system says so and reroutes to plain strain-level co-purchase.

That bottom row is the part nobody else ships. Most recommendation features are built so they can never be wrong out loud. This one publishes the case where its own signal is too weak to use, and sends the shop back to plain strain-level co-purchase instead. A number you can act on has to be a number that can come back empty.

Your customers do leave reviews — the post-purchase review surface is live in the customer app, and those reviews show on your strain pages. They are a customer-facing surface, not the input to this. We are not going to tell you an engine is reading them when it is not.

What the budtender sees at the cart is comparative and behavioural — what repeat buyers of this strain bought next, on your shelf. Never an effect claim, never therapeutic language. WAC 314-55-155 stays clean because the shop is describing purchases, not outcomes.

Your customers know this. Two pre-rolls labeled “Blue Dream” from two different cultivators don’t smoke the same. A national catalog can’t tell you which version your shop’s repeat buyers actually came back for.

Repeat-purchase data answers that, because a second purchase is a decision the customer made after they got home — and made without being asked. A buyer who came back for the same cultivator’s Blue Dream told you something. A buyer who switched told you something else. Neither had to fill anything in.

It also works on day one of a strain you have never carried, which is where review-based approaches are weakest — there are no reviews yet, but there are buyers who bought it and then bought something next.

No efficacy claims. No medical recommendations. No effect language at all. This is what was purchased, in what order, in your shop.