How it works
A return-behavior signal, built from your store's own data
The short version
- 1
Connect your store
Flokte reads order and return history from Shopify — order dates and amounts, return dates, return reason codes, refund amounts, and whether each return was attributed to store fault (wrong item, defect) or customer fault (changed mind, out of policy).
- 2
Classify return behavior
Over a 180-day window, Flokte weighs four things: how often orders come back, how much of that is customer-fault rather than store-fault, recent return activity, and how much order value is refunded. A clean store-fault history counts in the customer's favor. The output is Good, Watch, or Flagged.
- 3
Explain every classification
Each result comes with the specific factors that produced it — e.g. "4 returns attributed to customer fault in 60 days" — not just a number. Those factors are what a shopper sees if they ask why they were flagged, and the model version is recorded so past results stay attributable.
- 4
You choose what happens
You configure the actions: a tier tag on new orders and discount codes for good-faith customers today, and — when they're enabled — added friction (manual review, restricted returns) for high-risk ones. Any restriction always carries a customer notice, a plain reason, and a route to human review.
What the score is — and isn't
It is
- • A return-behavior signal for one merchant's store
- • Built from that store's order and return records only
- • Explainable factor by factor
- • Versioned, so past results stay attributable
It isn't
- • A credit score or financial-eligibility rating
- • A cross-merchant blocklist sold to third parties
- • Built from social media or off-platform data
- • A final decision made without human review
Fairness testing
Before general availability, and before any material change to the scoring logic afterward, Flokte measures flag rates across geographic slices (using shipping region as a proxy, since no demographic data is collected) and checks that no slice is flagged disproportionately. The results are retained as a durable record. Model activation is blocked in the system itself until a fairness test passes.
Human review
No customer is ever permanently restricted by an unreviewed automated output. When restriction rules are enabled, every restriction carries a customer notice, a plain reason, and a route to human review. Any shopper can contest one through our dispute path. Dispute a decision.