End-to-end customer intelligence: 2x dormant customers reactivated.
A jewelry marketplace was growing its catalogue and its customer base, but its marketing still ran on broad segments and generic email blasts, and churn only showed up after customers had already gone. Plexe built end-to-end customer intelligence across CRM, retention, cross-sell and search, reactivating twice as many dormant customers and lifting repeat purchase by 28%.
The challenge
- Marketing ran on broad segments and generic email blasts
- Churn surfaced only after customers had already left
- Cross-sell and upsell relied on hand-curated bundles
- Search returned poor results for vague or visual queries
What we built
- ML segmentation and AI agents that propose campaigns automatically
- At-risk customers flagged weeks before churn, with retention plays
- Personalized cross-sell and upsell on every touchpoint
- Natural-language and semantic search across the full catalog
Results
- 2x dormant customers reactivated
- +28% lift in repeat purchase rate
Why they came to us
Jewelry is a considered, occasional purchase, so the business lives or dies on bringing customers back. The marketplace had years of purchase and browsing data, but none of it shaped what customers saw. Campaigns went to broad segments, bundles were curated by hand, and the team only learned a customer had churned once they had stopped buying.
Search made it worse. Shoppers describe jewelry the way they see it: "a thin gold ring with a small green stone", "something like this but cheaper". Keyword search returned poor results for those queries, and customers left without finding what they came for.
Goals and success criteria
We set two headline metrics with the team: reactivate more dormant customers and raise the repeat purchase rate. Supporting those, every customer touchpoint (email, on-site recommendations and search) had to draw on the same customer intelligence, rather than four disconnected tools.
How it works
Plexe replaced broad segments with ML segmentation, and added AI agents that propose campaigns automatically: who to target, with what, and when. A churn model flags at-risk customers weeks before they would have lapsed, and triggers retention plays while there is still time to win them back.
On site, personalized cross-sell and upsell replaced hand-curated bundles on every touchpoint. Natural-language and semantic search now covers the full catalogue, so vague or visual queries return the pieces customers actually mean.
Results and what they own now
Twice as many dormant customers were reactivated, and the repeat purchase rate rose by 28%. Segmentation, churn prediction, recommendations and search all run in the marketplace's own stack, so the marketing team works from one shared view of the customer rather than a set of disconnected tools.