PERKLY · PERSONALIZATION

One home feed, every customer

A speculative product exercise — Perkly is a fictional deals marketplace.

The situation

Millions of people a month browse hundreds of thousands of local deals — services, things to do, travel and goods — across three customer segments, and yet they all get the same single experience.

The goal

Grow margin, and get more customers to buy again within 30 days — with a team of three engineers, half a data scientist and half a designer.

Decide what each customer sees, get smarter every time they return, and be able to show that it makes money.

What this prototype shows

  • The Explore home feed for different customers, and why each block is there.
  • The same customer returning three times — what the system learns each visit and how the page changes.
  • The logic that turns what it knows into a page: propensity × expected margin.
  • What a brand-new customer sees when it knows nothing yet.

Use it

Click around the phone in the middle — open and save deals, search — then press “Simulate interaction & reload app” in the panel on the right to start the next visit. That panel shows the customer profile, the event log and the ranking trace updating live.

All deals, merchants, margins and model coefficients are synthetic demo data.

9:41 Visit 1
29
Praha, TX
personalization-engine

1 · Customer profile

2 · Segment detection

How this estimate is made

3 · Ranking calibration (live)

4 · Behavior → updated profile

Simulate a visit · optional custom plan of customer clicks
    Event log (what each click changed)

      5 · Why this page

        Top tiles: why each
        Deal scores (P × margin)

        Prototype for a product case study — Perkly is a fictional deals marketplace. All deals, merchants, margins, promo campaigns, segment priors and model coefficients are synthetic demo data. The model is a hand-set, explainable logistic score; in production the coefficients would be learned from real behaviour.