Andy Ke

TaelorAI for stylists, Senior PM

Styling 45% faster with the AI workflow I launched, and 90% of stylists use it

Taelor's stylists hand-pick every shipment. I defined the requirements, the measures and the AI-assisted workflow for their new styling tool, then the guardrails that decide what reaches a customer. The AI drafts; the stylist decides.

Speed

45% faster

Styling with the AI pairing workflow (category intelligence, ranking criteria, fallbacks), adopted by 90% of stylists.

Quality

+60%

Recommendation success rate (recommendations stylists approved) in 2026, with zero customer-reported sizing errors, after I set quality thresholds and fallbacks.

Trust

Human check

Customer summaries show their sources and wait for a stylist's confirmation, so urgent requests and negative feedback get handled better.

How I built it. I used ChatGPT for prototyping, ideation and analysis; I owned the requirements, pairing rules, guardrails and measures. Engineering built the production tool.

Redrawn PairsWith screen: a picked item, ranked AI matches and thumbs ratings
PairsWith, redrawn from the shipped tool. Names and details are fictional.
  1. The stylist picks an item.
  2. The AI suggests matches, ranked by rules built from outfits customers kept.
  3. Stylist ratings shaped each rules release.

The decision

Chose

Pairing rules built with the stylists from outfits customers kept, before showing the tool more often.

Over

Pushing usage with more placements.

Because

A recommender nobody trusts doesn't get better by being seen more.

Deferred

Persona-based recommendations and whole-shipment drafts, held for design capacity.

Team

Engineering, design, data science, merchandising, operations and inventory.

I owned the stylist tool and AI roadmap, the measures, the pairing rules, the specs and CEO buy-in.

Redrawn AI Customer Brief
The AI Customer Brief, redrawn from the shipped tool. Names and details are fictional.
  1. One card instead of a stack of tabs: the AI reads texts, reviews and notes.
  2. It catches clashes: picked items with a print the customer dislikes.
  3. Each request becomes a button that filters or reranks the items.

When the AI is wrong

  • Below the quality bar, the system falls back to safer results instead of guessing.
  • Nothing ships to a customer without the stylist's approval.
  • Every summary line links back to the message it came from, so stylists can check it.
Redrawn AI text summary beside a raw text thread
AI text summary, redrawn.
  1. The raw text thread a stylist used to scroll through.
  2. The AI pins feedback and requests, each linked back to its message.
Redrawn AI styling-notes drafter
AI styling notes, redrawn.
  1. The stylist says what the note should do.
  2. The AI drafts it from the outfits actually in the box.
  3. It explains why the draft works, so the stylist learns, not just edits.