Back to work

Keeyu

AI agent for post-purchase ecommerce operations.

Timeline
2026
Role
Product Designer
Deliverables
End-to-end product, design system
01context

Keeyu runs the back end of ecommerce: payments, warehouses, carriers, and the support tickets they generate. I'm the product designer on a rebuild that turns it from a dashboard you check into an agent that tells you what's about to break.

Ops teams watch several tools and find out about problems when the customer tells them. Most of those are preventable. The data that would have caught them is already in the system, just not shown in time.

02insights

Three things I found looking at post-purchase support.

Insight 01

"Most of the tickets are the same question"

Where is my order is the single biggest category of ecommerce support tickets, about a fifth on an average day and over half in peak season1. Almost all of it is preventable. The tracking data that answers the question is already in the system before the customer thinks to ask.

Insight 02

"The customers who leave quietly cost the most"

Watch only tickets and reviews and you miss most of the damage. One bad delivery is enough to stop 55% of shoppers buying from a brand again, and close to a third go on to warn other people off2. Almost none of them file a ticket first. The only way to reach them is to catch the delivery before they do.

Insight 03

"Proactive help works, and almost nobody does it"

Gartner found proactive service lifted every satisfaction score they track, NPS and CSAT included, by a full point, yet only 13% of customers had ever received it3. Aim is the whole problem: untargeted alerts generate more follow-up than they prevent.

03the problem

Catch problems before the customer feels them, across payments, fulfillment, and shipping, without adding another dashboard to babysit.

That puts two decisions in the design: what's worth interrupting someone for, and how much the agent has to show before a person will act on it. Get the first wrong and it's noise. Get the second wrong and nobody trusts it.

04outcomes

What shipped, what changed.

  • Helped reframe the product around catching problems early across payments, warehouses, and carriers, rather than reacting once a customer reports one.
  • Designed the patterns for how the agent recommends and how a person approves, overrides, or hands it back, so the AI behaves the same way on every surface.
  • Advised on the wider UI uplift: tightening components and type, and pulling the surfaces back in line with the brand as the product grew.

A note

Keeyu is still in build. Specific product detail and metrics are under NDA.

References

  1. 1.Gorgias, ecommerce support ticket data (WISMO / order-status volume)
  2. 2.Bringg, Delivery Experience research (impact of a poor delivery on repurchase)
  3. 3.Gartner, Customer Service and Support, 2020 (proactive service impact and reach)

Next case study

Trusst AIA product design system rebuilt on a new brand.