Case 02 · pawaTech · operating model · Under NDA

Speed, without guessing

Whoever has an idea builds it as a working prototype on the real design system, and tests it with real customers — before anyone commits to building it.

Role
Head of Product Design & UX · operating model
When
Apr 2025 — present
Where
betPawa · design, product & engineering

Challenge

Ideas moved at the speed of documents, and the first honest signal arrived after engineering had already paid for it.

Approach

Give the people with the ideas a way to build a real prototype in hours, and put every prototype in front of customers before engineering starts.

Result

Ideas are built, tested and released to a small live audience before they reach everyone. The answer comes from customers rather than from the room.

2places AI was put to workbuilding the prototype, and testing it
0PRDs needed to share an ideaa working prototype does it
17markets the loop runs in

The problem

What was actually wrong

Most of the time between an idea and a release is not spent building. It is spent deciding what to build, and then finding out late that it was the wrong thing.

Here that happened in documents, and customers only saw the idea once engineering had committed to it — which makes every finding expensive and every disagreement political.

Before

A BRIEFA LONG-FORM PRDA SLACK THREADBUILD ITFIND OUT, LATE
A line. The answer arrives after the money has been spent.

After

PROTOTYPE ON THE SYSTEMTEST WITH CUSTOMERSITERATELIVE SLICE
A loop. Each turn costs hours, and something outside the team answers.

My role

Who did what

Mine

  • Decided where AI belonged: two points, not everywhere.
  • Made the design system the base for prototyping, so what people build behaves like the product.
  • Set up the loop — test with customers, iterate, then release to a small live audience.
  • Kept judgement human. AI shortens the distance; it does not decide.

The team’s

  • Product managers build their own prototypes and own the ideas they take to customers.
  • Research defined what a valid test looks like, so speed did not quietly become sloppiness.
  • Engineering owns the release mechanics that make a small live rollout safe and reversible.

Process

How it went

  1. 01

    Prototype on the real system

    Product people assemble something from the real components, so it behaves like the product instead of describing it — in hours, not in a sprint. Being wrong costs nothing at that price.

  2. 02

    Ask customers, not the room

    Usability testing with real customers. That is the whole reason the prototype has to be built on the real system.

  3. 03

    Iterate while it is still cheap

    Another round costs hours. The second and third versions are the ones customers actually asked for.

  4. 04

    Release to a slice first

    A prototype tells you whether people understand something. Only production tells you whether they do it.

Before

EVERYONE AT ONCE
One decision, taken before anyone has behaved.

After

A SLICEEVERYONEPRODUCTION DECIDES
Real money, real market, before the idea reaches the whole audience.

Decisions

What was chosen, and what it cost

Two points, not everywhere

Over
Rolling AI out across the whole process at once
Because
Speed is governed by how fast an idea becomes testable and how fast the answer comes back. The rest can wait.
It cost
Saying no to a long list of reasonable-sounding requests.

Prototypes built by whoever had the idea

Over
Prototypes built by designers on request
Because
A prototype you made yourself is one you will happily throw away. A queue turns the author into a reviewer of their own thinking.
It cost
The design system has to be genuinely good, because it is now used by people who cannot work around its gaps.

A live slice before the full audience

Over
Deciding on the strength of the usability test alone
Because
Only real money in a real market shows what people actually do, rather than what they understand.
It cost
Release mechanics, monitoring, and a real willingness to pull something back.

Outcome

What came of it

The loop closes inside the product instead of in a meeting: build, test with customers, iterate, release to a slice.

Product stopped queuing for design time to explore an idea, and design stopped making artefacts for ideas that were never going to survive a customer.

Direction and accountability stayed human. AI made the distance shorter, not the decisions.

Lesson

What it cost me to learn

The failure mode is looping instead of learning, and it feels productive the whole time it is happening. A loop only pays when something outside the team answers at the end of it.