#OPERATING MODEL #TEAM STRUCTURE #LEADERSHIP

Restructuring a 30-person decision bottleneck into three shipping pods

We'd grown from a 10-person tiger team to a real business unit, but kept a culture where every call had 30-50 people in it. Nothing could get decided. Here's how we restructured into three pods after Alpha — and what I'd do differently next time.

COMPANY
Compliance & Risks

ROLE
Senior Manager, Product Experience & Market Insights

TIMELINE
2024–2026

THE SITUATION

30-50 people on every call, and nothing could get decided

I started out leading a ~10-person tiger team alongside a delivery PM. When the product got traction and became a real business unit, a GTM PM and the rest of engineering joined — and we kept the same culture of radical inclusion that had worked at tiger-team size. 30-50 people on every call. Sales, CS, regulatory experts — everyone had a seat and an opinion.

The result: we couldn't make a single decision without consulting the room. I and two other PMs were working 12-hour days, seven days a week. We hit Alpha — it worked — but the team was burned out and the product was fragmented.

WHAT WE DID

Alpha was the forcing function. It proved we could ship. It also proved the structure wouldn't scale, and we had the momentum and mandate to redesign before bad habits locked in.

Testing the obvious models first

We looked at organising by customer journey stage, then by job steps. Both looked clean on a whiteboard. In practice the overlaps were huge and the dependencies unsolvable.

Reframing around what we needed to ship

I pushed for moving away from the academically correct model toward what we actually needed to ship in the next three months. Working with the CPO and the other two PMs, we landed on three time-boxed problem spaces — each with a clear 3-month goal, real independence, and a tight cross-pod sync limited to just the three PMs and the CPO.

The three pods: Pod 1 (mine) owned end-to-end UX, time-to-value, and cohesion across the whole product. Pod 2 owned the engineering data pipeline for V1. Pod 3 owned task management, a longer-runway feature.

For Pod 1, the goals were specific: reduce time from signup to the moment a user gets a meaningful result from the AI, increase onboarding completion without drop-off, and raise the rate at which users accepted AI outputs without modification — the core trust metric underneath everything else.

"Each one is a decision rule, not an aspiration.."

Systemising research instead of gatekeeping it

Centralising design had also meant centralising who had access to customer insight. I changed that: each pod got direct access to its own customer research rather than insight flowing through one person. That meant pods could apply the principles against their own evidence when making fast calls, instead of waiting on a central decision-maker who'd become a bottleneck by design.

Measuring quality instead of approving it

Rather than holding quality at the review gate, I put measurement on the output itself — tracking customer effort to complete a job step, alongside output quality. That gave pods a live signal on whether their decisions were working, without needing my sign-off to find out.

IMPACT

Three pods shipping independently, faster, with no central design checkpoint

  • Three pods operating independently with AI-assisted prototyping at higher volume and speed than before, without a central design checkpoint

  • A shared set of behavioural principles in active use for day-to-day decisions across pods, not published and ignored

  • Customer insight decentralised to each pod, removing a structural bottleneck I had previously been the centre of

  • Quality measured continuously (customer effort, output quality) rather than gated at review

WHAT THIS TAUGHT ME

Quality has to be built into principles and systems, not enforced after the fact

Design reviews don't scale when teams move quickly — by the time something reaches review, the decision has already been made ten times over by the AI-assisted prototyping happening upstream.

And principles only work as decision tools if they're behavioural and evidence-based. The job wasn't to write principles — it was to change how decisions got made when I wasn't in the room.