COMPOUNDING | T
Case Studies
v.2026.AI

Anonymous, but
specific.

Confidentiality is part of the contract. Below: three real engagements, anonymised, with the actual shape and outcomes.

Editorial illustration of a 1990s case-study archive — folder, polaroids and beige keyboard
FILE: case_studies.txt

Selected engagements

01—03
CS.01
Fintech · $500M ARR

Customer-ops team runs a weekly AI review, unprompted

Context

A scaled fintech with a large customer-ops function. AI licences had been distributed for months; usage was patchy and inconsistent across teams.

Challenge

The behaviour we needed to change wasn't tool usage — it was the team's default posture when a new workflow question came up. Most reached for the old runbook.

Approach

A 6-week embedded programme: weekly working sessions with team leads, a shared internal prompt library co-authored by the team, three production workflows shipped per pod, and a named AI owner per pod running a Friday review.

Outcomes

  • →Weekly AI review now runs without us in the room
  • →Named internal AI owner in every pod
  • →Team-authored prompt library, still edited quarterly
  • →Tier-1 ticket response habits changed, not just times

"We stopped treating AI like a magic button and started treating it like a tool with a manual. That made all the difference."

— VP of Operations, Fintech · $500M ARR
CS.02
VC-backed B2B SaaS

Non-engineers now prototype AI features before roadmap review

Context

A 120-person SaaS company had an over-long backlog of 'AI feature ideas' and no clarity on which were worth building.

Challenge

The cultural block wasn't ideas — it was who was allowed to try them. Product debates ran for months because only engineers could prototype.

Approach

A Sprint followed by an 8-week cohort programme. Cross-functional pods, live coaching, working prototypes as the deliverable. Product now runs 'prototype first, debate second' by default.

Outcomes

  • →Non-engineers now prototype ideas before roadmap review
  • →2 ideas promoted into the live roadmap; 3 dead ideas retired with evidence
  • →Cross-functional prototyping cadence runs quarterly
  • →'Prototype first, debate second' is now the default posture

"Our engineers stopped being the bottleneck on AI ideas. Everyone could prototype."

— Chief Product Officer, VC-backed B2B SaaS
CS.03
Asset manager · multi-billion AUM

Shadow AI disclosed, sanctioned, and shrunk — without a ban

Context

An investment team had quietly started using consumer AI for research summarisation. Compliance found out. The default reaction would have been a ban.

Challenge

Change the org's reflex from 'ban and forget' to 'sanction and monitor'. Keep the productivity gain. Don't push usage further underground.

Approach

Joint design with compliance, IT and the investment team. Plain-English policy, risk taxonomy, a sanctioned tool stack, a lightweight review rhythm, and a named AI owner per desk.

Outcomes

  • →Shadow-AI use disclosed and mostly rerouted to sanctioned tools
  • →Named AI owner on every desk, meeting monthly
  • →Compliance now co-authors policy updates with the team
  • →First audit cycle: zero surprise findings

"It was the first compliance document my team didn't groan about. We helped write it."

— Head of Compliance, Asset manager · multi-billion AUM

Want a similar engagement?

If your context resembles one of these, the work resembles the work. Let's scope it.

Start a conversationSee what we deliver