01
Readiness Assessment
Where AI pays, where it will not
We map the intake-to-learn loop and score every stage on evidence, not opinion.
- Duration
- 2 weeks
- You leave with
- A ranked map
Build the product. Fix the pipeline.Prove the number.
We build AI products for teams who need one shipped, and rebuild delivery for teams whose AI tooling has not moved a metric yet. Either way, you leave with evidence rather than an opinion.
Adopting AI tools is not the same as adopting AI.
Four things we see in almost every organisation that bought tooling first and asked what changed afterwards.
| What was done | What was measured | What actually changed |
|---|---|---|
| Rolled out AI coding assistants | Seats activated | Code arrives faster at a review queue that did not get faster |
| Ran an internal hackathon | Ideas generated | Three prototypes, no owner, none in production |
| Built agentic workflows | Agents shipped | Quietly switched off once the person who built them moved team |
| Published an AI policy | Policy approved | Nobody's decision on Monday morning was different |
Writing code was never the slow part.
Work moves through five stages. Speed in one stage is invisible if the next one is where the time actually goes.
01
Intake
Everything is a priority, so nothing is
02
Decide
The decision waits for a meeting
03
Build
The stage AI actually sped up
04
Release
Batched, manual, and feared
05
Learn
No number, so no learning
Prove it. Scale it. Run it.
02
Low-Hanging Fruit
Ship the cheap wins first
Two to four changes shipped inside the first sprint, each with its metric attached.
- Duration
- 4 weeks
- You leave with
- Shipped changes
03
Pilot & Feedback
One pilot, one owner, one number
One pilot in production with a named owner and a feedback loop that runs weekly.
- Duration
- 6 weeks
- You leave with
- A live loop
04
Repeat & Scale
The loop runs without us
The method becomes your team's, and our involvement shrinks on purpose.
- Duration
- Ongoing, quarterly
- You leave with
- Your team running it
What changed, and by how much.
3.4×
Deployment frequency
−61%
Merge to production
11 wks
Assessment to scaled pilot
4 of 5
Pilots reaching production
8×
European logistics group
Release cadence moved from monthly to twice weekly after intake and release were rebuilt together.
94%
Property operations, 6,000 units
After-hours intake given one owner and one weekly number; four queues scaled by quarter end.
−31%
Insurance claims platform
Two low-hanging changes in the first sprint removed a manual review step entirely.
Evidence first. Then scale. Never the reverse.
01
Outcome before output
Every engagement starts by naming the number it has to move and how we will know it moved. If we cannot name it, we do not take the work.
02
Evidence before scale
Nothing gets rolled out until a pilot has produced data a sceptic in your own organisation accepts. The sceptic is invited to the review.
03
One system, not one team
We change intake, decision-making and release together. Speed up a single stage in isolation and the queue just moves somewhere less visible.
04
The loop is the product
We design the feedback cadence first and pick the tooling to serve it. A weekly number on a wall has outperformed every platform we have seen bought to replace it.
“The first thing they did was tell us to stop two of our four AI projects. That conversation was worth the engagement on its own.”
“Two weeks in we were reviewing pull requests instead of slide decks. By the third month our own people were running the reviews.”
Half a day. No charge. A ranked list at the end.
- Two of our engineers, on site or on a call.
- Your delivery leads in the room, not just procurement.
- A ranked list of where AI pays, and where it will not.
- No deck, no obligation, no NDA needed to talk approach.