Start here
10 working days
The Diagnostic
A ranked map of what to automate, what it is worth, and what to leave alone. Fixed scope, written for the person who has to deliver it.
AI operations & commercial consultancy
Agentic workflows that survive contact with a real operation.
We find the work worth automating, build it into the operation, and leave your team owning it.
- 120+
- people led
- £100M+
- budget owned
- Regulated
- healthcare
- In production
- agentic workflows
The problem
The pilot works. The cost base doesn't move.
Most businesses now have an AI line in the plan and something running in a corner nobody depends on. It never changed the numbers, because it was never wired into the path the work actually takes.
Two groups sell into that gap: strategy firms, who produce a roadmap and leave, and development shops, who produce agents and leave. Neither has carried a rota, defended a headcount plan, or explained to a regulator why an automated decision was made.
Rework
The same case touched four times because nothing caught it early.
Chase
People employed to ask other people for things.
Wait time
Work sitting in a queue while cost accrues against it.
Premium cover
Agency, overtime and escalation paying for the gap.
How we work
One front door, then two ways forward.
The Diagnostic comes first. Everything after it is optional, and the plan is yours either way.
Then
8–16 weeks
The Build
One to three workflows running in the operation, owned by their team. Staged gates, shadow mode before cutover.
Then
Varies
Enable
Your engineers, ops and growth people able to do this without us. Three tracks, run against your own workflows.
What we think
We publish the method, not client logos.
No case studies and no pricing tables. Just what we believe about this work, written out.
Operations
Where pilots die
Most agentic pilots do not fail on model quality. They fail at the seam where the workflow meets an operation that has exceptions, rotas, regulators and staff turnover.
4 August 2026 · 8 min read
Worked problem
The real cost of a dropped offer
Post-offer dropout is treated as a recruitment metric. It is a capacity metric, an overtime metric and, in regulated services, a compliance metric. Here is how the number actually compounds.
21 July 2026 · 11 min read
Method
What we advised against building
A diagnostic that never returns a 'do not build' finding is a sales document. Three examples of work that looked automatable and was not.
2 July 2026 · 7 min read
Split of functions
Detection is automated. Judgement stays with named humans.
Every workflow we build has this line drawn explicitly, in writing, before anything reaches the live path.
Agentic
Detection, chase and routing
- Watching queues and flagging the exception before it ages
- Chasing the missing document, signature or data field
- Routing work to the right person with the context attached
- Reconciling systems that disagree, and reporting the delta
Human
Judgement and accountability
- Any decision a regulator, customer or clinician would question
- Pricing, escalation and exception approval
- Owning the workflow, its gates and its failure cases
- Deciding when the automation is switched off
I run an operation
Cost-to-serve is rising and the headcount plan you need is not affordable.
I own a portfolio
There is an AI line in the value-creation plan and it has not moved in two quarters.
We have run the operation we are being asked to fix. That is the whole proposition.
If you want a roadmap, a proof of concept for the board, or someone to be enthusiastic about AI with you, we are the wrong firm. We build things that go into the live path and stay there.