Home/Services/AI Transformation
Flagship engagement
AI Transformation Consulting for Companies That Need a Number, Not a Demo
AI transformation is the work of redesigning workflows, products and operating models around AI systems while measuring what that does to cost, revenue and decision quality. Done properly it starts by ranking candidate use cases on value at stake, data readiness and cost to serve, then funds only the few that beat an existing manual baseline.
In our experience most AI programmes stall between a convincing pilot and a line on the P&L. The reason is rarely the model. It is that nobody attached a number to the idea before building it.
- Senior-led engagements
- Written recommendation first
- Reply within two working days
When companies call us
The conversation that keeps restarting
Portfolio
“We have thirty AI ideas and no way to choose between them.”
Adoption
“Our pilots demo well and never reach production.”
Pricing
“The board is asking what our AI spend has returned.”
Sequence
“Security and legal keep stopping projects at week ten.”
Economics
“We cannot tell whether this is cheaper than doing it by hand.”
And usually
“We have been going round this for months and keep starting the argument again.”
The engagement
What we look at, and what you keep
What we analyse
- Every candidate use case, scored on value at stake, data readiness and cost to serve
- Whether the data actually exists today, in a system you control, with enough history to test
- Security, legal, client consent and audit-trail constraints, in week one rather than week ten
- The real current cost of the task, including the people who do it now
- Unit cost per task under a model, including human review and exception handling
- The accuracy level at which net saving peaks, which is rarely the highest accuracy
What you receive
- A ranked AI portfolio with a funding decision on every use case
- Two or three cases live in production, measured against a stated baseline
- Unit economics per task, and the routing rules that hold them
- An evaluation suite your team can keep running without us
- A short list of unglamorous data fixes, each with the use case it unlocks
- Governance and reporting your board and auditors will accept without translation
Worked example
Forty AI ideas. Three worth funding.
An industrial services company put forty AI ideas through the three gates. Twenty-three failed on data, nine on permission, five lost to something already in place.
Read the worked exampleComposite company · illustrative · no client data
How long it takes
Two-week diagnostic on a single use case, then 8–20 weeks to production.
Who it is for
Companies with several AI ideas, a senior sponsor, and an engineering team that can be involved early. We are a poor fit if you need hands to deliver a plan that is already fixed.
We do not resell models or seats, and we will not run a pilot that cannot state in advance what it would save or earn.
Related services
Common questions
AI Transformation, in practice
How do you prioritise AI use cases?
We put every candidate past three questions in order. Is the data here today, in a system you control, with enough history to test against? Are we allowed, once security, legal and client consent are considered? Does it beat the real current cost of doing the task by hand? Most ideas fail the first question, which takes the least time to answer. In our published worked example, 23 of 40 ideas failed on data alone.
How do you measure AI ROI?
Per task, against a measured manual baseline. We cost the current way of doing the work including the people, then cost the model version including inference, human review, exception handling and QA sampling. The difference multiplied by volume is the saving. Build cost is separate and produces a payback period, not a vague efficiency claim.
Do you build the product or only advise?
Both, alongside your engineers. We work on evaluation harnesses, data and retrieval plumbing, and the route from prototype to production that survives security, legal and procurement review. We do not take the code away with us.
Which AI models do you work with?
Whichever the economics and the constraints point to. Model choice usually matters less than people expect. On the largest use case in our worked example, inference was 35% of the running cost and human review was 57%, so the design question was how to reduce review, not which model was cheapest.
Do you work with our existing engineering team?
Yes, and we prefer to. Engagements where engineering is involved from week one reach production. Engagements where design is handed over at the end usually do not.
What does the diagnostic include?
Two weeks. We inventory your candidate use cases, test each against the three gates, cost the survivors against their manual baseline, and give you a written recommendation with a funding decision on each. The document is yours whether or not you continue with us.
How long before we see anything in production?
Typically 8 to 20 weeks after the diagnostic, depending on how many integrations and reviews stand between the prototype and live use. Security review is more often the constraint than engineering.
Tell us the decision you are facing.
A short paragraph is enough. Where another firm would serve you better, we will say so and suggest one.
- Senior-led engagements
- No model or software resale
- Reply within two working days