Services
AI consulting services
We provide five kinds of work: strategy, implementation, integration, evaluation and governance. Most engagements start with a short diagnostic, because the decision about what to build is worth more than the build. We are vendor-neutral by construction, and we will recommend against building when that is the honest answer.
The business problem
Almost every organisation now has AI activity. Far fewer can point to a number in a management account that moved because of it. The gap is not model quality, which by 2026 is rarely the binding constraint. It is that the decision about which workflow to change, what would count as success, and who owns the result is usually never made explicitly.
MIT's The GenAI Divide: State of AI in Business 2025 reported that around 95% of generative AI pilots deliver no measurable impact on the profit and loss statement. The same research found that builds involving an external partner succeeded roughly twice as often as internal-only efforts, largely because an outside team has to be specific before it is allowed to start.
That specificity is the service. Everything else follows from it.
When this is useful
Consulting is not always the right purchase. These are the situations where it repays.
Many candidates, no ranking
More use cases than capacity, and no agreed basis for choosing between them.
A pilot that will not move
Something demonstrates well but has not reached production, and nobody can say exactly what is missing.
A build with no measure
Work in flight with no baseline, no threshold, and no named owner for the outcome.
A system already live
Something in production that is drifting, unevaluated, or dependent on one provider in ways nobody has mapped.
A governance requirement
An obligation under the EU AI Act or an internal risk framework, and no evidence trail to satisfy it.
A build-or-buy decision
A vendor proposal on the table and no independent view of what it would cost to own instead.
The five kinds of work
Each of these is a separate engagement with its own page. Most clients do two or three of them, in this order.
- AI strategy consulting — use case selection, impact estimates, build versus buy, model and vendor choice, readiness, sequencing and roadmap.
- AI readiness assessment — a fixed-scope entry engagement that produces a ranked shortlist, one recommended first build, and a not-now list.
- AI implementation services — designing the workflow, building the production system, connecting it, testing reliability and measuring the outcome.
- AI integration services — connecting AI systems to CRM, ERP, support platforms, document repositories, databases and analytics.
- AI governance and evaluation — accuracy, reliability, hallucination testing, security, privacy, bias, oversight, auditability and monitoring.
- Two applied specialisms sit across them: generative AI consulting and AI agent development.
What you receive
Written artefacts rather than presentations, and all of them yours to keep.
- A named workflow with its current cost or cycle time, drawn from data you already hold.
- The number that should move, the threshold that would justify scaling, and the condition under which the work stops.
- An evaluation set of real cases with agreed correct answers, which outlives every model you use.
- Working code, prompts and configuration in your repository, under your licence.
- An architecture that treats the model as a replaceable component, and the documentation to change it.
- An explicit list of what we could not establish.
Expected business outcomes
What these engagements are designed to produce. Which of them applies depends on the workflow, and we will say so before starting rather than after.
A decision
One workflow chosen on evidence, or a documented decision not to build. Both are useful; only one is usually offered.
Measured change
Movement in a number the business already reports, against a baseline agreed in advance.
Capacity released
Volume absorbed without headcount, where the volume exists and the capacity is redeployed.
Fewer wrong decisions
A reduced error rate in a process where errors have a known cost.
An auditable system
Evidence of what the system did and why, sufficient for an internal or regulatory review.
Retained optionality
The ability to change model or provider in a fortnight rather than a quarter.
How an engagement runs
Short, sequential, and stoppable at each boundary.
Frequently asked questions
What does an engagement cost?
It depends on scope, and we will give a fixed price for the diagnostic before any commitment to build. We do not price a build before the diagnostic, because nobody can price a result they have not looked at.
How long before we see anything?
A fortnight for the diagnostic findings. Production timelines depend on the workflow, and the diagnostic is what makes that estimate honest.
Do you resell platforms or take vendor margin?
No. We hold no reseller agreements, which is why we are able to recommend buying, or not building at all.
Who owns the work?
You do: code, prompts, evaluation sets and data. Portability is a design constraint, not an add-on.
Will you recommend against building?
Regularly. It is the cheapest finding a diagnostic can produce, and we consider it part of the deliverable rather than a failure of it.
Do you work with our existing vendors?
Yes, including reviewing a proposal already on the table. We have no incentive either way.
Which regions do you work in?
India, the United States, and Europe and the UK.
Can you help with an AI system already in production?
Yes. Evaluation, monitoring and governance work on live systems is a large part of what we do.
Relevant insights
Start a conversation.
Tell us the workflow you would change and the number you would want to move. If we are not the right people, we will say so on the first call.