AI consulting services for mid-sized companies

Buying licences alone rarely changes that. A process that runs differently does, and that is what we deliver:

The first two stages end with a written decision: which use case to start with, what it needs from your systems and data, how it will be measured, who owns it, and what we found that argues against it. Every engagement then moves through the same five stages, and you decide after each one whether to continue:

• Readiness and use cases: checked on your own processes, data and sample cases
• Roadmap: short, with the first item built and measured before the rest is planned
• Pilot: on real work, against acceptance criteria agreed before the build
• Production: inside your systems, on your own servers or in your own cloud account
• Operation: a named owner, re-tests on every model change and a reported cost per case

From you we need a short list of candidate processes, real sample cases for each, the data involved as exports or through read access you approve, and time with the people who do the work today and with whoever runs the systems.

You receive a written ranking of the candidates with what each would need, a measured baseline for the strongest one, a test set built from your cases and a recommendation to build, wait or drop. The test set and the written decision stay with you, whether we build the system afterwards, your own team does or another supplier.

Building a system is not part of the assessment; it is a separate step with its own price. The assessment is also not a security certification or a compliance audit. Each candidate is checked on these points:

• Data: where the inputs live, their quality and who may grant access
• Systems: APIs, permissions and how a model's output reaches the place it is needed
• Infrastructure: whether the data needs your own servers or your own cloud account, or may go to an approved provider
• People: who reviews results, who owns the system after launch and what changes in their work
• Rules: data protection, the works council where there is one and your own policies, prepared for your advisers to decide
• Baseline: what the task costs today and how often it goes wrong, so a change can be measured

• Use cases ranked by value, risk and effort, measured on your cases where possible
• For each: the data, systems and permissions it needs, and where it runs
• For each: the owner, the measure of success and the criteria for stopping
• What we found that argues against each item

The pilot then runs on real work, with a person approving each result. At the end the figures decide: go into production, change the scope or stop. A stopped pilot with clear figures is a result too, and costs less than a system nobody uses.

The acceptance test runs once more on the production setup before go-live. Our pages on on-premise AI, data integration for AI and AI system testing describe the technical side in detail.

• A named owner on your side for each system, with budget and decision rights
• The acceptance test run again after every model, prompt or data change, before the change goes live
• Cost per handled case and spend per team reported, so the business case stays visible
• Runbooks for your team, or operation by Cloudsail if you prefer
• Logging and reporting rules agreed with your works council

AI transformation, as we use the term, means the first process runs on your systems with a measured result, and the next one is chosen on the same figures. Each person involved gets what they need to decide:

• Managing director: a decision paper with value, cost per case, risks and stop criteria
• Finance: the baseline, the measured change and the running cost
• IT: architecture, data flow, permissions, operation and an exit plan
• Data protection officer and works council: what is stored, where data is processed and what is logged, for them to decide

• AI strategy: which processes and products to start with, ranked by measured cost, risk and value
• Use case assessment: whether a task suits a rule, a workflow, a model or an agent
• AI readiness: data access, integration, security and the people who will own the result
• Model and vendor choice: products compared on your own tasks, with no commission from any vendor
• Implementation planning: architecture, test sets, approvals and how the system is run after launch
• Coding agents for engineering teams, where we also advise: rollout, harness, cost control and measurement

Extraction reaches the accuracy the team needs on typed documents but fails on handwritten forms, so those stay with people. The drafted replies need so much editing that they save little time. The recommendation is to build extraction with a review step and to drop the replies for now, with the figures behind both decisions.

Strategy consultancies are strong on business cases and change programmes, and some companies need exactly that. AI agencies often configure or resell a particular tool. We are an engineering company: the people advising you write the code, run the tests in your environment and hand over what they built. No model or tool vendor pays us, so a recommendation is not tied to a product.

If what you need is a company-wide change programme or a procurement process across many vendors, we will tell you so in the first call.

Two things are free: the 30-minute call with an engineer and the free AI usage review.