YOUR PEOPLE RUN IT AFTER WE LEAVE
AI Implementation
You own what we build, and your own people run it: taken from the leadership decision into everyday use, built to production criteria, handed over once it holds. We stay until it runs in operation, not only until it ships. Nothing after the handover depends on us coming back.
THE THESIS
Production is integration.
Most AI initiatives don't fail on the model. They fail on everything around it: permissions and security, integration with the systems people actually work in, data quality, evaluation, monitoring, and clear ownership of architecture and IP. The model is a fraction of the work. The value, and the risk, sit in the rest.
That is why so few pilots reach operation, and why the ones that do rarely got there on enthusiasm. They got there on integration discipline: production criteria from day one, an owner for every component, and a definition of working that was written before the build started.
BUILD, OPERATE, TRANSFER
The transfer has a date, not a hope.
Build-Operate-Transfer is how large infrastructure gets delivered without locking the owner in: one party builds and runs the asset until it is stable, then transfers it, whole, to the owner. We apply the same model to AI. It is the delivery model behind the Operation step of our method.
Our internal test is blunt: if it fails on Christmas Eve, the call should go to a runbook your team owns, not to a consultant you hope picks up.
- 01
Build
Designed and built against production criteria from day one: security, data protection, integration, evaluation. We lead the design and the build. Where deeper engineering is needed, a development partner or your own IT builds under our direction, and responsibility for the whole stays with us until handover.
- 02
Operate
Time-limited by design. We run the system in your real environment, with your people alongside, until it holds: evaluated against the criteria we agreed, adopted in the actual workflow, stable under real load. The phase has an end. It does not fade into a contract.
- 03
Transfer
Ownership moves to you in full: code and configuration, models and prompts, documentation, the evaluation framework, an operating plan that includes model-migration readiness, and people on your side who ran it with us. When we leave, nothing breaks. That's the point.
THE BOUNDARY
What we build, and what we won't.
We build the kind of AI an organisation can own: decision support and case preparation, document and workflow automation, internal assistants and agents on your own data, and the integrations that put them inside the systems your people already use.
What only a vendor could keep alive, we won't build. A system that needs a supplier on call forever isn't an asset, it's a dependency with your logo on it.
When a use case genuinely requires perpetual specialist operations, we say so, and we help you procure that from someone built for it, with your interests held through the procurement.
And sometimes the honest recommendation is a standard platform, a simpler automation, or not building at all. We are paid for the advice, not the build, so we can afford to give it.
THE RULES DURING THE BUILD
Compliant by design, not audited after.
Risk classification, logging and data protection are designed in during Build, not retrofitted. If your use case touches the high-risk list, the preparation work runs alongside the build, not after it. How we read the regulation, calmly, is on our EU AI Act page.
WHERE IMPLEMENTATIONS START
We don't build without a decision basis.
Almost every implementation we take on begins as an AI Readiness Assessment: the prioritised use case, the data reality, the risk classification, the plan. If you ran a pilot that stalled, the assessment starts there and scopes the path to production. If you already hold an equivalent basis, we start from yours.
AFTER TRANSFER
You won't need to call us again.
Yours to run, to extend, or to rebuild with whoever you choose. The documentation is written so that any competent partner can take it forward. Including one that isn't us.
If what you want after transfer is continued strategic direction rather than operations, that is a separate, voluntary decision: Fractional AI Leadership exists for organisations that want AI judgement kept close to the leadership team. No bundled continuation, and no retainer required to keep what we built running.
QUESTIONS
Before you commission an implementation.
We design and lead every AI implementation, and we build the lighter tools ourselves. Where deeper engineering is needed, a development partner or your own IT builds under our direction, with responsibility for the whole staying with us until handover. Your own developers and administrators work alongside us throughout, which is also what the AI literacy duty in Article 4 of the EU AI Act, in force since 2 February 2025, is there to promote among the people who will run the system.
The operate phase of an AI implementation is time-limited, and its length is set before the build starts: the criteria the system must hold in real use, and the period we run it together with you. Where an obligation carries a date, such as the transparency requirements in Article 50 of the EU AI Act that apply from 2 August 2026, the phase is planned so the criteria are met before it. The phase ends in transfer, never in an open-ended contract.
Before an AI implementation starts, we need a decision basis: a prioritised use case, an honest read of your data, and a risk classification. The classification asks whether the use case lands in Annex III of the EU AI Act, whose requirements apply from 2 December 2027. Recruitment, Annex III point 4, is the most common route into the high-risk rules for a company that otherwise stays clear of them. An AI Readiness Assessment produces that basis, or we start from an equivalent you already hold.
The price of an AI implementation is set once the decision basis exists, not before. Pricing a build before the use case, the data and the risk classification have been examined is guesswork, and the guess always ends up as your cost. So we scope the build and operate phases against a basis, with criteria for what counts as done and a date for the handover. The assessment that produces that basis does have a fixed scope and a fixed price, agreed before we start.
The handover of an AI implementation includes everything needed to run the system without us: code and configuration, models and prompts, documentation, the evaluation framework, an operating plan with model-migration readiness, and people on your side who already ran it during the operate phase. The documentation also serves as input if you later build a management system to ISO/IEC 42001, which is not a harmonised standard under the EU AI Act and gives no presumption of conformity.
Every model change in an AI implementation is a small migration: prompts drift, evaluations have to be rerun, and integrations rechecked, so the handover includes a migration plan and the evaluation framework to execute it. A change of model or supplier can also move personal data, and in Sweden the data protection authority Integritetsskyddsmyndigheten supervises AI that processes personal data. Your team, or any partner you choose, can move when the market does.
No, we sell no managed services and no operations contracts for an AI implementation. Operations belong to you, or to a partner built for it, and we design the handover so that the choice is genuinely yours. Obligations under the EU AI Act follow the provider and the deployer, so the ability to answer for the system needs to sit with you rather than with a supplier you would have to call. What we offer afterwards is advice, not dependence.
INSIGHTS
Recent analysis on this work.
Food manufacturing bought AI for language, not for production
The share of Swedish food producers using at least one AI technology went from 4.71 to 21.39 percent in two years. The share using AI for production processes fell over the same period, and that decides what the sector figure is worth to you.
Read the analysis
Machine logs can be analysed before the integration is built
A language model was run across 16,316 maintenance logs and corrected the maintenance type on one record in four. The logs carried the answer all along; the summaries built from them did not, and that is the difference between owning data and having asked it anything.
Read the analysisStart with the decision basis.
Implementation is where AI initiatives are won, and it is decided before the first line of code: in the use case, the data, and the ownership. One conversation tells you whether you have that basis or need it built. And if the honest answer is that you should not build, we will say so.