Regulatory applicability determination
A written and reasoned classification per system and per market. It includes the Art. 6(3) analysis where you are arguing out of high-risk.
Assessment, documentation and assurance run in sequence or on their own. All three produce artefacts that a regulator, a notified body, an auditor or an enterprise buyer can read. We do not deliver slide decks summarising what someone else should build.
Assessment
01
We look at the application as built: its models, data flows, human review steps and the points where it affects a person. We work out which regime applies in each market you sell into. Then we tell you what has to change in the product as well as in the policies.
Built against
A statement of intended purpose, deployment context and affected populations precise enough to survive a classification argument
Your role in the value chain as provider, deployer, importer or distributor. Most scoping goes wrong here first
Model and data lineage: third-party models, fine-tuning sets, retrieval sources and what your vendor contracts permit
Automated decision points, the degree of human involvement and whether that involvement is meaningful or nominal
Disclosure surfaces, synthetic-content marking, logging and retention against Article 50 and the equivalent state provisions
Existing controls: we build the programme on what you have instead of replacing it
A written and reasoned classification per system and per market. It includes the Art. 6(3) analysis where you are arguing out of high-risk.
Every gap tied to a specific provision, an owner, an effort estimate and a date. The format is sortable rows instead of narrative.
The interface, logging, oversight and data changes the product needs. We write them so an engineering team can size and schedule the work.
Sequenced against statutory dates and your release calendar. It separates what blocks launch from what can follow it.
Documentation
02
The regulator, the notified body, the enterprise buyer and the auditor all ask for the written record in different words. We write it in the structure the instrument specifies. We build it to survive the next model version without a rewrite from scratch.
Built against
What documentation already exists, in what form and how much of it can be recovered rather than replaced
Where the facts live (repositories, model cards, evaluation runs and incident logs) and who is accountable for keeping each current
The change surface: what a model update, a new data source or a new market does to each artefact
The evidence trail behind each claim: a technical file that cannot be substantiated is worse than none
All nine sections: system description, development process, monitoring, functioning and control, performance metrics, risk management, lifecycle changes, standards applied, declaration of conformity and post-market monitoring plan.
Scoped assessment of benefits and harms to individuals, groups and society. We record the analysis instead of asserting the result.
A continuous and iterative system covering identification, estimation, evaluation, mitigation and residual-risk acceptance. It is not a one-time register.
Sized to the organisation: policy, objectives, roles, Annex A control applicability, internal audit programme and management review.
Drafted with the conformity route and applied harmonised standards established and evidenced.
What is collected, how often it is reviewed, what triggers re-assessment and how it connects to serious-incident reporting under Art. 73.
Deployer-facing documentation covering capabilities, limitations, expected performance and required human oversight measures.
Assurance
03
Most AI risk assessments stop before anyone proves the system behaves as claimed. We design the evaluation: metrics, datasets, thresholds and cadence. Then we set up an audit process that produces evidence on a schedule instead of during a fire drill.
Built against
The system's claims restated as testable propositions that define the population and what counts as failure
Which metrics carry regulatory weight for your classification and which are only comfortable to report
Subgroup performance and disparity measures chosen to fit the decision before anyone sees results
Robustness under distribution shift, adversarial input and prompt injection where the system is exposed to untrusted content
Where oversight in practice departs from the design: override rates, time on task and automation bias
The independence of the review function relative to the team that built the model
We fix the metrics, datasets, thresholds, acceptance criteria and the statistical basis for each before the run.
Every metric with its definition, owner, collection method, reporting cadence and the decision it is allowed to inform.
Examination of possible biases likely to affect health, safety or fundamental rights. We record each mitigation and its residual effect.
Structured adversarial testing scoped to the generative-AI risk taxonomy. Findings are tracked to closure.
Scope, schedule, criteria, auditor competence and independence, reporting lines and the management review that consumes it.
Independent validation for regulated financial institutions. It covers conceptual soundness, data quality, outcomes analysis and ongoing monitoring.
The scope phase is meant to shrink the rest of the engagement. It routinely does.
We inventory the AI in the product, establish your role in each value chain and confirm which markets and regimes apply. Most engagements narrow here. The scope you feared is usually larger than the one you have.
Classification, gap analysis against the applicable provisions and a read of the controls you already run. We work from the system as built and with the engineers who built it.
We draft the artefacts and you review them. Then we fix the places where the record and reality disagree. Finding those disagreements is the useful output. The document follows from it.
Evaluation design, the first full run and the thresholds that decide what ships. If you have no measurement capability yet we build the harness with your team instead of handing over a specification.
The audit calendar, the monitoring plan and the review that keeps documentation current through model changes. Governance programmes usually fail at this stage. It is the part we hand over most carefully.
Fees are fixed against a fixed scope or set per phase once the scope is known. We do not price on hours.
Fixed fee for a fixed scope.
Classification, applicability across your markets and a prioritised gap register. That is enough to brief a board, size a programme or answer a buyer's security review honestly.
We build the full record once.
Assessment through documentation and evaluation. It ends with a complete conformity file and a working audit calendar. This is the usual shape for a team with a real deadline.
Standing capacity.
A named principal, a monthly review and documentation maintained through your release cycle. It suits organisations that ship model changes faster than a project engagement can absorb.
We specify them and work alongside the engineers who implement them. We do not take over a codebase. For a substantial change such as logging architecture, an oversight interface or an evaluation harness we write the specification at a level an engineering team can size. Then we review the implementation against it.
No. Conformity assessment by a notified body and certification against ISO/IEC 42001 must be performed by accredited bodies independent of the party that prepared the documentation. We prepare you for those assessments and support you through them. We cannot also sit on the other side of the table.
We design the artefacts around that question from the start. Version-dependent content sits apart from stable content. A documented change-assessment step decides whether a model update is a substantial modification that requires re-assessment. Teams that write a technical file as a single narrative document rewrite it every quarter.
Diagnostics are fixed-fee against a fixed scope. Programme work is priced per phase after the scope stage. You commit to the assessment before you commit to the documentation. Retained engagements are a monthly fee against a defined response time and review cadence. We do not price on headcount or hours.
We will tell you what attaches to it and whether you need us for four weeks or six months.