ADVISORY · AI CONTROL DESIGN
Decide what to build, and the rules it must follow.
Most AI programmes fail on the decision, not the code. AI Economics qualifies the use case, defines the economics, and turns supervisory obligations into rules an engineering team can actually enforce.
USE CASE DECISION CARD
Illustrative
Recommendation: proceed to pilot
Customer operations assistant
DECISION
Framed
ECONOMICS
Modelled
OBLIGATIONS
Mapped
HARD RULES
Derived
The rules leave the workshop as a catalogue your engineering team can enforce, whether you work with us or another vendor.
THE DECISION PROBLEM
The expensive mistakes are made before anyone writes code.
By the time a delivery team is briefed, the decision that determines the outcome has usually already been made informally, without economics and without considering the constraints production will impose.
01
The use case is never qualified
A candidate becomes a project because someone senior liked it, not because its value, owner and data were tested against a standard.
02
Economics arrive after the commitment
Inference cost, human review load and the cost of being wrong are modelled once the budget is already spent, if at all.
03
Obligations stay in policy documents
Supervisory requirements are written for auditors, not for engineers. Nobody translates them into constraints a delivery team can apply.
04
No one owns the decision to stop
Without an agreed gate and a named owner, a weak use case is reshaped indefinitely instead of being stopped and its budget released.
FIRST STEP
Use Case Qualification Workshop
A fixed-scope engagement that turns one candidate use case into a decision leadership can defend before delivery funding is committed.
1
Frame the decision
The problem, the business outcome, the named owner and the data actually available today.
2
Establish the baseline
What the process costs in time, money and quality now, so the improvement can be measured rather than asserted.
3
Map the obligations
Which supervisory and internal requirements bind this use case, and what each one demands of the solution.
4
Estimate and recommend
An initial effort and cost estimate with uncertainty made explicit, and a clear recommendation to proceed, reshape or stop.
You leave with a use case card, success metrics, production constraints, a hard rule catalogue, an initial estimate and a written recommendation. A qualified use case can then move to a working PoC in 3 days on MVP Fabric.
ADVISORY MODULES
Three modules. Separate scopes, separate outcomes.
Each module is bought on its own and produces its own deliverable. There is a decision gate after the third: continue to build, or stop with everything the engagement produced already in your hands.
MODULE 01 · 45 DAYS
Discovery
Which AI use cases in the portfolio deserve investment, and in what order.
- Use case inventory and qualification
- Value and cost baseline per candidate
- Ranked portfolio with a recommended first move
MODULE 02 · 60 DAYS
Framework
The AI-SDLC framework and the catalogue of hard rules your delivery must obey.
- AI-SDLC operating model end to end
- Hard rule catalogue derived from your obligations
- Control points, owners and evidence requirements
MODULE 03 · 45 DAYS
Gap analysis
The distance between how you deliver software today and what the framework requires.
- Assessment against the agreed framework
- Prioritised remediation plan with effort
- Decision gate: build, or stop and keep the work
REGULATORY GROUND
Named obligations, not “regulated enterprises”.
Every obligation below is translated the same way: what it demands, what it means for this use case, and which hard rule it becomes for the engineering team.
KNF
Recommendation D
IT area management and ICT security. The document your architecture standards, change management and documentation are actually audited against.
KNF
Cloud processing
What may be processed where, under which classification and with what exit plan, all settled before a deployment target is chosen.
EU
DORA
ICT risk, third-party dependency and resilience testing obligations that reach any AI component sitting in a critical process.
EU
EU AI Act
Risk classification, human oversight and record-keeping duties, resolved for this use case rather than for the organisation in general.
Also covered where relevant: banking outsourcing requirements, internal model risk policy, cloud policy and the organisation’s own AI policy.
DECISION METHOD
Two instruments behind every recommendation.
BOMM
What matters economically
Separates the part of a process where AI changes the cost structure from the part where it only changes how the work feels. Produces the baseline every later estimate is measured against.
DEAL
When a human must intervene
Defines the points at which an AI decision stops being autonomous: what triggers review, who owns the call, and what evidence the intervention leaves behind.
WHAT YOU KEEP
Everything this engagement produces describes your process, not our tooling.
If you never buy the platform, you keep the framework, the hard rule catalogue and the economics. You can implement them with any vendor or with your own teams. The advisory work stands on its own. That is deliberate: a rule catalogue that only works inside one supplier’s product is not governance, it is lock-in.
NEXT
Rules are only real when something enforces them.
Advisory defines the rules. MVP Fabric is the product that applies them to the engineering work itself, so a coding agent cannot merge code that breaks one. If you want to see the enforcement rather than the framework, start there.