0110 business days
Diagnostic
Determine whether one workflow is valuable, measurable and controllable enough to justify a pilot.
€4,000 HTView diagnostic ↘
Controlled AI workflow engineering
AI systems should have a known state.
State Method diagnoses, pilots and builds AI workflows with measurable outcomes, explicit authority and recoverable failure.
Live state / human gateCycle 024
- 01Input
- 02Interpret
- 03Validate
- 04Approve
- 05Act
- Stop
- Log
- Review
- Recover
Approval remains human-owned
One controlled workflow 10 business days Diagnostic · €4,000 HT
Problem / control gap
The model can improvise. The workflow should not.
A plausible model output is not enough. The surrounding workflow also needs an explicit state, defined permissions, validation, a human approval owner, observable failure and a recovery path.
Ambiguous / unbounded
Plausible outputUnknown ownerHidden failureNo safe stop
Controlled / observable
- 01State
The workflow exposes where each case is.
- 02Authority
Tools act only inside defined permissions.
- 03Approval
A named person approves consequential outputs.
- 04Recovery
Failure stops, logs and routes for review.
Protocol / five states
A controlled route from ambiguity to decision.
The method turns one operational workflow into a bounded system whose evidence, authority and failure paths can be inspected.
Apply it to one workflow- 01
Bound
Define exactly where the workflow starts, where it ends and who remains responsible for its outcome.
- 02
Evaluate
Test the workflow on evidence that resembles real operating conditions.
- 03
Control
Decide what the model may propose, what software must enforce and what a person must approve.
- 04
Recover
Make failure visible and recoverable before the workflow can affect real operations.
- 05
Decide
Use the evidence to choose the next controlled step instead of assuming every prototype deserves production.
Commercial path
Start with the smallest stage the evidence supports.
Diagnostic → Controlled Pilot → Production System is a decision path, not a guaranteed funnel. A valid outcome can be to change the design, gather better data, keep the workflow manual or stop.
02Evidence-led scope
Controlled Pilot
Build the smallest useful version with real tools, representative cases and explicit controls.
From €14,000Review pilot scope ↘
03Only after proof
Production System
Harden integrations, reliability and ownership only when the evidence supports it.
Custom scopeUnderstand readiness ↘
Controlled pilot / evidence in use
Prove the smallest useful workflow before production.
The pilot uses real tools and representative cases to test whether the proposed workflow works under controlled conditions. It is a bounded implementation, not an open-ended production build.
- Starting price
- From €14,000
- Boundary
- Smallest useful version
Typical build
- One bounded workflow
- One or a small number of real integrations
- Sandbox or staging environments
- Logs, timeouts, retries and recovery behavior
What it must prove
- The workflow creates enough operational value
- Model performance is adequate on representative cases
- Permissions and deterministic checks work
- Human review and recovery are practical
Pilot decision
- Proceed to production hardening
- Change the design and test again
- Gather better evidence
- Stop without a production build
A pilot does not guarantee production readiness. The result is an evidence-backed go, change or stop decision.
Discuss a controlled pilotFixed-scope diagnostic
One workflow. Five decision artifacts. No obligation to continue.
The diagnostic determines whether one AI-assisted workflow is valuable, measurable and controllable enough to justify a pilot. It is a decision engagement, not a small production implementation.
- Boundary
- One workflow
- Elapsed time
- 10 business days
- Fee
- €4,000 HT
- Decision
- Go / change / stop
Artifacts delivered
- 01Bounded workflow map
- 02Evaluation summary
- 03Permission matrix
- 04Observable failure register
- 05Go / no-go recommendation
What happens next
You receive a proceed, change, gather-more-evidence or stop recommendation. A controlled pilot specification is included only when justified.
Systems already shipped
Technical proof, not borrowed credibility.
Selected work led by Hugo Sequier before the launch of State Method. The projects are construction-related; the control method applies across sectors.
01AnalyzTech
An AI system connecting document reading, floorplan analysis, verification actions and a review interface.
02Floorplan Automation Pipeline
A production pipeline for PDF intake, polygon extraction, post-processing, rule-based checks and reporting.
03Video to Report
A video pipeline that segments uploads, extracts report evidence and prepares a structured draft for office review.
Qualification boundary
Clear boundaries are part of the deliverable.
→Good fit
- One repetitive workflow with a named owner
- A measurable manual baseline
- Representative cases or inputs
- Outputs a person can review
- A real decision about whether to build
×Outside the method
- “Build an autonomous employee”
- No named owner or usable examples
- Unrestricted production authority
- Professional or regulatory sign-off by AI
- Production deployment inside the diagnostic
- Guaranteed ROI or accuracy
H.S. / 07 YEARS
Founder-led engineering
State Method is led by Hugo Sequier.
Hugo is an AI engineer and data scientist with seven years in software development and production experience across AI agents, document intelligence, computer vision and full-stack systems.
View Hugo’s workWorkflow assessment / first step
Bring one workflow.
Describe what happens today, what the AI-assisted step should produce, how the result would be reviewed, and the current volume or baseline.
Business context only No credentials No confidential documents Human-reviewed intake