Methodology
The Junyr Method™ Scale: assess an SME’s 5 AI maturity levels
· Updated · 12 min read · Paul-Antoine Tual
Spectateur, Artisan, Orchestre, Architecte and Pionnier describe five cumulative organisational capabilities; the scale helps an SME choose its next stage from observable evidence without treating the adoption of a tool as proof of maturity.
- The level is assessed across nine complementary dimensions, not by counting purchased licences.
- Each stage requires the foundations of the previous one before autonomy or scope expands.
- The self-assessment guides a diagnosis; it certifies neither performance nor compliance.
Why use a scale designed for SMEs?
Large maturity frameworks offer useful language, but an SME needs to translate their ambitions into responsibilities, data, controls and decisions proportionate to its size rather than copy the organisation of a large corporate.
- Starting point: inventory official and hidden uses before designing a target.
- Unit of analysis: examine a process, its data, its owner and its business outcome.
- Evidence: request a verifiable artefact — policy, baseline, log, recovery test or value review.
- Decision: invest in the next missing capability without imposing a disproportionate role or platform.
Available surveys describe adoption and value creation, but their populations and definitions cannot establish how French SMEs are distributed across the five Junyr levels.
- France Num 2025 found that 26% of microbusinesses and SMEs used at least one AI solution in a sample of 11,021 firms.
- The public “Osez l’IA” plan aims for 80% of SMEs and mid-caps to use AI by 2030; this adoption target does not correspond to a Junyr level.
- KPMG classified 11% of its global large-company sample as “AI leaders” and reported meaningful outcomes for 64% overall; its study does not measure French SME maturity.
- The scale therefore remains qualitative until a representative survey applies its criteria and publishes its method.
Overview: the five levels
Each level answers a management question and ends with exit evidence, making progress verifiable without attaching a national prevalence, universal budget or guaranteed duration to it.
The Spectateur
“Which uses already exist, and which ones escape our view?”
Uses are absent, rejected or hidden, so the company does not yet control its exposure or understand its starting point.
- Initial inventory of tools and accounts.
- Minimum rule for data and uses.
Priority: make visible
The Artisan
“How do we turn individual practices into a first controlled use?”
Staff obtain local gains through uneven practices, but the company cannot yet reproduce or govern them.
- Priority case and business owner.
- Baseline, safeguards and decision criteria.
Priority: scope
The Orchestre
“Are our recurring uses coordinated, measured and reviewed?”
Several uses operate in production with human validation, named owners and a shared review of value, quality, cost and risk.
- Documented policy and data practices.
- Periodic review with recorded decisions.
Priority: standardise
The Architecte
“Can the platform and governance remain controlled at scale?”
Integrations, registers, controls, evaluations and continuity procedures form a common architecture that supports wider use without losing control.
- Supervised agents within defined scopes.
- Portability, recovery and exit tested.
Priority: industrialise
The Pionnier
“Have several processes been redesigned and improved durably?”
The organisation redesigns several processes around governed, reversible human–AI cooperation that is independently assessed when the risk warrants it.
- Durable outcomes and continuous learning.
- Risk-adjusted controls that are reassessed.
Priority: renew
The nine dimensions of maturity, level by level
The original nine dimensions remain distinct and complementary: the overall level is the highest stage whose essential capabilities are established across the full scope being assessed.
- Organisational capabilities: augmented staff, cultural adoption and governance.
- Technical capabilities: sovereignty, resilience, technical debt and security.
- Operational capabilities: workflows and agents, value and FinOps.
| Dimension | Spectateur | Artisan | Orchestre | Architecte | Pionnier |
|---|---|---|---|---|---|
| Sovereignty & control of models | Uses remain hidden | A provider is chosen locally | Data classified and providers approved | Governed architecture, portability tested | Dynamic arbitration, substitutable components |
| Resilience & imperviousness to the splinternet | Dependencies unknown | Informal manual fallback | Documented backup and degraded mode | Recovery and exit tested | Multi-scenario continuity exercised |
| Augmented staff | No official use | Individual practices | Priority roles equipped and supported | Assistance integrated into eligible roles | Human–AI cooperation redesigned at scale |
| Cultural adoption & AI literacy | Limits poorly understood | Self-directed learning | Role-based paths and understood rules | Verification and supervision mastered | Continuous learning connected to incidents |
| AI governance | No accountable owner | Isolated sponsor | Owners, policy and periodic review | Operational register and arbitration | Proportionate independent review and continuous improvement |
| Technical debt | Not observed | Scattered scripts and connections | Backlog and minimum standards | Versions, tests and documentation controlled | Debt managed against maintainability goals |
| Security | Attack surface unknown | Uneven local controls | Identities, secrets and logs governed | Threats, incidents and providers tested | Controls reassessed as risks evolve |
| Workflows & agents | Outside processes | Occasional assistants | Production uses with validation | Supervised agents within defined scopes | Governed, reversible multi-agent processes |
| Value & FinOps | No starting point | Claimed gains, scattered costs | Baseline, budget and impact reviewed | Cost per outcome and attributable ROI | Continuous allocation by value, risk and learning |
Read each level as a decision
Level 1: the Spectateur
The Spectateur does not yet know the true extent of use and must create visibility before selecting a project, because the absence of an official tool can conceal individual accounts and data already being exposed.
- Expected evidence: an initial inventory of tools, accounts, functions and data concerned.
- Minimum rule: uses that are prohibited, permitted or subject to approval.
- Next decision: select a measurable business friction and its owner.
Level 2: the Artisan
The Artisan turns the energy of individual uses into a first controlled process, retaining what works while addressing data, quality and dependence on one person as separate issues.
- Data: restrict sources and information according to their classification.
- Quality: define test cases, human validation and stop criteria.
- Transfer: document the practice so that the capability belongs to the company.
- Economics: measure a baseline before attributing a gain to AI.
Level 3: the Orchestre
The Orchestre coordinates several recurring uses and subjects them to the same decision loop, so that adoption, quality, cost, impact and incidents are reviewed together.
- Accountability: sponsor, business owner, technical lead and reference users are named.
- Operations: production uses have support, fallback procedures and versioned changes.
- Value: attributable benefits are compared with TCO over an explicit period.
- Governance: policy, register and periodic review reflect organisational roles and risks.
Level 4: the Architecte
The Architecte builds a shared, reversible capability around its data, integrations and controls, allowing a new use case to reuse tested foundations without depending on one provider or an opaque setup.
- Architecture: documented components, controlled access, evaluated models and common observability.
- Agents: explicit scopes, budgets, validations, prohibited actions and escalation paths.
- Continuity: backup, degraded mode, recovery, model change and provider exit are tested.
- FinOps: reliable cost per outcome, allocation by use and quality-cost-latency trade-offs.
Level 5: the Pionnier
The Pionnier demonstrates across several processes that redesigned human–AI cooperation produces durable outcomes while maintaining proportionate supervision, reversibility and independent review where the risk warrants it.
- Reach: several critical processes share capabilities without concentrating uncontrolled risk.
- Outcomes: value, quality and adoption persist beyond the pilot and its first champions.
- Learning: incidents, evaluations and model changes improve the controls.
- Review: a function independent of the project verifies outcomes and risks when their severity requires it.
Why the levels are cumulative
“No level can be skipped” is a sequencing rule: an SME may explore an advanced technology early, but it should expand it only after establishing the responsibilities, data, measurements and controls required by the use.
- An isolated agent can be tested with synthetic data and blocked actions.
- Production requires authorised sources, supervision and a fallback procedure.
- Expansion to several teams requires shared operations, standards and value review.
- The decision follows the project’s evidence; no generic rate predicts its failure or success.
Define a transition without promising a timetable
The timetable and budget are built from the missing evidence and actual scope because two companies of the same size can have very different data debt, integrations and obligations.
- Spectateur → Artisan: uses inventoried, rule understood and first process scoped.
- Artisan → Orchestre: recurring production uses, baseline and joint review of outcomes and risks.
- Orchestre → Architecte: platform, register, controls, continuity and exit made operational.
- Architecte → Pionnier: processes redesigned at scale, durable outcomes and proportionate independent review.
Self-assessment in twelve questions
This questionnaire prepares a diagnostic discussion by gathering ordered evidence; higher levels require all their prerequisites, so a high total cannot compensate for a critical gap.
- Are the tools actually used, including individual accounts, inventoried?
- Do staff know which data and uses are prohibited, permitted or subject to approval?
- Does each priority case have a sponsor, a business owner and an indicator?
- Does a time, quality, cost or risk baseline exist before the pilot?
- Does at least one recurring use operate in production with a fallback procedure?
- Are adoption, quality, incidents, costs and impact reviewed together each month?
- Does the register identify the company’s role, use, data, provider and risk?
- Are identities, permissions, secrets, logs and integrations controlled and tested?
- Are TCO and ROI calculated for an explicit period and scope?
- Do model changes, incidents and provider exits have an owned procedure?
- Have several processes been redesigned with limits on autonomy and proportionate supervision?
- Are outcomes, risks and controls verified by a function independent of the project when the risk requires it?
The interpretation assigns the highest level for which every gate has been passed, then adds missing evidence to the roadmap for the next stage.
- Spectateur: Q1 or Q2 is still missing.
- Artisan: Q1–Q2 are established, but one or more of Q3–Q6 are missing.
- Orchestre: Q1–Q6 are established.
- Architecte: Q1–Q10 are established.
- Pionnier: Q1–Q12 are established across several processes, with durable outcomes and proportionate independent review.
Turn the diagnosis into a roadmap
The diagnosis becomes useful when it names a missing piece of evidence, an owner and a decision date, then connects each investment to a reliable business outcome rather than an abstract race to the highest level.
- Select: retain a small portfolio based on value, feasibility and risk.
- Prepare: assemble data, rights, evaluation criteria, controls and a fallback procedure.
- Test: measure representative cases with real users.
- Decide: continue, redesign or stop based on quality, adoption, cost, impact and incidents.
- Expand: reuse the foundations only after a documented review.
Frequently asked questions
Can an AI maturity level be calculated with a simple score?
A points total can provide an initial orientation, but the assigned level should be the highest one for which every essential piece of evidence is present.
- Answers are not interchangeable: a technical platform cannot replace a business owner or a usage rule.
- The assessment covers an explicit scope and is then read dimension by dimension.
- A critical gap in security, governance or measurement limits the overall level until it is resolved.
Must an SME reach Orchestre before testing AI agents?
An SME can explore an agent in an isolated environment, while moving it into production requires the foundations that match the process’s risk and reach.
- The test needs an owner, authorised data and evaluation criteria.
- Production adds access controls, logging, supervision and a fallback procedure.
- Investment increases after evidence from the previous stage, without applying a generic failure rate.
How long does it take to move from one level to the next?
No universal duration describes a maturity transition because the timeline depends on scope, data, risk, integrations and the organisation’s capacity for change.
- The Spectateur-to-Artisan transition ends when uses are visible and governed.
- The Artisan-to-Orchestre transition requires recurring uses that are measured and reviewed.
- Architecte and Pionnier require durable evidence across several processes and stronger controls.
Sources for interpreting the surveys
These sources support the retained external figures and should be read with their population, definition and date; they do not validate a national distribution of Junyr levels.
- France Num, 2025 Barometer — AI-solution adoption among French microbusinesses and SMEs.
- French Ministry for the Economy, “Osez l’IA” — the public AI-adoption target for 2030.
- KPMG, Global AI Pulse 2026 — outcomes and the “AI leaders” group in large organisations.