AI readiness is not the number of tools a company has tested. It describes whether a measurable business case, suitable data, connected systems, accountable roles and capable users are strong enough together to move a solution into safe operation.
A business is AI-ready when it has at least one material, measurable process; lawful and sufficiently reliable data; systems that can be integrated; an accountable owner; and users who can review and adopt the result. If one of these foundations is missing, a broad rollout should not begin, although a tightly scoped learning pilot may still be useful.
The six dimensions of AI readiness
| Dimension | Core question | Evidence |
|---|---|---|
| Strategy | Which business objective improves measurably? | Prioritised business case |
| Process | Is the workflow stable, frequent and decidable? | Process map and baseline |
| Data | Are relevant, lawful and current data available? | Inventory and quality sample |
| Technology | Are integrations, identities and operations viable? | Architecture and security review |
| Governance | Who approves, monitors and stops the system? | Owner, controls and escalation |
| People | Can and will users review the result? | Role plan, training and feedback |
A simple 0–3 scorecard
Score each statement 0 = absent, 1 = partial, 2 = reliable and 3 = proven in operation. Do not rely on the average alone: a zero for lawful data access or accountable ownership is a stop condition.
- The objective has a measurable KPI
- A business owner owns the outcome
- The current process is documented
- Representative test cases exist
- Sources and access rights are known
- Data quality has been sampled
- Integrations and identities are understood
- Security and operations have capacity
- Failure impact and human review are defined
- Users were involved early
- Training and support are planned
- Success and stop criteria are measurable
| Score | Interpretation | Next step |
|---|---|---|
| 0–12 | Exploration | Build foundations and run a small learning case |
| 13–24 | Pilot-ready | Run a tightly controlled pilot |
| 25–31 | Rollout-ready with gaps | Close gaps, then scale in stages |
| 32–36 | High maturity | Expand portfolio, monitoring and reuse |
Go and no-go criteria
Go
- A recurring problem and a real user are identified.
- A baseline exists for time, error, quality or cost.
- The solution may lawfully access the required data.
- A person can review, correct and stop critical outputs.
- An owner has time, budget and authority.
No-go or remediate first
- The process is rare, unstable or not understood.
- Sources conflict and nobody maintains them.
- The provider's handling of confidential data is unclear.
- Errors could create unnoticed, irreversible harm.
- The business case is simply “we need AI too”.
Why readiness is more than data
OECD research highlights reliable IT infrastructure, data management and workforce skills as important adoption conditions. Many firms train employees, use external data and still struggle to find the right expertise. Readiness is organisational and technical at the same time.[1]
Good data cannot rescue a poorly governed process. Equally, a sound workflow fails if users do not trust the system or lack time to review it. The dimensions must therefore be assessed together.
Choosing the first use case
| Criterion | Favourable | Unfavourable |
|---|---|---|
| Frequency | Daily or weekly | A few exceptions per year |
| Measurability | Time, errors, quality or conversion | Only subjective improvement |
| Risk | Error is caught before impact | Automatic irreversible decision |
| Data | Bounded, maintained and authorised | Unknown, fragmented or contradictory |
| Adoption | Team feels a clear pain point | Solution is imposed from above |
Four assessment phases
Objectives and portfolio
Clarify business goals, collect processes and prioritise by value, feasibility and risk.
Evidence over opinion
Sample process data, documents, access paths, system boundaries and real user questions.
Gap plan
Give every gap an owner, action, effort and due date—from data cleanup to training.
Pilot contract
Set scope, metrics, controls, security boundaries and stop criteria before development.
Readiness differs by use case
A company may be ready for automated proposal drafts and not ready for AI in recruitment. Assessment should therefore attach to a concrete process. Company-wide statements such as “our data is good” or “we are not digital enough” are too broad to approve or reject an investment.
| Maturity | Characteristic | Suitable activity |
|---|---|---|
| 1 · orient | Objectives and ownership are open | Process interviews and portfolio |
| 2 · prepare | Use case clear, data or operations incomplete | Data, integration and governance work |
| 3 · pilot | Controlled scope and measurable baseline | Bounded end-to-end pilot |
| 4 · scale | Quality and operations repeatedly proven | Rollout with monitoring and support |
Which metrics belong in the pilot?
A pilot is not a demonstration; it is a comparison with today's workflow. Alongside speed, measure quality, rework, exceptions and adoption. A system that is 30 per cent faster but doubles review work may not be an improvement.
- Handling and waiting time before and after
- Share of outputs usable without major correction
- Errors by type and potential impact
- Minutes spent on review and rework
- Unanswered or escalated cases
- Cost per successfully completed case
- User adoption and actual usage
- Incidents, access and data issues
Misleading readiness signals
Many logins do not prove adoption, a polished demo does not prove process quality, and high model accuracy does not prove commercial value. A pilot that tests only easy examples is equally misleading. The evaluation set should contain normal cases, edge cases, poor input, missing data and deliberate counterexamples.
Scale only when the use case meets its target range, material errors are controlled, operations and support are clear, and accountable business teams are willing to own the solution.
Frequently asked questions
How long does an AI readiness assessment take?
A clearly bounded business area can often be assessed in two to four weeks. A company-wide portfolio across locations and systems takes longer.
Do we need perfect data?
No. Data must be sufficient, lawful and testable for the use case. A pilot may reveal gaps but must not hide them.
Can a company be ready without in-house AI developers?
Yes. It needs business ownership, integration capacity and internal or external capability for selection, testing and operation.
What is the most important output?
Not a single score, but a defensible decision: which use case starts, which gaps come first and how success will be measured.
