An AI readiness assessment measures your enabling environment: data governance, platform, funding, skills.
It returns one score for the whole organization. But nothing gets built by an organization. Work gets built on specific data, inside a specific workflow, with a specific person accountable when the output is wrong.
Readiness is a property of the opportunity, not the company, and the same company is usually ready for one and nowhere near ready for another.
What does a readiness score of 78 actually tell you?
Picture a COO at a specialty insurance carrier with her AI readiness assessment open on her laptop. Seventy-eight out of a hundred. Ahead of her peer group on data governance, ahead on executive sponsorship, middling on skills.
Her CEO has already forwarded it to the board with a note saying the company is in good shape.
On Monday she has to decide what the company builds, and the report does not help her. Everything in it is true. None of it answers the question she was handed, which is: ready to do what?
That gap is not a defect in her particular assessment. It is what you get whenever you score an organization on a question that only has answers at the level of the work.
Is the same company ready for every AI opportunity?
No, and the distance between the best and worst candidate is usually measured in quarters. Three were in front of the carrier at 78.
An internal knowledge assistant over the policy and procedure library. Four thousand documents in SharePoint, already permissioned, already owned by an operations manager who can change them without asking anyone. When it is wrong, an underwriter asks a second question. Ready now, and it could start this quarter.
Claims status automation in the contact center. The status data lives in a policy administration system from 2004 and reaches the service layer by overnight batch, so the assistant would confidently tell customers yesterday's answer.
Claims owns half the workflow and billing owns the other half. When it is wrong, a customer acts on it. Partially ready, with gaps that are known, fixable, and not fixable in a sprint.
An agent that issues refunds and premium adjustments on its own. It writes to the general ledger. No one at the carrier has ever granted write access to a non-human actor.
There is no control owner for it, no reversal path, and no settled answer to who is accountable when it pays a claimant twice. Not close, for reasons that have nothing to do with the model.
Same company, same 78, three different answers.
Which gives you a test you can run on your own report this afternoon: swap the use case, keep the company, and see what changes in the findings. If nothing does, it is not measuring readiness for anything you are about to build.
What changes between one AI opportunity and the next?
Seven things, and an organization-level score looks at none of them.
This is not a contrarian reading. Gartner's July 2026 guidance on the AI-ready data stack tells leaders to translate strategy into prioritized use cases first, then "identify the unique data, metadata, modeling and governance needs" of each (Rita Sallam, July 17, 2026).
KPMG's roadmap from September 2026 says it more bluntly still: start with the AI use case, work backward to the data.
Cost of being wrong is the line most often skipped, and it is the same move we make with hallucination risk, which is assessed per use case rather than per model.
Is an enterprise AI readiness assessment still worth doing?
Yes, for what it actually measures. It describes your enabling environment: the platform you would build on, the governance you would inherit, how procurement behaves, whether leadership will fund a second year.
Those conditions are shared across every opportunity, they change slowly, and a company at 40 will be slower at everything than a company at 78. Our earlier piece on what AI readiness looks like at the level of people and timing makes that case; this one is about the work rather than the workforce.
What an enterprise score cannot do is carry information it never collected. The carrier's 78 says nothing about whether model risk policy covers an autonomous actor with ledger access, because the assessment did not ask.
Enterprise readiness gives you context. Opportunity-level readiness tells you what you can pursue.
How do you assess AI readiness for a single use case?
Take one opportunity and fill in eight names, not eight ratings. A name can be checked by someone else. A rating cannot.
1. The data. Which fields, in which system, and can someone show them to you in production rather than in a clean extract? Is the use cleared, or is clearance assumed?
2. The fit. Is the data good enough for this particular question? That is a different test from being good in general.
3. The workflow owner. Who owns the process this changes, and can that person authorize the change without escalating?
4. The write path. Where does this sit in the system people actually work in, what can it alter, and what is the reversal procedure when it alters the wrong thing?
5. The control. Which existing governance, security or compliance control covers this, and who owns it? "We will work out governance later" means you do not have one.
6. The reviewer. Who checks the output before it reaches a customer, an auditor or a ledger, and for how many months does that stay true?
7. The people. Whose work changes, how much of it changes, and what has been done to prepare them for it?
8. The number. What moves if this works, and is it measured today? If it is not instrumented before you start, you will spend the following year arguing about the return instead of reporting it.
Where you cannot supply a name, you have found a gap. Gaps beat scores because a gap is a task with a duration.
"Legal clearance not requested" is two weeks. "No process owner" is a quarter and a political conversation. "No control framework for non-human actors with ledger access" is a program with its own sponsor.
Those three sequence very differently, and the sequence is what turns a ranked list of ideas into a roadmap.
What does this look like on a real engagement?
A national commercial real estate firm ran GenROI before building anything. Discovery surfaced four opportunities: enterprise knowledge access across deal data spread over multiple systems, document generation for BOVs and offering memoranda, title and lease abstraction, and a foundation for AI at scale.
They were not equally ready, and the roadmap said so. The work shipped in three phases across 24 weeks: ten weeks for the core platform, eight for document generation, six for abstraction. Seventy-plus users are in production and the rollout is continuing.
That phasing was not a delivery convenience. It was the readiness answer, converted into dates.
How does GenROI assess AI readiness?
GenROI evaluates every opportunity on potential value, feasibility, strategic fit and organizational readiness, and returns four things: an AI Opportunity Assessment, an AI Readiness Report, a prioritized AI roadmap, and an effort and value summary.
The process runs discover, assess, prioritize, roadmap, in that order, because you cannot assess readiness until you know which opportunities you are assessing it for.
Discovery is AI-guided, which is how it reaches more roles and functions than a workshop can cover, and our strategists then validate, challenge and synthesize what comes back. AI gives the process scale. The judgment stays with people.
The COO's honest answer looks like this. Ready now for the knowledge assistant. Four months from the claims work, and here is what the four months are made of. A year from the refund agent, and most of that year is governance rather than engineering.
Less satisfying than a single number, and it is the version she can take into a budget conversation.
Ready, or not yet
Frequently asked questions
How do you know if a specific AI use case is ready?
Run the eight checks above and count the blanks, because the count is what tells you which of three situations you are in. Zero blanks means build.
One or two means you have a scoped piece of work with a date on it, not a readiness problem, and the build can often start in parallel. Four or more means the opportunity is not ready in a sense that a project plan can fix, and it belongs later in the roadmap behind something that is.
What are the biggest AI readiness gaps?
Ownership and measurement, by some distance. Data problems surface early because somebody tries to query the data and fails.
Nobody notices that the process has no owner until the change needs authorizing, and nobody notices the target metric was never instrumented until someone asks what the return was.
Do you need perfect data before starting?
No, and waiting for it is how programs stall. You need data that is good enough for one specific question.
A dataset that is unusable for demand forecasting can be entirely sufficient for retrieving policy language, and the only way to find out is to test it against the question rather than against a general standard.
Can you start AI initiatives while improving your overall AI readiness?
That is usually the right order. The enabling environment improves slowly, and it improves fastest under the pressure of a real delivery.
Pick the opportunities you are already ready for, use them to build the governance and measurement habits you lack, and let the harder opportunities wait for conditions the first ones helped create.
Related
How to identify AI use cases that are worth building: the selection step that comes before this one. That piece decides which opportunities are worth assessing; this one decides whether you can act on them.
