What happens after an AI readiness assessment.

Most organisations that commission an AI readiness assessment get a report. It has a score, a maturity band, a radar chart with five or six axes, and a list of recommendations phrased as verbs: uplift, establish, embed. It is presented once, it is well received, and then very little happens. Six months later the same organisation commissions another one, because the first told them where they were but not what to do on Monday.

That is a scoping failure, not an analysis failure. An assessment earns its cost in what it commits you to, not in what it tells you. Here is what should come out the other side, and what happens next.

The report is not the deliverable.

A score is a communication device. It is useful for a board paper and almost useless as an instruction. "You are 2.4 out of 5 on data governance" does not tell a delivery team what to build, and it does not tell a CFO what to fund.

The real deliverable of a readiness assessment is a sequenced, costed first engagement - a specific piece of work, with a scope, a price, a duration and a named business outcome, that is defensible as the right thing to do first. Everything else in the report exists to justify that recommendation. If the assessment cannot produce it, the assessment has not finished.

What the output should actually contain.

Four things, in our experience, separate an assessment that moves from one that sits on a shelf.

• A blocking-constraint diagnosis, not a gap list. Every organisation has dozens of gaps. Only one or two of them are actually stopping the next thing. The assessment should name which constraint binds first - usually data access, data quality, or ownership - and say plainly that the others can wait.
• A named first use case with a business owner. Not "customer analytics" but a specific decision or workflow, with a person whose number changes if it works. Use cases without an owner do not survive contact with a budget cycle.
• The honest disqualifications. Which of the use cases on your wish list are not viable yet, and what specifically would have to change. This is the most valuable page in the document and the one most often missing, because it is the one that costs the consultant follow-on work.
• A cost and duration for the first step. A range is fine. An absence is not. If the assessment cannot price what it recommends, it has not thought it through to delivery.

Three findings that come up almost every time.

Across assessments we have run for Australian organisations, the same three constraints recur far more often than the maturity-model literature suggests.

The data is accessible but not agreed. The pipelines work. The warehouse exists. What is missing is a settled definition of the handful of measures the business argues about - active customer, revenue recognised, churn. AI built on contested definitions produces confident answers that different parts of the business reject for different reasons. This is a governance problem wearing a technology costume, and it is usually the cheapest of the three to fix.

Ownership is unassigned. There is no one whose job it is to say whether a dataset is fit for a given purpose. Without that, every AI initiative re-litigates data quality from scratch. Naming owners is close to free and reliably unblocks more than a platform migration does.

The ambition is mis-sequenced rather than too large. Organisations are rarely too ambitious about AI. They are ambitious in the wrong order - starting with the use case that has the best story rather than the one whose data is already trustworthy. Re-sequencing costs nothing and is frequently the single highest-value output of the whole exercise.

Who does the work next.

This is where most assessments quietly break. The people who wrote the report are frequently not the people who would deliver the work, and the recommendations show it: they are written at a level of abstraction that never has to survive implementation.

Ask, before you commission anything, whether the assessors will be on the delivery team. If the answer is no, you are buying a document from one firm and the risk from another. Our own assessments are run by the engineers and consultants who would do the follow-on work, which is a constraint on what we are willing to recommend - you do not write a cheerful roadmap you know you personally have to deliver against. What that engagement then looks like in practice, stage by stage, we have set out separately in what an AI consulting engagement looks like.

Why the first engagement should be small.

The instinct after a readiness assessment is to launch a program. The better move is one well-defined piece of work, fixed in scope, that produces a visible business outcome in weeks rather than quarters.

Not because ambition is bad, but because a readiness assessment is a set of hypotheses about your organisation, and hypotheses should be tested cheaply. A small first engagement tells you whether the constraint diagnosis was right, whether the business owner really engages, and whether your data behaves the way the assessment assumed. All three are frequently wrong in small ways that are survivable at the scale of one workflow and expensive at the scale of a program.

The question of who runs it.

Readiness assessments are sold by global consultancies, by local boutiques, and increasingly as a templated product with a questionnaire behind it. The differences that matter are not the framework - most frameworks are similar and none of them are the hard part.

What differs is whether the assessor has seen the specific failure modes of Australian organisations: the Privacy Act obligations that shape what you can do with customer data, APRA's expectations if you are regulated, the practical reality of a data team of six rather than sixty, and the fact that most Australian enterprises are running a mix of systems accumulated over decades rather than a clean modern stack.

RUBIX has been doing data and AI work in Australia since 2011 - fifteen years - for organisations including ANZ, NAB, AustralianSuper and MUFG. That matters less as a credential than as a source of priors. An assessor who has seen the same constraint resolve badly at four other Australian organisations gives you a different recommendation than one applying a framework for the first time. The evidence for that is in the work itself rather than in any claim we could make about ourselves.

It is also worth being clear about what a readiness assessment cannot do. It cannot tell you whether a given agent will work, only whether the conditions for building one exist. We have written separately about where AI agents do not work yet, and a good assessment should be equally willing to tell you that the answer to your question is "not this one, not yet."

A practical test.

Before you sign off on any readiness assessment, ask the assessor one question: what would you have us do first, what will it cost, and what will be different when it is done? If that answer requires another engagement to produce, the assessment is not finished. If it comes back specific, costed and uncomfortable in places, you have something you can actually act on.

General information only, not legal or regulatory advice. Current as at September 2026.