Every published definition of AI readiness describes the same thing: how prepared an organisation is to adopt and scale AI. Data, governance, skills, infrastructure, strategy. All of it is useful and none of it answers the question that decides whether the money is well spent, which is whether this organisation should be deploying AI at all, and against which specific decision.
- 01AI readiness is a measure of capability to deploy, not of whether deploying is worthwhile.
- 02The common pillars are data, infrastructure, governance, skills and strategy. They are reasonable and they are incomplete.
- 03Almost every readiness framework is published by a party that sells AI adoption, so readiness is framed as a gap to be closed by buying something.
- 04A readiness score of zero on data quality is usually the real finding. The data problem is the project, and the AI is a label on it.
- 05The question a readiness assessment should answer first is which decision the output will change. If nobody can name one, readiness is irrelevant.
What is AI readiness?
AI readiness is a measure of how prepared an organisation is to deploy artificial intelligence and keep it running. It is usually expressed as a score or a maturity level across several dimensions, and it is meant to tell leadership where the gaps are before a project starts rather than after it stalls.
That is a genuinely useful thing to know. A model pointed at inconsistent data, in a business with nobody accountable for the output, will fail for reasons that have nothing to do with the model. Readiness assessments exist because that failure is common and expensive, and because it is visible in advance to anyone who looks.
The limitation is in the question being asked. Readiness measures whether you could. It does not measure whether you should, and those are different findings with different consequences.
What are the pillars of AI readiness?
Published frameworks differ in count but converge on the same ground. Data, meaning whether the information a model would read is accessible, consistent and trustworthy. Infrastructure, meaning whether anything can be deployed, monitored and rolled back. Governance, meaning whether someone owns the decision when the output is wrong. Skills, meaning whether the people who will maintain it exist. Strategy, meaning whether the work connects to something the business is actually trying to achieve.
Some frameworks split these into five pillars, some into three, some into seven. The number is a presentation choice rather than a finding. If you are comparing frameworks and the pillar counts differ, you are comparing packaging.
What matters more is the weighting nobody states out loud. In practice the data pillar decides most outcomes. An organisation strong on strategy and weak on data is not sixty percent ready. It is not ready, and the honest output of the assessment is a data project.
How is AI readiness measured?
Typically by structured interview and document review against a scoring rubric, producing a level per pillar and an overall figure. The rubric is the important part and it is rarely published, which makes most scores difficult to compare across providers.
Treat a single number with suspicion. Averaging across pillars hides exactly the thing you need to see, because readiness is not additive. A blocking weakness in one dimension is not offset by strength in another, and an average is designed to produce the impression that it is.
The more useful output is the list of what would have to be true, in order, before the first deployment is sensible. That is a sequence rather than a score, and it can be acted on.
What does a readiness score not tell you?
Whether the thing is worth building. A high score says the organisation can deploy AI competently. It says nothing about whether this particular application earns the engineering and the permanent maintenance behind it.
That gap matters because the costs arrive after the score. Integration with existing systems commonly runs a third to a half of initial development and is the most frequent source of overrun. Running costs for AI specifically scale with use rather than with headcount, so the bill grows when the thing succeeds. A readiness assessment that stops at capability leaves all of that unexamined.
It also will not tell you which decision changes as a result. That is the question worth asking first, and it is the one most likely to end a project early and cheaply. If the honest answer to "and then what happens" is that somebody looks at the output, there is no value yet, whatever the score says.
Who publishes these frameworks, and does it matter?
Almost all of them are published by organisations that sell AI adoption: platform vendors, cloud providers, implementation firms. That is not dishonesty, it is incentive, and it shapes the frame rather than the facts.
The effect is consistent. Readiness is presented as a gap, gaps are presented as closable, and closing them involves a purchase. A framework produced by a party with no stake in the answer would be free to conclude that the correct next step is to do nothing, and you will notice how rarely that conclusion appears.
We build AI systems, so this applies to us as well. The reason we can write it down is that the assessment work is structured to be able to end in a no, and a no costs us the build.
When is the honest answer not yet?
When the data the model would read is contested. If two systems disagree about the same fact and nobody owns the reconciliation, that is the project. Putting a model on top produces confident output derived from an argument.
When the process is not written down. Automating an undocumented process encodes whatever it currently does, including the parts that are wrong, and makes them harder to see afterwards.
When nobody will own it in year two. Models drift, dependencies change, and an AI system with no owner degrades quietly rather than failing loudly. The first person to notice is usually a customer.
When is the answer no, and how would you know?
When the volume is low. Engineering, monitoring and maintenance do not get cheaper because the task is rare, and a rare task rarely repays them however irritating it is.
When the real driver is that a competitor announced something. That is a genuine force in these decisions and it is worth naming in the room, because it is frequently the one actually deciding. It is also the one that produces projects nobody can defend eighteen months later.
And when a simpler thing would work. A rule, a form, a fixed report or a cleaned up spreadsheet will solve a surprising share of what arrives described as an AI problem, at a fraction of the cost and with none of the maintenance. Checking that first is not a lack of ambition. It is the cheapest test available, and a readiness assessment that never considers it is not assessing much.
We design and implement AI and automation for specific, identified work: research, document handling, workflow routing, decision support and internal operations.