Most companies do not have an AI problem. They have an AI readiness problem: the appetite is real, the budget is approved, but the data, the skills, the governance, and the use cases are not in a state where AI can produce anything durable.
This is a practical self-assessment you can run this week. Six dimensions, honest signals for low, medium, and high readiness, and a clear next move for each weak score. Use it to stop guessing whether you are ready, and start knowing exactly where you are not.
Readiness is the gap between wanting AI and being able to keep it
Buying a model is easy. Running it in production, safely, with people who trust it, is the part that fails.
The pattern repeats across companies of every size. A leadership team commits to AI, a few pilots get built, a demo lands well, and then nothing reaches production. The cause is rarely the model. It is the conditions around the model: data nobody can vouch for, teams who were never trained, no owner for risk, and a backlog of ideas that were never scoped against value.
Readiness is simply the honest measure of those conditions. A company that scores high can take a good use case and ship it in weeks. A company that scores low will burn a quarter on a pilot that quietly dies, then conclude that AI does not work for them. The technology was never the variable.
The assessment below splits readiness into six dimensions: strategy and sponsorship, data quality and access, skills and literacy, governance and risk, tooling and security, and a backlog of real use cases. Score each one low, medium, or high using the signals provided. Be strict. A generous self-score is the most expensive mistake in this whole exercise.
Score yourself honestly across all six
Strategy and sponsorship. Low: AI is a line in a slide and no named executive owns the outcome. Medium: a sponsor exists but funding is per-pilot and success is undefined. High: a named owner, a budget tied to specific outcomes, and a written view of where AI is worth it and where it is not.
Data quality and access. Low: data is scattered, undocumented, and a new project starts with weeks of plumbing. Medium: core systems are integrated but quality and lineage are inconsistent. High: the data that matters is accessible, documented, and trusted enough that teams reach for it by default. Skills and literacy. Low: a handful of enthusiasts, everyone else watching from a distance. Medium: pockets of capability that do not transfer between teams. High: broad literacy on what AI can and cannot do, plus enough hands-on skill to build, evaluate, and maintain.
Governance and risk. Low: no policy, no named owner, AI use happening informally and invisibly. Medium: a draft policy exists but is not enforced or measured. High: clear ownership, documented controls, and a process that maps to a recognised framework like the NIST AI Risk Management Framework. Tooling and security. Low: shadow use of consumer tools, no review of what data leaves the building. Medium: approved tools exist but access, logging, and secrets handling are uneven. High: sanctioned tooling, scoped access, audit logs, and a security review that AI systems actually pass. Backlog of real use cases. Low: a wish list of buzzwords with no value attached. Medium: a list of ideas, none scoped or sized. High: a ranked backlog of concrete use cases, each with an owner, a value estimate, and a rough cost to build.
What to do with every low score
A low strategy or sponsorship score is the one to fix first, because it gates the rest. Without a named owner and a written view of where AI is worth it, every other investment is a guess. The move is not a bigger strategy deck. It is one accountable executive and a one-page thesis you can argue with.
Low data or tooling scores are slow to fix but predictable. Treat them as engineering work with deadlines, not as a precondition you wait on forever. You rarely need perfect data. You need the specific data behind your top two use cases to be accessible and trusted, which is a far smaller job than a company-wide cleanup.
Low skills or governance scores are the ones companies most often skip, and they are exactly why pilots fail to stick. Literacy is what makes a team trust and maintain a system instead of abandoning it. Governance is what lets you ship without a quiet, growing risk you cannot see. Both can start small: one training cohort, one named risk owner, one policy that is actually enforced.
From a self-score to a prioritised plan
A self-assessment tells you where you are weak. Audit and Consulting turns that into a sequenced plan with owners, costs, and a first use case worth shipping.
An outside read on all six dimensions
We score the same six dimensions you do, but with the questions a team cannot ask itself and access to what is actually happening across tools and data. The output is a clear, defensible picture of where you stand, not a generous self-portrait.
A ranked backlog tied to value
We turn the vague wish list into concrete use cases, each sized for value and cost, then ranked. You leave with a short list of things worth building first and a longer list deliberately parked, so effort goes where it pays.
A plan you own, not a dependency
The deliverable is a sequenced plan your team can run: what to fix, in what order, who owns it, and what good looks like at each step. If you want help building the first use case, Build and Run is there, but the plan stands on its own.
You know exactly where you are, and what to do next
Readiness work has paid off when the conversation stops being whether you are ready and starts being what you ship first.
A ready company can name its sponsor, point at trusted data behind its top use cases, show a workforce that understands AI, and produce a governance owner and a policy that is enforced. It has a ranked backlog where every top item has a value and a cost. None of that is glamorous, and all of it is what separates companies that keep AI in production from companies that keep restarting pilots.
You do not need to be high on all six to start. You need to know your real score, fix the dimension that gates the others, and pick one use case that the rest of your readiness can actually support. That is the difference between an AI program that compounds and one that stalls.