AI Readiness Assessment: What to Check Before You Build

An AI readiness assessment is a structured check of whether your data, workflows, decision owners and people can support a specific AI build. A good one ends in a decision: build this one thing first, fix this data before anything else, or do not start yet. If it ends in a maturity score and a slide deck, it has not done its job.

That is the whole argument of this post. The rest is what we check, what the output should look like, what the published numbers say about how ready companies really are, and when a small team should skip the formal assessment and just build a thin slice.

What is an AI readiness assessment?

It is the step before you spend real money on AI. Someone looks hard at how the business actually runs and asks whether a given AI idea has what it needs to work: data it can reach, a process it can follow, a person who owns the result, and people who will use it.

Two things it is not. It is not a maturity model that grades your company from one to five. A grade feels useful and changes nothing on Monday. And it is not a vendor pitch with a questionnaire attached. If every answer leads to the same product, it was a sales call.

The useful version is a decision input. It narrows a long list of AI ideas down to the one or two that will actually ship, and it names what has to be true before the others can.

What does a good AI readiness assessment check?

Different firms slice this differently, and that is fine. The questions underneath are the same. Here is what we look at in the first phase of our Figure it out, Build it, Ship it method.

Data you can query today. Not data that exists somewhere. A surprising amount of company data lives in a tool nobody has API access to, in a spreadsheet one person maintains, or in PDFs that were never meant to be read by a machine. If the AI cannot get to the data without a human copying it, the project starts with plumbing. We wrote more about that in our post on data readiness for AI.

A workflow someone can describe, exceptions included. Ask the person who does the work to walk you through it. If the answer is "it depends" five times, the exceptions are the real workflow. AI handles the happy path well. The exceptions decide whether it saves time or creates cleanup.

A named owner for when it is wrong. Every AI system is wrong some of the time. Someone has to own what happens next: who reviews the output, who can override it, and who answers when a customer gets a bad result. If nobody can be named, that is the first gap to close.

The people who will use it. Were they asked? What do they do today, and what do they already use on their own? The fastest way to kill an AI project is to build it for a team that was never in the room.

Other published frameworks cover the same ground with different labels. The Rundown's guide to AI readiness assessment (published September 11, 2026, read October 12, 2026) scores five areas: data, current usage, skills, process clarity and risk management readiness. AE Studio's AI readiness page (ae.studio, fetched October 12, 2026) lists six: systems and data, institutional knowledge, workflows and operations, org structure and decision rights, team capabilities, and opportunity mapping. Different cuts, same questions.

What should the result of an AI readiness assessment look like?

A useful result fits on one or two pages and has three parts:

• A ranked backlog. Candidate AI projects in order, each with a rough sense of effort, value and what it depends on.
• A no-go list. The ideas you are not doing yet, and the specific reason for each. This is often the most valuable page, because it stops the team from re-litigating the same ideas every quarter.
• One thin slice to build first. The smallest version of the top project that runs on real data with real users, small enough to ship in weeks.

For comparison, AE Studio's page says its readiness assessments typically take 2 to 4 weeks depending on scope, and lists the deliverables as an AI readiness map, a prioritized opportunity backlog, a data readiness assessment and recommendations for next steps. The page does not publish a price. The Rundown's guide aims for a single page with a score per area, the two biggest gaps and the first three actions.

What all of these share is a next step. A score with no next step is a deck.

How ready are businesses for AI, really?

Fresh numbers are rare, and most of them come from surveys with narrow samples, so read them carefully.

One of the newest is the 2026 Data Confidence and AI Readiness Survey from Veeam and ITWeb, released October 7, 2026. It surveyed 140 data and cyber security leaders at medium and large organisations across 14 African countries, online in June and July 2026. Only 29 percent said they consider themselves advanced or fully ready to govern and scale AI using trusted data. The same survey found that 85 percent had at least moderate confidence they could recover usable data after a cyber incident, yet almost one third had not recently tested that ability.

That sample is specific: large and medium organisations, in one region, answered by security and data leaders. It does not tell you how ready a 30-person company in Texas is. What it does show is a pattern we see everywhere, which is a gap between how confident people feel about their data and how much of that confidence has been tested. Finding that gap before you build is exactly what a readiness check is for.

When should you skip the assessment and build a thin slice?

Sometimes a formal assessment is overkill, and saying so is part of being honest about it.

If you are a small team with one clear workflow, one owner and data you can already reach, you will learn more from building a thin slice than from weeks of interviews. Put a simple version in front of the people who do the work, measure how often it is right against real cases, and let the results tell you what is missing. A small evaluation set and a review step do more than a questionnaire at that size.

Our rough rule:

• Assess first when several teams, several systems or regulated data are involved, or when the list of AI ideas is long and nobody agrees on the order.
• Build first when the scope fits in one person's head and the cost of a wrong first try is a few weeks, not a compliance problem.

In our method, "Figure it out" is the assessment, and we size it to the project. For a single workflow that can be a few days of interviews and data checks. For a whole company it takes longer. Either way, it ends with something we can build. Teams that skip this step entirely tend to hit the common ways AI integrations fail: the model works in the demo and breaks on the first messy record.

How do you run a basic AI readiness check yourself?

You do not need a consultant to get started. A plain version looks like this:

1. List the candidate workflows. Every task someone has said "AI should do this" about. Write them down without judging them.
2. For each one, answer two questions. Can a system reach the data it needs today? Is there one person who owns the result? Anything with two "no" answers moves to the bottom.
3. Talk to the people who do the work. Ask them to walk through the task, including the weird cases. Listen for how often the process changes.
4. Pick one. The top candidate is the one with reachable data, a clear owner and a team that wants it. Build the smallest version of that.

This will not catch everything. It will stop the most expensive mistake, which is building the wrong thing first. If you want a second set of eyes on the list, that is the work we do as part of AI integration for business.

Frequently asked questions

How long does an AI readiness assessment take?

It depends on scope. Published guides range from about two weeks of part-time effort by one owner (The Rundown) to 2 to 4 weeks for a firm-led engagement (AE Studio). For a single workflow, a few days of focused work is often enough.

Do we need clean data before we start?

No. Finding out which data is usable is one of the main jobs of the assessment. You need to know where the data lives, not to have fixed it already.

Who should own the assessment inside the company?

One person with enough authority to make a decision at the end, usually an operations or business lead, with IT or security involved. If ownership is spread across a committee, the assessment tends to end without a decision.

Is an AI readiness assessment worth it for a 20-person company?

Often a light version is. A 20-person company rarely needs a formal multi-week engagement, but a few days spent picking the right first project is cheap compared to building the wrong one. If you would rather talk it through, you can book a consultation or tell us about your project.