AI Readiness Assessment: 12 Checks Before You Build

An AI readiness assessment is a check of whether a specific piece of work in your business is ready to hand to an AI system, and what has to change first if it is not. For a company whose team runs on Slack, email, a task app and a CRM, the useful version tests one workflow at a time. Below are the 12 checks, how to score them, and what to fix when one fails. Start with the job your team copies and pastes between tools all week.

Key Takeaways

• Readiness belongs to one workflow. Test a specific, repeatable job before asking whether the company is ready for AI.
• Checks 1 to 4 decide whether a job is worth building: written steps, weekly volume, nameable exceptions and a known cost.
• Checks 5 to 7 decide whether it can be built now. The inputs must live in connectable systems with consistent records.
• A failed check usually points to a process fix that costs attention and makes any later build cheaper.
• Set approval rules before building. The system takes the repeatable steps; people keep approvals, money and anything a customer sees.

What is an AI readiness assessment?

An AI readiness assessment answers one practical question: can this work be handed to an AI system safely? AI readiness means four things are in place for that work. The steps are clear, the information it needs is reachable, someone owns it, and there are rules for what the system may do alone.

There are two versions. The enterprise version scores a whole organization on broad areas like strategy, culture and infrastructure, and it suits a company with a data team and an IT department. The owner version checks one workflow at a time, which suits a business of 20 to 200 people where the operations lead also runs half the tools.

Most businesses sit in the second group. As of May 2026, the U.S. Census Bureau reported that between 17% and 20% of U.S. businesses were using AI. Use rose among firms with at least 20 employees and showed no significant change among smaller ones. What most smaller firms lack is a clear first job to start with.

If you want a second opinion on which job to test first, book a call with us.

Why do enterprise readiness assessments miss for smaller companies?

Enterprise readiness assessments measure the organization around AI, and a smaller company needs a measure of the work itself. Microsoft's AI Readiness Assessment scores seven pillars, including AI governance and security, organization and culture, and model management, over about 45 minutes. Cisco's covers six areas: strategy, infrastructure, data, talent, governance and culture. Those questions assume a platform team is there to answer them. A 40-person company with one operations lead gets a low score and no plan for Monday.

The evidence points to the work and the data as the places projects break. Gartner predicts, in a February 2025 press release, that through 2026, organizations will abandon 60% of AI projects unsupported by AI-ready data. In its survey of 248 data management leaders, 63% did not have, or were unsure they had, the right data management practices for AI. One widely cited 2025 MIT report on enterprise AI reported that only about 5% of custom enterprise AI tools reached production, and it named brittle workflows and poor fit with day-to-day operations among the barriers.

Both findings lead to the same test: can this one job, with these inputs, be handed over?

We are happy to talk through what your stack can connect to before you spend anything.

The AI readiness checklist: 12 checks before you build

Pick one workflow, such as new-client intake, weekly reporting or routing requests to the right person. Sit with the person who does that job today and answer each check yes or no. A "sort of" counts as no.

The work

1. Can you write the steps on one page? Start to finish, in order, including where the work comes from and where it ends. If two people describe it differently, you have two processes, and a system can only follow one.
2. Does it happen many times a week? Volume is what pays for a build. A task that comes up twice a month is usually cheaper to leave with the person who already does it well.
3. Do most cases follow the same path? Exceptions are fine when you can name them. "Refunds over a set amount go to the owner" is a rule. "It depends on the client" is judgment, and judgment stays with a person.
4. Do you know what it costs today? Count the hours per week, how often it goes wrong, and how long work waits in someone's queue. That number is how you will know later whether the system helped.

The inputs

5. Does the information live in your systems? That means the CRM, the inbox, the task app or a shared drive. Anything that lives in one employee's head or on paper has to move into a system first.
6. Can those systems be connected, and do you hold the admin access? Many business tools can share data with other software, but plans and permissions vary. You need the admin logins, or the person who has them, in the room.
7. Are the records consistent enough to trust? Look for one record per customer, required fields actually filled in, and the same status names used by everyone. A system reads duplicates and blanks as facts.

The people

8. Does one named person own the workflow? That person decides how exceptions are handled and gets the alert when something looks wrong. Shared ownership usually means a quiet failure runs for weeks before anyone sees it.
9. Have the people doing the job been asked? They know the shortcuts, the edge cases and the reasons behind odd steps. They will also review the output, so their buy-in decides whether it gets used.

The guardrails

10. Have you decided what the system may do alone? Drafting, sorting, filling in fields and posting internal updates are common starting points. Sending anything to a customer, moving money or deleting records should wait for a person's approval.
11. Do you know which data it must never touch or send? Think contracts, payroll, and health or financial details about customers. Write that list down before any access is granted.
12. Have you agreed a pass or fail number? Pick it before the build: hours saved per week, errors per hundred cases, or time from request to response. A number chosen afterward lets everyone grade their own work.

How do you score the checklist?

Score it with a decision rule instead of a points total, because the checks do different jobs. Checks 1 to 4 decide whether the job is worth building. Checks 5 to 7 decide whether it can be built now. Checks 8 to 12 are normal decisions that get settled during scoping, so a no there simply goes on the to-do list.

That gives three verdicts:

• Yes to all of checks 1 to 7: the workflow is worth scoping for a build.
• Any no in checks 1 to 4: fix the process first.
• Any no in checks 5 to 7: fix the data or the access first.

Here is an illustration, using an invented company. A 35-person firm runs new-client intake like this: a web form lands in a shared inbox, someone creates the company in the CRM, builds a project in the task app, posts in a Slack channel and sends a welcome email. It happens 15 times a week and takes about 40 minutes each time, so checks 1, 2, 4, 5 and 6 pass. Check 3 fails, because each account manager handles unusual contracts their own way. Check 7 fails, because the CRM holds three records for some companies.

The verdict is two weeks of process and data cleanup, then scoping. That is a good result, because the build that follows will be smaller and more reliable.

Bring your results to a call and we will tell you what we would do next.

What should you fix first when a check fails?

Fix the process first, and expect most of those fixes to cost attention, and little else.

Start with the steps. Sit with the person who does the job and write the workflow down as it really runs, workarounds included. Where people handle the same exception differently, pick one way and write that down too. That single page often clears up a surprising amount of back-and-forth on its own.

Then clean the records the workflow reads. Merge duplicate customers, make the fields the process depends on required, and agree on one set of status names. Collect the admin logins for every system involved and keep them where the owner can reach them.

Next, name the owner and measure a baseline. For two weeks, track how many times the job runs, how many hours it takes, how often it goes wrong, and how long requests wait. A shared spreadsheet is enough.

This work pays twice. The process usually gets faster just from being written down and cleaned up. A later build also costs less, because the rules are already decided and the data can already be trusted. When the fixes are done, run the checklist again on the same workflow.

If a check is hard to answer, that is a good conversation to have with us.

How does an AI adoption strategy follow the assessment?

For a smaller company, an AI adoption strategy is the order in which you hand workflows over, and the checklist sets that order. Run it on the three or four jobs your team complains about most. The ones that pass checks 1 to 7 go first, ranked by the cost you measured in check 4. The rest wait until their fixes are done.

The AI implementation plan for that first workflow should widen trust in stages. Start with a person approving every output. As the number from check 12 holds week after week, let the system handle routine cases alone and send only the exceptions to its owner. Anything a customer sees keeps its review step. People stay in charge of judgment, relationships and approvals, and the system takes the copying, sorting and chasing.

For the guardrail checks, the NIST AI Risk Management Framework, released in 2023, is a free reference intended for voluntary use, built around four functions: Govern, Map, Measure and Manage. A smaller company can apply its questions to one workflow at a time.

Examples of what these systems handle day to day deserve their own article, so this one stays on the decision.

A short call is often enough to turn your checklist into a first project.

Who should run your AI readiness assessment?

The owner or operations lead should run it, together with the person who does the job. Give it about an hour per workflow. The people doing the work can answer the first four checks on their own, because they know the job best.

An outside view helps in two places: when checks 5 to 7 are unclear because you cannot tell what your systems can connect to, and when a workflow passes and you want the build scoped. Expect a good consultant to tell you when a workflow should wait.

GetLocalLeads.AI consults on AI systems and also builds and maintains them. You can engage us for advice only, or for scoping followed by a build. Our AI consulting services page explains both options.

Book a call when you want the build scoped.

Frequently asked questions

How long does an AI readiness assessment take?

The self-check in this article takes about an hour per workflow when the person who does the job is in the room. Formal assessments that review a whole organization run longer, depending on how many departments and systems they cover.

Does my data need to be perfect before we use AI?

Your data needs to be consistent for the one workflow you are handing over, which is a much smaller job. That is check 7: one record per customer, required fields filled in, shared status names. Gartner's prediction about abandoned projects concerns data that was never made AI-ready.

What is the difference between AI readiness and an AI adoption strategy?

AI readiness is a verdict on one workflow: can it be handed over now? An AI adoption strategy is the order you hand workflows over in, built from several readiness checks and ranked by what each job costs you today.

What should the assessment give you at the end?

You should end with a verdict for each workflow you tested, a short list of fixes to make first, and a baseline for hours and errors. For any workflow that passes, add written rules for what the system may do alone and what a person approves.

Do I need a consultant for an AI readiness assessment?

You can answer checks 1 to 4 yourself. A consultant earns their place on checks 5 to 7, where you need someone who can tell what your systems can connect to, and when you want a passing workflow scoped. The first call with GetLocalLeads.AI is free.

Where to start this week

Pick the job your team complains about most. Write its steps with the person who does it, count one week of volume and mistakes, and run checks 1 to 7. By Friday you will know whether that job needs fixing or is ready to build. Keep the page of steps, because it becomes the starting spec for whoever builds the system. When you want help with either path, our AI consulting page has a Book a Call button, and the first conversation is at no cost.