An AI readiness assessment answers a simple question: if you started using AI on real work next month, what would break, and what would pay off? Most mid-market companies are more ready than they think in some areas and badly unprepared in others. The point of the assessment is to find out which is which before you buy tools, hire an agency or announce an initiative.

We run these assessments for owners and executive teams, and we use the same framework on our own operations. This article gives you the checklist so you can run a first pass yourself. It takes a few days of honest conversation, not a six-month study.

Why most AI projects stall

AI projects rarely fail because the model is not smart enough. They stall for ordinary reasons:

  • The data the AI needs lives in a system nobody can get an export from.
  • The process being automated is not written down, and three people do it three different ways.
  • Nobody owns the result, so the pilot works and then quietly stops being used.
  • Legal or IT finds out late and freezes everything.

An AI readiness assessment is mostly about surfacing those problems early. The technology questions are usually the easiest part.

The checklist: six areas to score

Score each area from 1 (not ready) to 3 (ready). Be honest. A 1 is not a failure; it tells you where the first dollars should go.

1. Systems and access

AI is only useful if it can reach your information. List the systems where real work happens: email, file storage, accounting, CRM, e-commerce platforms, ad accounts, ticketing, your ERP.

For each one, ask:

  • Does it have an API, or at least a reliable export?
  • Who holds the admin login, and is it a person or a shared account?
  • Is there an existing connector for the AI tools you are considering?

The connector question matters more than it used to. Protocols like MCP (Model Context Protocol) now let an AI assistant read from and act in business systems through a standard interface. We connect Claude to Gmail, Outlook, Google Drive and Sheets, Shopify, QuickBooks, Search Console, GA4 and ad platforms this way. If your core systems have connectors, you can move fast. If they are a locally installed program with no API, that is a 1, and it shapes everything else. Our article on what MCP servers are explains the mechanics.

2. Data quality and ownership

You do not need a data warehouse to start. You do need data that is roughly right and has an owner.

  • Can you pull last month's revenue by product or customer without a week of cleanup?
  • Do your systems agree with each other (for example, does the e-commerce platform roughly match the accounting system)?
  • When a number is wrong, does someone know whose job it is to fix it?

A useful test: ask two people for the same figure from two systems and compare. The gap tells you how much cleanup sits ahead of any dashboard or AI analysis.

3. Process clarity

Pick the five tasks that eat the most staff hours each week. For each one, ask whether a new hire could do it from written instructions. If the answer is no, the task is not ready to automate yet, but it may be ready to document, and AI can help with that.

Good first candidates share a few traits:

  • They repeat on a schedule (daily reports, weekly reconciliations, monthly close prep).
  • The inputs are digital and live in systems you can connect.
  • A wrong answer is caught before it reaches a customer.
  • Someone currently spends hours on them and would gladly stop.

Poor first candidates are tasks that depend on judgment that has never been written down, or where a mistake goes straight to a customer or regulator.

4. People and ownership

Every AI workflow needs a named owner who uses it, notices when it drifts and decides when to change it. Ask:

  • Which leaders already use AI tools in their own work? They are your early adopters.
  • Who would own a new automated report or agent day to day?
  • Is there time in anyone's week to learn, or is the team already underwater?

Executive use is the strongest predictor we see. When the owner or CEO uses AI daily, adoption spreads. When it is delegated to a junior employee as a side project, it usually fades. This is why we offer executive AI coaching alongside strategy work.

5. Risk, security and policy

This is the area that freezes projects late if you skip it.

  • Do you have a written policy on which AI tools staff may use and what data can go into them?
  • Are you on business plans for AI tools, with the data-handling terms that come with them, or on personal accounts?
  • Which data is off-limits (health records, payment card data, personal data subject to privacy laws, data under customer contracts)?
  • Do your customer contracts restrict where their data can be processed?
  • For any AI that sends email, posts content or moves money, is there a human approval step?

You do not need a 40-page policy. A one-page rule set that names approved tools, banned data types and required approvals is enough to start, and it gives staff permission to use AI instead of hiding it.

6. Budget and measurement

Decide in advance how you will know a project worked. Hours saved per week, faster month-end close, fewer missed follow-ups, more meetings booked. Pick one number per project and measure it before you start, so you have a baseline.

Budget should cover more than software. Plan for setup, connecting systems, testing, and someone's time to own the result. Tool prices change often, so check current pricing rather than relying on what you saw last quarter.

Reading your scores

Add up the six scores. Then look at the pattern, not just the total.

Pattern What it usually means Where to start
High systems, low process Data is reachable but nobody agrees how the work is done Document two or three core processes first
High process, low systems Clear procedures trapped in hard-to-reach software Fix access: exports, APIs, connectors
Low risk and policy score Staff are probably using AI already, without rules Write the one-page policy this month
Low ownership Pilots will work and then fade Name an owner before you build anything
Mostly 2s and 3s You are ready for real projects Pick one high-value workflow and build it

What a good first project looks like

The best first project is boring, frequent and measurable. Examples we have built or seen work well:

  • A daily sales report that pulls from Shopify, Amazon and accounting and lands in the inbox every morning.
  • A monthly billable-hours reconstruction that reads sent mail, file activity and the calendar, then drafts a line-item timesheet for review.
  • An audit of an advertising account that produces a prioritized list of fixes in a spreadsheet.
  • Drafting routine email replies in the owner's own voice, with a person approving every send.

Each of these saves real hours, uses data that already exists, and keeps a human in the loop. None requires a new platform. Once one is running and trusted, the second and third come much faster because the access, policy and ownership questions are already answered. Our guide to Claude skills shows how repeat tasks like these get packaged.

Running the assessment in practice

A first pass can be done in about a week:

  1. Day 1: list systems and who holds admin access.
  2. Days 2 and 3: interview the people who do the five most time-consuming tasks. Ask them to walk through the work on screen.
  3. Day 4: pull two or three test figures from different systems and compare them.
  4. Day 5: review current AI use (ask openly; people will tell you if it is safe to), draft the one-page policy, and score the six areas.

The output should fit on two pages: your scores, the top three blockers, and one recommended first project with an owner and a success measure.

If you want an outside view, our AI strategy consulting engagements start with exactly this assessment, run by operators who have connected these systems in their own businesses.

Frequently asked questions

How long does an AI readiness assessment take?

A useful first pass takes about a week for a mid-market company, mostly spent in short interviews and testing system access. A deeper version that covers multiple business units or a formal security review can take a few weeks.

Do we need clean data before we start using AI?

No. You need data that is roughly right and reachable for the specific workflow you pick. Many good first projects, like drafting replies or summarizing documents, need very little structured data. Cleanup can happen project by project.

Who should be involved in the assessment?

The CEO or owner, whoever runs finance, whoever owns IT or systems access, and the people who actually do the high-volume tasks. Skipping the last group is the most common mistake, because they know how the work really gets done.

What if our staff are already using AI on personal accounts?

That is common and worth addressing quickly. Move them to business plans with appropriate data terms, publish a short policy on what data can be used, and treat their experience as a head start rather than a violation.

Should we hire someone full-time to lead AI?

Usually not at first. Most mid-market companies get further by naming an internal owner for each workflow and bringing in outside help for setup and connections. A dedicated hire makes more sense once several workflows are running and need ongoing care.