AI readiness assessment: a checklist for service businesses

What an AI readiness assessment is for

The phrase sounds like a maturity survey for a large company. For a 30-person firm it should be simpler and more useful. The question is not "are we ready for AI" but "which of our recurring workflows would pay back if we improved it, and what is in the way."

So pick the two or three workflows that cost the most time or lose the most revenue. Typical candidates are lead response, proposals, client onboarding, document intake, invoicing follow-up and recurring client reports. Score each against the checklist below.

Involve the people who actually do the work. The owner knows where money leaks, but the office manager, dispatcher or paralegal knows where time goes and which steps are skipped when things get busy. Thirty minutes with each of them usually changes the ranking.

1. The workflow

A workflow is ready when you can describe it and measure it. Score one point for each statement that is true.

• It happens at least weekly, ideally daily.
• We can write the steps on one page, including who does each step and in which system.
• We know roughly how long it takes per instance, or can time five examples this week.
• Most cases follow the same pattern, with exceptions we can name.
• There is a clear owner who would use and defend an improvement.

2. The data

AI and automation depend on inputs that are findable and consistent. This is where small firms most often score low.

Data is also where a little effort goes a long way. Making job type a required field in your field service tool, or agreeing that every new client is created in the CRM before anything else happens, can move a workflow from fixable to ready in a few weeks. It costs almost nothing and helps reporting even if you never add AI.

• The inputs live in a system, not in someone's inbox or head.
• Key fields are filled in consistently, for example job type, service area or matter type.
• We have 20 or more past examples of good outputs, such as approved proposals or sent reports.
• We know which data is sensitive: client financials, health information, privileged material.
• One system is the agreed source of truth for customers and jobs.

3. Systems, people and risk

The last three areas decide whether an improvement can be built and adopted. Score one point for each.

• Systems: our core tools, such as QuickBooks Online, HubSpot, Clio, Jobber or ServiceTitan, are current and we have admin access.
• Systems: those tools have integrations or an API, or work with Zapier, Make or n8n.
• Systems: we have checked whether an existing tool already has the feature we want.
• People: the staff who do the work have time set aside to test a change.
• People: at least one person uses AI tools weekly and can help others.
• People: a manager will review results and make decisions on what to keep.
• Risk: we have an AI use policy, or will write one before the pilot.
• Risk: we know which steps must stay with a person, such as pricing, legal advice or anything sent to a client.
• Risk: we know whether any regulation applies, for example HIPAA, the FTC Safeguards Rule, or Colorado's SB 26-189 if software would score people.

Reading your score

Add up the points for each workflow out of 24. The exact number matters less than where the gaps are.

18 to 24: a good pilot candidate. Write acceptance criteria and test with real examples.

11 to 17: fixable. The gaps are usually data consistency or an undocumented process. Spend two to four weeks fixing those first. Often that alone saves time.

Below 11: not ready, and that is useful to know. The right move might be a settings change in software you already own, cleaning up the CRM, or simply writing the process down. Buying an AI tool now would automate confusion.

Be wary of averaging scores across the whole business. A firm can be very ready in billing and not ready at all in sales. The point is to find the specific workflow where the gaps are small and the payoff is meaningful.

An illustrative example: two workflows at an HVAC company

Here is how the scoring might play out for a 40-person HVAC company. The details are illustrative, not drawn from a real client.

Workflow A is after-hours call intake. Calls go to an answering service that emails a message to the on-call technician. It happens nightly, the steps are clear, and the company can count missed or late callbacks from its phone logs. But the message details are free text, nothing lands in ServiceTitan until morning, and nobody owns the process. It scores 15 out of 24: fixable, mainly on data and ownership.

Workflow B is maintenance agreement renewals. Renewal dates sit in the field service system, the steps are simple, and the office manager owns them. The data is reasonably clean, and the main gap is that reminders are sent by hand when someone remembers. It scores 20 out of 24: a good pilot. The first step may not even need AI. Scheduled reminders from the existing system, plus an AI-drafted personal note for larger accounts that a person approves, could be enough.

The assessment's job was to show that B is ready now and A needs a process owner and a structured intake form first.

Turning the assessment into a plan

The output of a readiness assessment should fit on two pages. It names the workflows you looked at, ranks them by likely value and readiness, picks one pilot, and records a baseline so you can tell later whether anything improved.

Baselines should be simple measures the team already understands: minutes per proposal, hours from enquiry to first reply, days from signed engagement to kickoff, percentage of invoices paid within 30 days, or how often a draft needs correcting. Time five to ten real examples rather than guessing.

Watch for one trap. Assessments sometimes produce a long list of AI ideas with no owner. Choose one workflow, one owner and one measure. Everything else waits.

InTheory runs this kind of assessment as the first 30 days of an engagement, and sometimes as a paid diagnostic on its own. You can also run the checklist yourself with this page and a spreadsheet.