An AI readiness assessment is a scored check of whether your business can use AI safely and profitably now. For a service company it covers five areas: your data, your systems and logins, how clearly your processes are written down, your people, and your risk controls. Score each area out of 8, and the total tells you where to start.
Most AI readiness assessments online are written for enterprise IT teams: cloud platforms, data lakes, governance boards. A 12-truck HVAC company or a two-bay auto shop needs a different test. The questions that matter are plainer. Are your jobs and customers in one system? Does anyone else know how a quote gets followed up? Could you give a piece of software its own login without handing it the keys to everything? This page gives you that test, a point scale you can fill in over a lunch break, and what each score band means.
What is an AI readiness assessment?
Definition: an AI readiness assessment measures how prepared a business is to put AI to work on real tasks. It looks at the inputs AI needs (clean, reachable data), the access it needs (systems with logins and permissions), the instructions it needs (written processes), the people who will review its work, and the controls that keep mistakes away from customers and money.
It is not a technology shopping list. The output of a good assessment is a short list of tasks where AI can help this quarter, the order to do them in, and the gaps to fix first. If you search for an "AI audit" or an "AI audit for business," you will find the same idea under a different name. Some firms use "AI audit" to mean checking an existing AI system for bias or compliance. For an owner-run service business, it almost always means: where would AI save time or recover money, and are we set up for it?
Why a service business should score itself before buying AI
AI tools are cheap to try and expensive to get wrong. A chat assistant that drafts a review reply costs little if it misses. An agent that reschedules jobs, texts customers or touches invoices costs a lot if it works from a messy job board or a shared password. The readiness score tells you which kind of AI you can use today.
- Low score: use general assistants for drafting and research, and fix the basics.
- Middle score: run one narrow pilot on a task with clear rules and a person approving every output.
- High score: connect agents to your field software for follow-up, reporting and monitoring.
Scoring first also protects you from a common trap: buying a tool to fix a process that was never written down. AI does not fix an unclear process. It runs the unclear process faster.
The AI readiness checklist (score yourself out of 40)
Answer each question with 0, 1 or 2 points. 0 means no, or you do not know. 1 means partly, or only in one person's head. 2 means yes, and you could show someone today. Each area has four questions, so each area is worth up to 8 points and the full checklist is worth 40.
1. Data: is the information AI needs in one place and accurate?
| Question | 0 points | 1 point | 2 points |
|---|---|---|---|
| Are customers, jobs and invoices in one field service, shop or CRM system? | Spread across paper, texts and spreadsheets | Mostly in one system, with side spreadsheets | One system of record |
| Is each job's status kept current (scheduled, done, invoiced, paid)? | Statuses are rarely updated | Updated by some techs or on some days | Updated the same day, every job |
| Do estimates live in the system, with their outcome (won, lost, open)? | Estimates are on paper or in email | In the system, outcome not tracked | In the system with a clear outcome |
| Are phone calls and web leads logged somewhere you can look back at? | No call or lead log | Call log exists, nobody reviews it | Calls and leads are logged and tied to customers |
2. Systems and logins: can software get in safely?
| Question | 0 points | 1 point | 2 points |
|---|---|---|---|
| Does your main software have an API, an integrations page or a Zapier listing? | Do not know | Yes, but limited or unclear | Yes, documented |
| Can you create a separate user with limited permissions? | Everyone shares one login | Separate users, but all are admins | Separate users with roles |
| Does each person have their own login, with two-step sign-in on email and money accounts? | Shared passwords | Own logins, no two-step sign-in | Own logins and two-step sign-in |
| Do you know who owns each account (phone system, website, Google Business Profile)? | A former employee or vendor owns some | Mostly known | All owned by the company and listed |
3. Process clarity: could you explain the job to a new hire?
| Question | 0 points | 1 point | 2 points |
|---|---|---|---|
| Is there a written rule for following up an unsold estimate (when, how, who)? | It happens when someone remembers | A habit, not written | Written and followed |
| Is there a written rule for what happens to a missed call? | No rule | Someone calls back eventually | Written callback rule with a time limit |
| Is it clear who chases unpaid invoices and when? | The owner, when there is time | Someone, with no schedule | A named person on a set schedule |
| Could you list the exceptions you want to hear about every morning? | No | A few, in your head | Yes, written down |
4. People: who will review and own the work?
| Question | 0 points | 1 point | 2 points |
|---|---|---|---|
| Is there one person who can own an AI pilot for 30 days? | No one has time | Maybe the owner | A named office manager or lead |
| Can that person approve drafts within one business day? | No | Most days | Yes, it is part of the job |
| Does your office team use the field software daily without workarounds? | Lots of workarounds | Some | Daily, as intended |
| Have you told the team why you are trying AI and what it will not do? | Not discussed | Mentioned | Explained, with questions answered |
5. Risk: what keeps a mistake away from customers and money?
| Question | 0 points | 1 point | 2 points |
|---|---|---|---|
| Do you have a rule about what staff may paste into public AI tools? | No rule | Informal | Written rule |
| Will a person approve every outbound message, payment and record change at first? | No | Some of them | Every one |
| Do you know the data settings on the AI tools staff already use? | Do not know which tools are in use | Know the tools, not the settings | Know both, and have set them |
| Could you turn off an automation in one step if it misbehaved? | No | With help from a vendor | Yes, yourself |
On data settings, the defaults differ by plan. OpenAI, for example, says it may use content from its individual plans (Free, Plus, Pro) to train its models unless you turn off "Improve the model for everyone" under Settings, then Data controls. It also says that by default it does not use inputs or outputs from ChatGPT Business, ChatGPT Enterprise or its API to train models. Check the current terms for each tool your team uses. Our secure AI checklist goes deeper on this area.
What does your AI readiness score mean?
Add the five area scores. Then look at your lowest single area, because one weak area can sink a project even when the total looks good.
| Total score | Band | What it means | What to do next |
|---|---|---|---|
| 0 to 14 | Not ready yet | Data is scattered or processes live in people's heads. An agent would have nothing reliable to read. | Use AI for drafting only. Spend 60 days getting jobs, estimates and calls into one system and writing down two processes. |
| 15 to 26 | Ready for one pilot | The basics exist but have gaps. You can run one narrow task with a person approving everything. | Pick the task with the clearest rule, often estimate follow-up or a missed-call callback list, and run it for 30 days next to your current way of doing it. |
| 27 to 34 | Ready for agents | Data is current, logins are separate and someone owns the work. Agents can read your software and draft for approval. | Connect two or three agents with their own logins: follow-up, review requests and a morning exception report. |
| 35 to 40 | Ready to scale | Strong across all five areas. | Add monitoring across systems, and loosen approvals only on tasks with a clean track record. |
The floor rule: if any single area scores 0 to 2, fix that area before anything else, whatever your total. A shop that scores 8 on data and 1 on risk is not ready for an agent that texts customers.
A worked example
This is an illustration, not a client result. A plumbing company with nine trucks runs Housecall Pro. Jobs and invoices are in the system and statuses are current (data: 6). Everyone has their own login, but all office staff are admins and the Google Business Profile is owned by a former marketing vendor (systems: 4). Estimate follow-up happens when the owner remembers, and missed calls get returned "when someone is free" (process: 2). The office manager could own a pilot (people: 5). Nobody has a rule about pasting customer details into AI tools (risk: 3).
Total: 20, so one pilot. But process scored 2, which triggers the floor rule. The first step is not software. It is writing two rules: unsold estimates get a follow-up on day 2 and day 7, and missed calls get a callback within 15 minutes during business hours. Once those rules exist, an agent can watch for estimates and calls that break them and draft the follow-up for the office manager to send.
Common AI readiness mistakes
- Scoring the plan, not the reality. If statuses are updated "usually," that is a 1, not a 2.
- Starting with the hardest task. Dispatch and pricing are tempting, but they carry the most risk. Start where a wrong draft costs little.
- Skipping the risk area. It is the one owners score last and the one that does the most damage when ignored.
- Buying a platform before a pilot. If the task works with drafts and approval for 30 days, then decide what to buy.
- Treating it as a one-time test. Re-score every quarter. The score should rise as you close gaps.
What the free Cortex AI Operations Audit covers
If you would rather not score yourself, the free AI Operations Audit is the done-with-you version of this checklist. It is a 30-minute call about how work moves through your company: how calls come in, how estimates get followed up, how jobs close, how reviews get asked for and how invoices get chased.
What it produces:
- Which tasks to automate first for your company, and which to leave alone.
- Confirmation of which connections are possible with the software you already run (ServiceTitan, Jobber, Housecall Pro, Tekmetric, QuickBooks and others) before you pay anything.
- If Cortex can help, a written plan and one fixed price. If it cannot, you are told so.
The audit itself costs nothing. If you go ahead, the work starts by recording three starting numbers with you (estimates won per month, total Google reviews and how often Google shows your business), and the same numbers are compared on day 90. If none of them is up and you met the written conditions, you get back what you paid. You do not switch software, and nothing reaches a customer without your approval.
Where Cortex fits
Cortex builds AI agents inside the software service businesses already run. Each agent has one job, its own login with the least permission it needs, and a person approves anything that matters: sends, payments and record changes. That maps directly onto the checklist above. If you score in the "ready for one pilot" band or higher, one of our AI agents can start on follow-up or a morning exception report. If you score lower, the honest answer is to fix data and process first, and a general assistant is enough for now.
If you want other routes before you spend, our guides on what an AI consultant does and building an AI roadmap lay them out. Or bring your score to the free audit and start from there.
