The AI automation audit checklist

An AI automation audit is a structured review of how work moves through your business, used to rank which processes to automate first. This checklist is the one we run in the first week of every engagement.

Work through it with the people who actually do the work, not only the people who manage it. The gap between those two sets of answers is usually where the best automation candidates are hiding.

Answer in writing. Vague answers are the finding.

Sales

1. Where do leads arrive? List every channel: forms, calls, DMs, email, WhatsApp, referrals, walk-ins. Most businesses name three and then remember two more.

2. How long until the first reply — on average, and at worst? The worst case is the number that matters. That is the lead you lost.

3. Who follows up, and what happens when they are busy? If the answer is "it waits", you have found revenue leaking on a schedule.

4. Where does lead data live, and is it the same place for every source? Different sources landing in different places means no one can see the whole pipeline.

5. How many times does a lead get contacted before you give up? Compare the written policy against what actually happens. They rarely match.

6. What happens to a lead that says "not right now"? If the answer is nothing, that is a list of warm prospects being thrown away.

Operations

7. Which tasks are done more than five times a week by hand? Frequency beats complexity. Small, boring, constant tasks pay back fastest.

8. Which tasks require copying data from one tool to another? Every copy-paste between systems is an automation candidate and a typo waiting to happen.

9. Which reports are built manually? Ask how long each takes, and whether anyone reads it.

10. What breaks when a specific person is on holiday? That is a process living in one person's head. Write it down before you automate it.

11. Which approvals create a queue? Find where work sits waiting for someone to say yes, and how long it waits.

12. What do you do twice because the first attempt was not recorded? Duplicated effort is usually a data problem, not a discipline problem.

Customer handling

13. What are the ten most common questions you get? Write the actual answers too. This becomes the knowledge base for any agent you build.

14. How many channels do customers use to reach you? Then ask which of them nobody is formally watching.

15. Is there a written response standard? A target time and a target quality. If it is not written, it is not a standard.

16. What happens to a message received outside working hours? For many businesses this is the single largest, cheapest fix available.

17. How do you know when a customer is unhappy — before they leave? Most businesses find out at the cancellation, which is far too late.

Data

18. Which numbers do you look at weekly, and where do they come from? If a number takes more than five minutes to produce, it is not being looked at weekly.

19. Which spreadsheets are the "real" source of truth? Every business has one. It is usually on someone's desktop and backed up nowhere.

20. If you had to prove what happened on a specific job six months ago, could you? The answer tells you whether your systems are recording or just processing.

Scoring

For each answer, estimate two things:

• Hours per week currently spent on it.
• Revenue at risk if it goes wrong or does not happen.

Multiply them. Sort the list. The top three are your roadmap.

Then apply one filter before you build: cross off anything you could not safely run manually for a day if the automation broke. Those are not first projects — they are third projects, after you have built trust.

How to use the result

The output is not a to-do list. It is an order.

Take the top item, write its process down properly (trigger, steps, tools, decisions, exceptions), name an owner, agree the number you expect to move, and build that one thing. Measure it for a month.

Only then start the second. A business that automates one process properly every quarter will finish the year far ahead of one that started six and shipped none.

Running it well

Three things make the difference between an audit that produces action and one that produces a document:

• Do it with the doers. Managers describe the process as designed. Staff describe it as it runs.
• Time-box it. Two hours, one session. A week-long audit turns into a project of its own.
• Write the exceptions down. "Except when the customer is on account" is the sentence that breaks automations six months later.