An AI audit answers one question: where does AI save your business the most time and money, and what should you build first? It comes before any build. Done well, it stops you from paying for a chatbot nobody uses.
This guide walks through the method we use with clients. You can run a version of it yourself with a spreadsheet and a few conversations.
Signs Your Business Needs an AI Audit
An audit pays off when at least one of these sounds familiar:
- Your team copies the same data between email, spreadsheets, calls and chat every day.
- You tried AI tools, but nobody uses them a month later.
- Leadership wants a cost, a timeline and an expected return before approving any AI spend.
- Different teams are buying their own AI tools, and nobody knows what data goes where.
Step 1: Agree on Goals and Scope
Start with outcomes, not technology. Write down 2 or 3 goals in plain numbers. For example, "answer customer emails within 1 hour" or "cut the time to prepare a quote in half."
Then pick the teams and tools in scope. An audit that covers every department at once ends up shallow. Start with the 1 or 2 teams where work repeats the most.
Step 2: Map How Work Flows Today
Interview the people who do the work, not only their managers. Ask them to show you a real task from start to finish, on screen. You learn more from watching one quote get built than from an hour of meetings.
For each workflow, note:
- The trigger: what starts the task, such as an email, a call or a form.
- The steps: each tool opened and each piece of data copied.
- The volume: how often it happens each day or week.
- The time: how long one run takes, and where people wait.
- The judgment: which steps need a person's decision.
Step 3: List Every Repetitive Task
From the map, pull out each task that repeats with little variation. Good candidates share three traits. They happen often, they follow a pattern, and a mistake is cheap to catch.
Typical examples:
- Reading inbound emails or documents and entering the details into a CRM.
- Answering the same 20 customer questions by phone, chat or WhatsApp.
- Researching a lead or a supplier before a call.
- Writing first drafts of quotes, reports or follow-up emails.
Step 4: Score Each Idea
Give every idea a score, so the ranking comes from numbers, not from whoever argues loudest. We use four measures:
| Measure | Question to Ask | Score |
|---|---|---|
| Hours saved | How many hours a month would this free up? | 1 to 5 |
| Cost | What would it cost to build and run each month? | 1 to 5, cheaper scores higher |
| Risk | What happens when the AI gets it wrong? | 1 to 5, lower risk scores higher |
| Data readiness | Is the data it needs clean and reachable? | 1 to 5 |
Multiply hours saved by the other three scores, or add them if you prefer. Then sort the list. The top 3 to 5 ideas become your shortlist.
Step 5: Check Your Data and Systems
Most AI projects stall on data, not on the model. For each shortlisted idea, check three things:
- Access: can a tool reach the data through an API, an export or a shared inbox?
- Quality: is the data complete and consistent, or scattered across personal spreadsheets?
- Permissions: who is allowed to see it, and does any of it fall under privacy rules?
An idea with a high score but no reachable data moves down the list, or gets a data fix added to its plan.
Step 6: Test the Top Ideas
Before you commit to a build, test each top idea with a small prototype on real examples. Use last month's emails, calls or documents. Measure how often the AI gets it right, and how much a person still needs to fix.
This step turns guesses into evidence. It also shows your team what the tool would feel like to use, which matters as much as accuracy.
Step 7: Build a 90-Day Roadmap
Turn the results into a plan with dates and owners. A useful roadmap shows:
- The first tool to build, with its cost, timeline and expected hours saved.
- What to build next, and what to leave for later.
- Which data fixes need to happen first.
- How you'll measure success, such as hours saved, cost per task and accuracy.
Step 8: Set Guardrails Before You Build
Agree on guardrails now, so every tool you build follows them. On every build, we include:
- A spending cap on each AI task.
- Sources shown with every answer, so people can check it.
- A person approves anything risky, such as refunds or contract changes.
- A log of every AI decision.
Common Questions
How Long Does an AI Audit Take?
It depends on how many teams are in scope. A focused audit of 1 or 2 teams can finish in a few weeks. Your team leads spend a few hours in interviews and reviews across that time.
Do We Need Technical Staff to Run One?
No. Mapping workflows and scoring ideas needs people who know the work. You need technical help for the data checks and the prototypes.
What If the Audit Finds That AI Isn't Worth It Yet?
That's a useful result. It saves you from a build that wouldn't pay off. Often the audit finds a data or process fix that delivers value first, with AI to follow.
