GuidesAI audit
AI audit: finding the use cases that pay off in your business
What an AI audit is for
Avoiding two mistakes: launching an over-ambitious project that never lands, or piling up trials without ever putting anything into production. The audit answers one question: what to start with, and why.
Step 1: the task inventory
We interview each function, one hour per team. For each repetitive task, we note:
- who does it, and how often;
- how long it takes;
- which data it relies on, and where that data is;
- what happens when it is done badly.
We look for the tasks people do reluctantly: copying from one tool to another, compiling a report, sorting requests, chasing.
Step 2: the state of the data
A task can only be automated if its data is accessible. For each one:
| Question | Good sign | Bad sign |
|---|---|---|
| Where is the data? | In a tool with an API | In emails or in people’s heads |
| Is it up to date? | Someone is responsible for it | Nobody knows |
| Is it sensitive? | No, or it can be anonymised | Health data, contracts |
Step 3: scoring the use cases
Each case gets two scores from 1 to 5.
Gain: time freed, errors avoided, revenue protected.
Feasibility: data available, result checkable, risk if it goes wrong.
We keep those with high gain and high feasibility. Those with high gain and low feasibility become data projects, to carry out before talking about AI.
Step 4: the roadmap
Three projects, not ten.
- 1The firstIs the safest: quick result, zero risk. It builds confidence.
- 2The secondIs the most profitable.
- 3The thirdPrepares what comes next: often a data project.
For each: the scope, the success measure, the build cost and the monthly running cost. See How much an AI project costs.
What an audit should rule out
- Rare tasks: automating something that happens twice a year pays nothing.
- Decisions with high human stakes: hiring, sanctions, credit.
- Broken processes: fix first, automate afterwards.
- Ideas without an owner: an agent nobody is responsible for is abandoned within three months.
An example of a result
For a merchant running several online stores, the inventory brings up the same task everywhere: opening a dozen tools to find out what is going on. That is the problem our project Alfred addresses, with automatic findings at each sync and a Monday report.
The logical next step after the audit is described in Bringing AI into a small business.
Frequently asked questions
- How long does an AI audit take?
- From a few days to two or three weeks depending on the size of the company. Beyond that it is no longer an audit but a study, and the conclusions arrive too late.
- What does the deliverable contain?
- The scored list of use cases, the state of the data for each, an estimate of the gain and the running cost, and a roadmap of three projects at most.
- Can you do the audit yourself?
- Yes for the task inventory, which your teams know better than anyone. An outside view mostly helps judge technical feasibility and rule out ideas that only look good.
