How do I measure ROI from AI automation?

Without a record of how the task works today, any result after launch is a guess. Before anything is built, measure the task as it currently runs.

Time per task

How long it takes a person from start to finish, including switching between inboxes, spreadsheets and the CRM.

Volume

How many times the task happens in a normal week.

Errors and rework

How often something is entered wrong, missed or has to be redone.

Response time

How long a customer or colleague waits between asking and getting what they need.

Measure over a few ordinary weeks rather than an unusually quiet or busy one. A simple tally, or an export from your inbox or CRM, is usually enough. If you are still deciding which task to start with, what tasks to automate first can help.

The return only means something against everything the system costs, not just the build.

• –Build: design, development, connecting your systems and testing.
• –Running costs: AI model usage, software subscriptions and hosting.
• –Maintenance: fixes, updates when a connected tool changes, and small improvements.
• –Review time: staff time spent checking, approving or correcting the system's output.
• –Setup time for your team: testing with real examples and learning the new way of working.

Review time is the cost most often left out. A system that drafts replies still needs someone to read them, at least at first. Our note on what AI automation costs a small business covers what drives each of these.

If a task takes X minutes and happens Y times a week, it uses X multiplied by Y minutes of staff time each week. After automation, the time per task is rarely zero, because someone may still review, correct or approve the result.

1. 01Multiply your current minutes per task by your weekly volume. That is the current weekly time.
2. 02Do the same with the new minutes per task, including review and corrections. That is the new weekly time.
3. 03Multiply the difference by what an hour of that person's time costs you. That is the weekly saving.
4. 04Compare a year of savings with the build cost plus a year of running and maintenance costs.

For example, a physiotherapy clinic's front desk copies each online booking request into its scheduling system and emails a confirmation. The clinic times this over a few weeks and counts the requests. After an automation prepares the booking and a staff member only approves it, the clinic times the approval step the same way. Putting its own numbers through the steps above shows the time difference directly, and whether it covers the running costs.

• –Faster first replies to enquiries, including ones that arrive in the evening or at the weekend.
• –Fewer missed leads, because every enquiry is recorded and followed up rather than lost in a busy inbox.
• –Consistency: every enquiry gets the same questions, the same information and the same follow-up.
• –More complete records in the CRM, which makes reporting and later follow-up easier.
• –Staff time moved to work that needs judgement, such as speaking with customers.

These are harder to price but can still be measured. Track the time from each enquiry arriving to the first reply, the number of enquiries with no follow-up and how complete CRM records are, and record them in the baseline too.

Judge the system after it has run on real volume for long enough to include ordinary weeks, busy weeks and some unusual cases. The first weeks after launch usually include adjustments and people getting used to a new routine, so they rarely show the settled picture.

Review again whenever the task or the volume changes. If the numbers do not justify the running cost, it is reasonable to narrow the system, simplify it or switch it off. Some tasks are better left with a person, and finding that out is a useful result in itself. Our business automation work starts from this kind of baseline for that reason.