AI ROI for Small Firms: Price One Task Before You Buy

Why AI ROI starts with mapping the process

An ROI figure is only as good as its "before" number. For a small firm the useful before number isn't revenue or headcount. It's the cost of one finished task: one invoice processed, one client onboarded, one proposal sent, one contract reviewed.

To price a task you need to know how it actually runs: who touches it, how long each step takes, where it waits, and how often it comes back for rework. That's a process map. Without one, people estimate from memory, and memory usually leaves out the chasing, the re-keying and the checking, which is often where the time goes.

That's why we map before we automate. Our process mapping consultant page explains how a map is built. Once you have it, you can measure AI ROI honestly, because you know exactly which steps changed.

How to measure AI ROI in four steps

1. Pick one task. Frequent, rule-bound and easy to count. "Supplier invoices processed" works. "Better decisions" doesn't.
2. Price it today. Time per task, including checking and rework, multiplied by the loaded hourly cost of the people doing it. Add any per-task fees you already pay.
3. Price it with AI. The remaining human time (someone still reviews), plus the AI and automation costs: seats, usage, the automation platform and any setup.
4. Compare after a month of real work. Not a demo. Count the tasks, time a sample, and check the error rate stayed the same or improved.

That's the whole method behind any honest AI cost benefit analysis. The hard part is step 2, and it's the step most vendors skip.

AI ROI calculator: cost per task, before and after

Enter your own figures. Nothing is pre-filled, because the right numbers are yours: time a few real examples rather than guessing. The same calculator works as an automation ROI calculator for workflows with no AI step at all.

Automation ROI vs AI ROI: what's the difference?

The ROI of automation and the ROI of AI are worked out the same way: cost per task before, cost per task after. The difference is in what can go wrong.

• Plain automation (moving data between systems, reminders, routing) is predictable. Once it works, it keeps working, and the automation ROI calculation is mostly about setup cost and platform fees.
• AI steps (reading documents, drafting, sorting messy input) vary. You need a person reviewing the output, and that review time belongs in the "after" column. Leave it out and your generative AI ROI looks far better than it really is.

Many good projects mix the two: automation moves the work, AI handles the one messy step, and a person approves the result. For how we build that, see AI implementation consultant.

AI productivity gains: what to count and what to ignore

AI productivity gains are real in some tasks and imaginary in others. Count these:

• Minutes removed from a task you can count, measured on real work.
• Fewer handoffs or less waiting, if it shortens turnaround for clients.
• Rework avoided because errors are caught earlier.

Be wary of these:

• "Hours saved" with nothing to show for them. If saved time doesn't turn into more billable work, faster delivery or fewer overtime hours, it's not money yet.
• Vendor averages. A percentage from someone else's customers tells you nothing about your process.
• Seat counts as adoption. Paying for 10 seats isn't the same as 10 people saving time. Check usage.
• Time moved, not removed. If the AI drafts in seconds but a senior person now spends longer checking, the cost may have gone up.

Why so many AI projects show no return

The research is sobering. An MIT report from 2025 found that most generative AI pilots it studied showed no measurable effect on profit and loss; we summarised it in MIT GenAI Divide: why 95% of AI pilots fail ROI. The reasons are familiar from small-firm projects too:

• No baseline. Nobody priced the task before, so nobody can show a change.
• Wrong first process. A showy use case instead of a frequent, boring one.
• Tool bolted on. AI added beside the workflow instead of inside it, so people copy and paste between systems.
• No owner. Nobody is responsible for the result, so the pilot drifts.

Every one of those is a process problem, not a model problem. That's why mapping comes first.

Writing an AI business case for a small firm

An AI business case doesn't need to be long. One page that answers these questions is enough to decide:

Put your figures into the calculator above and you have the numbers section. If the answer is "not worth it," that's a useful result too: you've saved the subscription.

Where small firms usually find a return

We won't quote other firms' savings as if they were yours. But the tasks that most often pass the cost-per-task test share a pattern: they're frequent, rule-bound, and currently involve copying information from one place to another. Typical examples:

• Accounts payable. Invoice capture, coding suggestions and approvals. AP automation ROI is usually easy to measure because invoices are countable. See AI for accounting firms.
• Client intake. Enquiries captured once and routed, with missing details flagged.
• Document drafting. First drafts from approved templates, always reviewed.
• Chasing. Reminders for missing documents and unpaid invoices.

Tasks that rarely pass: one-off strategy work, anything that happens a few times a year, and judgment-heavy advice where the review takes as long as doing it.

Frequently asked questions

Get your before number in 90 minutes

Pick one task. In a free Sketch session we'll map it with you and work out what it costs today, so any AI ROI claim, ours or a vendor's, can be checked against a real number.

Book your free Sketch session