Strategy

How to calculate the ROI of an AI project before you build it

25 October 2026 · 7 min read · Hidden Node

Leaders rightly ask what an AI project will return before approving it. The good news is that a useful estimate does not need complex models. It needs a clear baseline, realistic assumptions and an honest list of costs.

Step 1: measure the baseline

Pick one workflow and measure how it runs today:

• Volume: how many cases per month (claims, invoices, tickets, applications)?
• Handling time: average minutes of staff time per case.
• Cost per hour: fully loaded cost of the people doing the work.
• Quality: error or rework rate, and what an error costs.
• Speed: how long a case waits from arrival to completion.

If you do not have these numbers, sample a few hundred recent cases. A rough baseline is far better than none.

Step 2: estimate the change, conservatively

For each metric, estimate what changes with AI in place. Be conservative:

• Assume only a share of cases can be fully handled automatically, and the rest still need a person, faster than before.
• Assume people still review outputs in the first months.
• Assume adoption ramps up over time rather than starting at 100%.

Time saved per month = volume × share of cases affected × minutes saved per case.

Convert time saved into value only where it really turns into money: lower overtime, absorbed growth without new hiring, or staff moved to higher-value work you can name.

Step 3: count benefits beyond time

Some of the largest returns are not hours:

• Faster turnaround, which can raise conversion (for example, loan approvals) or customer satisfaction.
• Fewer errors, such as overpayments caught in invoice audits.
• Better compliance, through consistent checks and complete audit trails.

Only include these when you can explain how they will be measured.

Step 4: list all the costs

Business cases often forget:

• Discovery, design and build effort.
• Integration with existing systems.
• Model and cloud usage costs, which grow with volume.
• Evaluation and testing on real cases.
• Change management and training.
• Ongoing monitoring, maintenance and model updates.

Step 5: compare and decide

Calculate payback period (months until cumulative benefit exceeds cumulative cost) and a simple yearly return. Then test the case: what happens if volume is half, or if the automation rate is lower than expected? If the project only works under optimistic assumptions, choose a different first use case.

A practical rule

A good first AI project pays back within a reasonable time even with conservative assumptions, has a baseline you trust, and has a named owner responsible for the metric. Our readiness assessment produces this estimate for your top three use cases.