How to Calculate ROI Before You Implement AI

Why most AI ROI calculations are wrong

The standard approach is to count hours, multiply by an hourly rate, and divide by the automation cost to get a payback period. A process that costs $4,000 per month in labor and can be automated for $40,000 pays back in 10 months. Simple.

What this misses is error cost. A manual data entry process doesn't just consume labor time — it also produces errors at a predictable rate, and those errors cost money. Wrong quantities in a purchase order lead to short shipments. A missed clause in a contract leads to a renegotiation or a dispute. A delayed report means decisions get made on stale data. These costs are real, but they show up as freight charges, legal fees, or lost deals rather than as a line item called 'manual process cost.'

For most of the workflows we automate, the error cost is between 30 and 60 percent of the labor cost — meaning a $4,000/month labor cost has an accompanying $1,500 to $2,400/month in downstream error cost that never gets included in the ROI model. Miss that, and you're understating the value of automation by a third to a half.

Variable 1: What is the true labor cost?

Labor cost is not just salary. The fully-loaded cost of an employee — including benefits, payroll taxes, management overhead, and office space — is typically 1.3 to 1.5 times their base salary. A $60,000/year employee costs the business $78,000 to $90,000.

For the purposes of an ROI calculation, identify the people involved in the process being automated, estimate what fraction of their time goes to it, and multiply by their fully-loaded cost. If two people each spend 25% of their time on a process, and their fully-loaded annual cost is $90,000 each, the process consumes $45,000 per year in labor.

Be precise about the fraction. Ask the people involved to track it for a week, or walk through a typical week with them and estimate it together. The number is almost always higher than the gut estimate because repetitive tasks expand to fill available attention.

Variable 2: What is the error cost?

Error cost is harder to measure but not impossible. Start by asking: what goes wrong in this process, and what does it cost when it does? For an order intake process, this might be wrong quantities (short shipments, expedited freight to cover), wrong part numbers (returns, rework), or missed deadlines (contractual penalties, lost customers).

Estimate the frequency of each error type and the average cost when it occurs. A 2% error rate on 500 orders per month at $150 average rework cost is $1,500 per month in error cost — $18,000 per year that doesn't appear in the direct labor calculation.

You don't need precise numbers. An order of magnitude is enough to see whether error cost is a significant factor. If your gut says errors are costing you less than $10,000 per year, include it as a rough estimate and move on. If it feels more like $50,000+, it's worth spending more time measuring it.

Variable 3: What is a realistic efficiency gain?

Automation is not 100% efficient. A well-built order intake automation will handle 85 to 95 percent of orders automatically and route the rest to a human exception queue. That exception handling still takes time — but instead of the full process, it's just reviewing and approving flagged items.

A realistic efficiency gain for a document-based workflow is 70 to 90 percent reduction in total time. For a data entry workflow with consistent inputs, it can be 90 to 95 percent. Build your model around the lower end: 70 to 80 percent. If the ROI works at 70 percent efficiency gain, it almost certainly works in practice.

Also model the ramp period. In the first 4 to 8 weeks after go-live, while the team is building confidence in the system and the exception handling is being tuned, efficiency gains will be lower. Model month 1 at 40%, month 2 at 70%, and month 3+ at steady state. This gives you a more realistic payback calculation.

A simple framework to use right now

Step one: write down the process you're considering automating. Estimate the labor cost per month (fully-loaded). Estimate the error cost per month. Add them together — this is your monthly baseline cost.

Step two: estimate the automation efficiency gain (use 75% as a default). Multiply your baseline cost by 0.75 to get the monthly savings.

Step three: get a fixed-fee quote for the implementation. Divide the quote by the monthly savings to get the payback period in months. If the payback period is under 18 months, the project is financially straightforward. If it's under 12 months, it's a strong case. If it's under 6 months, it's urgent.

A note on how to get a reliable quote: insist on a fixed fee, paid after a working proof of concept is delivered. Any consultant who quotes without first building a working demo on your data is quoting a guess. Get the demo first, then evaluate the quote.