The business case for an AI coworker: how to calculate the ROI

Four variables

The ROI of an AI coworker is a straight-line calculation. You need four numbers:

• The annual case volume (number of invoices, orders, service tickets, documents).
• The current handling time per case.
• The weighted hourly rate of the people doing the work today.
• The implementation and operational cost of the AI coworker.

The saving is: (volume × handling time × hourly rate) − operational cost. The payback time is: implementation cost / annual saving. Anything else that gets mentioned, quality, lead time, customer satisfaction, is upside, not a prerequisite.

What realistic numbers look like

For well-scoped processes at Dutch mid-market organizations, we see recurring numbers. Implementation cost for a first process: on the order of €15,000 to €30,000 one-off, depending on complexity, exceptions and integrations. Operational cost: a monthly platform fee (starting around €800 per month) plus a per-case price that depends on process type and volume.

For a service-ticket process at 3,500 tickets a year, ten minutes per ticket saved, in our experience that comes down to a payback of under a year. The same holds for invoice processing at 10,000 invoices a year.

The question that comes before the business case

Whether you should do it isn't in the spreadsheet. The first question is whether the process is a fit. Three criteria:

• Repeatability. The process needs a recognizable structure. Variation per case is normal; every case being entirely unique is a red flag.
• Scale. Below a few thousand cases a year the business case gets thin. Between 2,000 and 100,000 cases a year is the sweet spot for most processes.
• Integration. The process has to land somewhere, an ERP, an administrative system, a ticketing system. No working endpoint, no automation.

If the answer to all three is yes, the business case usually isn't a point of discussion.

What not to count

Two traps. One: counting the employee's full salary as 'saved'. In practice the work redistributes, the person takes on higher-value work, they don't leave. So count the hours of the process, not the whole contract.

Two: guessing at revenue impact. Faster order processing sometimes leads to more returning customers and higher margin, but that's a hard claim to prove within a year. Anchor the base case on the hard cost side; treat revenue effects as a bonus.

How fast you know whether it works

A Quick Scan takes about 1.5 hours. In that time we map the process with you, walk through the existing data (volumes, variants, current tooling) and give a realistic range for implementation time and payback. From kick-off to production a first engagement averages eight weeks. In those eight weeks there's no ROI; after them the meter ticks every day.