Tool
AI ROI calculator: what an AI agent is worth in your workflow
Work out in two minutes how many hours an AI agent frees up in one process, what that is worth net per year, and which build budget pays back within twelve months.
Result per year
The workflow carries a build. The number that matters most is the share the agent can prepare safely; the assessment measures it on real cases.
Assumptions: 1,600 productive hours per full-time position per year; the saving applies only to the share of cases the agent prepares. Faster answers, fewer errors and more consistent decisions are not included.
Check this calculation against real numbers
In the strategy call we go through your assumptions and tell you which one is the shakiest. 30 minutes, free of charge.
Request a strategy call with these numbersHow the calculator works
Working time today is cases per month times twelve times minutes per case. The agent saves only on the share of cases it can prepare inside the rules, and there only on the part of the time a person no longer spends searching and writing, because they review and release instead.
Hours saved times fully loaded cost per hour gives the gross saving. Minus the running cost for model usage, hosting, monitoring and refinement, that leaves the net saving. The one-off build divided by the monthly net saving gives the payback in months.
Where to get the numbers
The case count almost always sits in a system already: the ticket system, the ERP as the number of quotes or complaints, the invoice register. Use the average of the last twelve months, not the peak month.
Handling time is best measured rather than estimated: have two people handle ten cases each with a stopwatch and take the median. Waiting time for answers does not belong in it; the agent shortens that in a different way.
The two most uncertain values are the share the agent prepares and the time saved per case. Set them cautiously. These are exactly the two numbers a Workflow Assessment measures on real cases.
| Workflow | What counts as a case | Where the volume is | What drives the time |
|---|---|---|---|
| Service triage | One incoming request by email, portal or phone note | Ticket system, service inbox | Reading, categorising, searching history in CRM and ERP |
| Quote preparation | One request for quote with its line items | ERP or CRM, quote statistics | Matching items, checking prices and availability |
| Claims and complaints | One complaint or damage claim | QM system, ERP, complaint log | Collecting evidence often takes longer than deciding |
| Document processing | One document: invoice, order confirmation, certificate | Invoice register, DMS | Retyping and matching against order and master data |
What the calculation does not show
The calculator counts working time only. In many workflows the larger value lies elsewhere: a quote that reaches the customer the same day instead of after four days wins more often. A complaint decided by rule rather than by handler costs less goodwill. An invoice posted correctly saves the correction at month-end close.
On the cost side, the internal time of the business and IT during the build and the support for the people affected are missing. Both can be planned, but neither is zero.
When a dedicated build pays off
As a rule of thumb from the calculation itself: if a workflow frees less than half a full-time position, it rarely carries a build of its own with integration, control set and shadow mode. Then it is worth bundling similar cases from adjacent processes, or starting with a workflow that has more volume.
Conversely, if even the cautious calculation shows a payback under a year, the most expensive mistake is to wait too long with a pilot that never writes back to the system of record.
Questions