Automation proposals often arrive with an attractive headline: this task takes ten minutes, happens hundreds of times a month, therefore automating it will save a large amount of money. That arithmetic is useful, but incomplete.
Time released is only valuable if the work really disappears, the automation is dependable, exceptions remain manageable and the organisation can redeploy the capacity. Build cost, integration work, licensing, monitoring, support and future change all belong in the denominator.
Deloitte's 2025 AI ROI research is a useful warning against assuming that implementation quickly becomes value. In its survey of 1,854 executives across Europe and the Middle East, most respondents reported that satisfactory ROI on a typical AI use case took two to four years. The research is AI-specific, but the lesson applies more broadly: value realisation depends on workflow redesign, infrastructure, adoption and operating change—not technology alone.
The business case should describe how automation changes the economics of a process, not how impressive the automation looks in a demo.
Build the baseline first
Measure the current process before designing the solution. At minimum, capture transaction volume, average handling time, waiting time, error or rework rate, exception rate, labour cost, service-level impact and any direct cost of delays.
Do not rely only on workshop estimates if system logs or work samples are available. A task that “takes five minutes” may take five minutes of handling but two days of waiting. Automation may create more value by reducing cycle time than by removing labour.
Separate hard value from soft value
Hard value can include avoided contractor spend, reduced overtime, lower transaction-processing cost, prevented penalties, lower software cost or headcount capacity that genuinely does not need to be added. Soft value may include faster response, more consistent control, better employee experience or increased management visibility.
Both matter, but label them honestly. Do not count every saved minute as cash unless that capacity can actually be removed or redirected to measurable output.
Include the full automation cost
A useful total-cost model includes discovery, process redesign, build, integrations, licences or usage charges, testing, security review, documentation, training, monitoring, exception handling, maintenance and future modification when surrounding systems change.
AI-based automations may add model usage, evaluation, guardrails and human-review costs. Traditional rule-based automation may have lower variability but still needs ownership and failure handling.
Model three scenarios
Use a conservative, expected and upside case. Vary transaction growth, adoption, exception rate, maintenance effort and percentage of theoretical time savings that becomes real economic value. This makes the decision less dependent on one optimistic assumption.
Annual net benefit = realised labour or capacity value + avoided error or delay cost + other measurable benefit − annual run and support cost.
Then compare the upfront investment to cumulative net benefit to estimate payback, and use NPV or an internal hurdle rate for larger programmes. The specific finance method can vary; the discipline of modelling the whole lifecycle should not.
Compare automation with alternatives
The strongest business case includes the option not to automate. Could the same result come from removing a process step, changing a policy, configuring existing software, improving a form, training users or integrating two systems? If a simpler intervention solves most of the problem, automation may not be the best first investment.
Vendor-commissioned economic studies can illustrate possible value but should not be treated as a forecast for your organisation. Microsoft cites Forrester-modelled ROI for Power Automate and Dynamics deployments; those results depend on the composite organisations and assumptions in the studies. Your own baseline should drive your case.
What better looks like
A credible automation proposal can answer six questions: what problem exists now, how often it occurs, what it costs, what will change, what the automation will cost over its lifecycle, and how the organisation will verify value after launch.
If those answers are missing, the project is still an automation idea—not yet an investment case.
Sources and further reading
- Deloitte: AI ROI—the paradox of rising investment and elusive returns.
- Microsoft: Maximize productivity with AI automation (includes Microsoft-commissioned Forrester modelling).
- Microsoft Dynamics 365: 2026 Forrester TEI summary (vendor-sponsored modelling).