ROI calculations require a baseline
A time-saving claim is only meaningful if you know how long the task takes now. Most AI ROI calculations fail because they start with the tool's promise rather than the business's current cost.
The UK government's AI Adoption Research found that 75% of AI-using businesses self-reported improved workforce productivity, but most had not yet seen a change in revenue. That gap between productivity improvement and revenue impact is worth understanding before building a business case.
The most honest ROI calculation for AI automation is a time-cost comparison: how long does the workflow take now, what is the realistic time saving from the AI intervention, and what is the cost of implementing and maintaining the change?
A practical ROI calculation framework
| Step | What to measure | How to get the number |
|---|---|---|
| 1. Baseline time | How long the workflow takes per instance, per week, and per year. | Ask the person who does the task to time themselves on three to five real examples. Use the average, not the best case. |
| 2. Frequency | How many times the workflow runs per week or month. | Count from a log, inbox, or system record rather than estimating. |
| 3. Labour cost | The hourly cost of the person performing the task, including employment overhead. | Use the fully loaded cost, not the headline salary. |
| 4. Annual time cost | Baseline time × frequency × 52 weeks × hourly cost. | This is the current cost of the workflow. |
| 5. Realistic saving | The proportion of the workflow time that AI can genuinely replace or assist. | Test the AI on real examples and measure the actual time with the tool, not the vendor's claim. |
| 6. Implementation cost | The cost of setup, integration, testing, training, and ongoing maintenance. | Include consultant time, platform subscription, and internal time to manage the change. |
| 7. Payback period | Implementation cost ÷ annual time saving (in cost terms). | If the payback period is under twelve months, the case is usually straightforward. |
What a realistic saving looks like
Drafting and writing tasks
AI can reduce the time to produce a first draft, but review and editing time remains. A realistic saving is 30–60% of the drafting time, not 100%.
Classification and routing
For high-volume, well-defined classification tasks, AI can handle a large proportion of cases. Exceptions still need a human. Measure the exception rate before claiming a saving.
Research and summarisation
AI can compress the time to produce a summary from a defined source. The saving depends on how much of the original research still needs human judgement.
Data entry and re-keying
Automation can eliminate manual re-entry between systems. The saving is close to 100% of the re-keying time, but integration setup and exception handling have their own costs.
Costs that are often underestimated
- The time to build, test, and refine the automation before it is reliable.
- The ongoing cost of maintaining the automation when connected systems change.
- The time to train staff on the new process and manage the transition.
- The cost of reviewing and correcting AI outputs that are wrong or incomplete.
- The platform subscription cost at the volume the business actually runs.
When the ROI calculation changes the decision
A careful ROI calculation sometimes reveals that the expected saving is smaller than the implementation cost, or that the payback period is longer than the business's planning horizon. That is a useful result — it prevents a poor investment.
It can also reveal that a simpler change, such as a template, a checklist, or a process redesign, would produce most of the saving without the complexity of an AI implementation. The calculation is a tool for making a better decision, not a justification for a decision already made.
From my workflow
I ask clients to time themselves on three real examples of the task before we discuss tools. The actual time is almost always different from the estimate — usually longer for complex tasks and shorter for simple ones. The measurement changes the conversation.
The most common mistake in AI ROI calculations is using the vendor's claimed time saving rather than a measured one. I always run a small test on representative examples before putting a number in a business case.
Frequently asked questions
How do you calculate the ROI of AI automation?
Measure the current time cost of the workflow, estimate the realistic time saving from the AI intervention based on a real test, subtract the implementation and maintenance cost, and calculate the payback period.
What is a realistic time saving from AI automation?
It depends on the task. Drafting and writing tasks typically save 30–60% of the drafting time, not 100%. Classification and routing tasks can save more for well-defined cases. Always measure on real examples rather than using vendor claims.
What costs are often missed in AI ROI calculations?
Setup and integration time, ongoing maintenance when connected systems change, staff training, the cost of reviewing incorrect outputs, and the platform subscription cost at actual volume.
Sources and further reading
Yiannis — Freelance SEO & AI Consultant, London
Semantic SEO Specialist · Technical SEO Auditor · AI-Assisted Search Strategist
I help London and UK businesses turn search intent into a clear, prioritised SEO programme. About me →
Related guides
Build a realistic AI business case
An AI Tools Assessment maps the workflow, measures the baseline, and produces an honest ROI estimate before you commit to a build.
See the AI Tools Assessment