“Can AI save us money?” is a good question, but it needs a number to become a decision. Many AI projects are approved on enthusiasm and cancelled when nobody can show the result. Others never start because nobody made the business case.
This guide gives you a simple way to calculate the return on investment (ROI) of AI and automation, with a worked example.
The basic formula
ROI = (Annual benefit - Annual cost) / Annual cost x 100%
And the payback period:
Payback period (months) = Initial investment / Monthly net benefit
The hard part is not the formula. It is estimating benefits and costs honestly.
Step 1: Measure the current process
Pick one process and measure:
- Volume: how many times per month does it happen?
- Time per case: how many minutes does it take?
- Who does it: what is the fully loaded hourly cost of that role (salary plus employer costs and overhead)?
- Error rate: how often does something go wrong, and what does an error cost?
- Delay cost: does waiting cost money (lost sales, late payments, unhappy customers)?
Measure, do not guess. A week of simple time tracking is usually enough.
Step 2: Estimate the benefit
AI and automation create value in four ways:
- Time saved: hours freed for other work
- Errors avoided: fewer mistakes, returns, corrections
- Speed: faster response or processing, which can increase sales or customer satisfaction
- Scale: handling more volume without hiring
Be conservative. If a pilot shows 60% time savings, use 40-50% in the business case to cover review time and exceptions. Research on real tasks shows wide variation, see how much time AI really saves.
Step 3: List all costs
People forget these:
| Cost type | Examples |
|---|---|
| One-time | Analysis, development, integration, data cleanup, testing, training |
| Recurring | Software licences, AI API usage, hosting, monitoring |
| Human | Review time, exception handling, process ownership |
| Change | Training sessions, documentation, lost productivity during rollout |
| Maintenance | Updates when connected systems change, prompt and model updates |
Worked example: processing supplier invoices
The numbers below are a hypothetical example to show the method. Use your own figures.
Current situation:
- 800 invoices per month
- 6 minutes per invoice (opening, checking, typing into accounting software)
- Fully loaded cost: €40 per hour
- Monthly cost: 800 x 6 / 60 x €40 = €3,200
With AI document extraction and automation:
- AI extracts data, automation checks against purchase orders, a person reviews exceptions
- Estimated time per invoice: 1.5 minutes (review and exceptions)
- Monthly labour cost: 800 x 1.5 / 60 x €40 = €800
- Monthly saving on labour: €2,400
Costs:
- One-time setup: €12,000
- Recurring (software, API usage, maintenance): €400 per month
Result:
- Monthly net benefit: €2,400 – €400 = €2,000
- Payback period: €12,000 / €2,000 = 6 months
- First-year ROI: (€24,000 – €4,800 – €12,000) / (€12,000 + €4,800) ≈ 43%
- From year two (no setup cost): (€24,000 – €4,800) / €4,800 = 400%
This does not yet include fewer errors or faster payment processing.
Step 4: Add the benefits that are hard to measure
Some benefits are real but harder to put in euros:
- Employees doing more interesting work (retention)
- Faster customer responses (satisfaction and reviews)
- Better data quality for decisions
- Ability to grow without hiring at the same pace
Mention them in the business case, but do not rely on them to justify the investment.
Step 5: Start with a pilot
Do not build the full solution first. Run a pilot of a few weeks on a part of the volume, measure the real numbers and update the business case. This reduces risk and gives management real data.
Which processes have the best ROI?
The best candidates are:
- High volume: happens hundreds or thousands of times per month
- Rule-based or pattern-based: clear structure, even if the input is messy
- Time-consuming per case
- Easy to check: errors are visible quickly
Examples: invoice processing, support ticket classification, order status questions, CRM data entry, product data enrichment, report generation.
Poor candidates: rare, high-risk decisions that require deep judgement every time.
Common mistakes
- Counting saved hours that are not really freed. Ten minutes saved per day per person is real, but it only becomes value if it is used well
- Ignoring review time
- Forgetting maintenance
- Comparing against a perfect process instead of the real one
- No baseline: without measuring before, you cannot prove the result after
Frequently asked questions
What is a good ROI for an AI project?
It depends on your company and risk. Many automation projects aim for payback within a year. Pilots help you check that before committing.
Should I count saved time if nobody is let go?
Yes, as capacity. Show what the freed time will be used for: more customers, faster growth, better service or avoided hiring.
How do I estimate API costs?
Run the pilot on real data and measure tokens per case, then multiply by volume. See how to reduce LLM API costs.
Need a business case for AI?
I help businesses select the right processes, calculate realistic ROI and run pilots that prove the value. See my AI integration service or request a consultation.