AI in practice · 6 min

How to calculate ROI on AI automation before you commit

Published
Updated
Reading time
6 min

Want the numbers without reading the whole framework? Use the free AI Automation ROI Calculator: payback, 3-year ROI, and net savings in 60 seconds, including oversight overhead and ongoing API costs.

The problem with most AI ROI estimates

Most AI ROI calculations are either absurdly optimistic ("we'll save 10,000 hours!") or so vague they convince nobody. The result: promising automation projects die because nobody built a credible business case, or bad projects get funded because the numbers were fabricated.

A good ROI estimate does three things: it quantifies the current cost of the workflow, estimates the realistic cost of automating it, and accounts for the things that are easy to forget (maintenance, error handling, adoption friction).

Step 1: Measure the current cost of the workflow

Before you can calculate what AI saves, you need to know what the manual process actually costs. The formula is straightforward:

Monthly workflow cost = Time per task x Tasks per month x Fully loaded hourly rate

"Fully loaded" means salary plus benefits, equipment, management overhead, and office costs. A common rule of thumb: multiply gross salary by 1.3 to 1.5 to get the fully loaded rate. For a team member earning €3,500/month gross, that is roughly €25 to €30/hour fully loaded.

Example (invoice processing):

  • Time per invoice: 5 minutes (0.083 hours)
  • Invoices per month: 200
  • Hourly rate: €28 fully loaded
  • Monthly cost: 0.083 x 200 x €28 = €465/month

Do not forget the hidden costs that don't show up in the time-per-task number: context switching (the person handling invoices is probably doing other work too), error correction (manual data entry has a 1 to 3% error rate), and delayed processing (invoices sitting in a queue affect cash flow visibility).

A conservative estimate: add 20 to 30% to your calculated cost to account for these.

Step 2: Estimate the AI automation cost

AI automation costs fall into four buckets:

Build cost (one-time): Engineering time to build the pipeline, integrate with your systems, and test. For a straightforward workflow like invoice processing or ticket triage, expect 2 to 4 weeks of senior engineering time. For something more complex like a RAG knowledge base, expect 4 to 8 weeks.

API and compute cost (ongoing): The cost of running AI inference. This varies wildly by model and volume. Current ballpark for common tasks:

  • Document extraction: €0.02 to €0.15 per document
  • Text classification (ticket triage): €0.001 to €0.01 per item
  • RAG query: €0.01 to €0.05 per query
  • Lead enrichment: €0.10 to €0.50 per lead (includes data source costs)

Infrastructure cost (ongoing): Vector databases, hosting, monitoring. For most small-to-medium automation projects: €50 to €300/month.

Maintenance cost (ongoing): Prompt tuning, handling edge cases, updating when your processes change. Budget 2 to 4 hours/month of engineering time for a stable automation. More in the first 3 months as you tune.

Example (invoice processing automation):

  • Build: 3 weeks of engineering at €800/day = €12,000
  • API: 200 invoices x €0.05 = €10/month
  • Infrastructure: €80/month
  • Maintenance: 3 hours/month x €60/hour = €180/month
  • Total monthly run cost: ~€270/month

Step 3: Calculate the realistic payback

Monthly savings = Current workflow cost - AI run cost

Payback period = Build cost / Monthly savings

Using the invoice example:

  • Monthly savings: €465 + 25% hidden costs (€581) - €270 = €311/month
  • Payback: €12,000 / €311 = ~39 months

That is honest math. For 200 invoices/month, the ROI is real but slow. The calculus shifts dramatically at higher volume. At 1,000 invoices/month, the manual cost is €2,900/month while the AI cost barely moves (API costs go to €50/month), making the payback period under 5 months.

This is the critical insight: AI automation ROI scales with volume. Low-volume workflows may not justify custom automation. High-volume workflows almost always do.

Step 4: Account for what the spreadsheet misses

The quantifiable ROI is only part of the picture. Three factors that are harder to measure but often tip the decision:

Speed. AI processes things in seconds, not minutes or hours. For support ticket triage, this means faster first-response times. For lead qualification, it means contacting prospects while they are still warm. Speed advantages compound. They affect customer satisfaction, conversion rates, and team morale.

Consistency. Humans have variable attention. The 200th invoice of the month gets less careful review than the first. AI applies the same logic every time. In compliance-sensitive domains (healthcare, finance, legal), consistency has regulatory value that is difficult to quantify but very real.

Scalability. Manual workflows hit a ceiling. You need to hire more people to process more volume. AI workflows scale with minimal marginal cost. If your business is growing, the avoided hiring cost is often larger than the direct time savings.

Common mistakes that kill the business case

Comparing AI to zero cost, not to the current process. The question is not "is AI expensive?" but "is AI cheaper than what we do now?" A €270/month automation cost sounds like a lot until you compare it to €580/month in manual processing.

Ignoring the adoption curve. Automation that nobody uses has zero ROI. Budget time and effort for change management: training, feedback collection, iterating on the interface. The first month of deployment is about learning, not savings.

Assuming 100% automation. Almost no workflow should be fully automated from day one. Plan for 80 to 90% automation with human review on the rest. Your ROI model should reflect this. The human is still spending time, just less of it.

Forgetting maintenance. AI systems need ongoing tuning. Models change, your processes evolve, edge cases appear. A system that works perfectly in month one and is never touched again will degrade. Budget for it.

When to skip the ROI calculation entirely

Sometimes the value of automation is not about cost savings at all. If you are automating because:

  • The current process is a bottleneck blocking growth (you literally cannot process more volume with current headcount)
  • Errors in the current process carry outsized risk (wrong data in a compliance report, missed urgent ticket)
  • The manual work is destroying team morale and causing turnover

...then the ROI case is strategic, not financial. Frame it that way. Not "this saves €X" but "this removes the constraint that prevents us from growing" or "this eliminates the risk that could cost us €Y in a single incident."

What to do with this framework

If you are evaluating a specific workflow, run the numbers with the formula above. Be conservative on savings, generous on costs, and honest about volume. If the payback period is under 12 months, it is a strong case. Under 6 months, it is almost certainly worth doing.

The fastest way to apply this framework is our free AI Automation ROI Calculator. It runs the formula above with the things vendor calculators leave out: oversight overhead, ongoing API costs, and best/worst-case sensitivity.

If you want someone to run these numbers on your real workflows and then build the automation, that is a fixed-price project, from €5,000. We start by checking whether it pays, and we tell you if it does not.

For a broader view of which workflows are good automation candidates, we cover five common ones in detail.

Want this in your business?

Fixed-scope, fixed-price AI projects, from €5,000. We work out first whether it pays, then make it work with real users.