Blog · October 2026

How to Measure ROI on AI Automation (With Real Numbers)

A framework that survives a CFO conversation.

Most automation projects fail to prove their value — not because they don't deliver, but because nobody measured the "before". By the time you want to show ROI, the baseline is a memory.

This framework fixes that. Measure these five things consistently, and you'll always be able to defend the investment.

The five metrics that matter

1. Hours saved per week

The most direct metric. Track the time spent on the automated workflow before and after. Average across 3–4 weeks to avoid one-off weeks.

2. Error rate reduction

Count mistakes per 100 transactions before and after. Errors cost time (fixing) and money (wrong invoices, missed follow-ups).

3. Response time improvement

For lead response, customer support, or quote turnaround — measure the average time to first response. This has direct revenue implications.

4. Conversion rate change

For lead-nurturing or sales automations, measure the percentage of leads that become customers, before and after.

5. Cost per transaction

Total cost (people + tools) ÷ number of transactions. This is what actually improves when automation works.

The basic ROI formula

Automation ROI is straightforward:

ROI = (Value created – Cost) ÷ Cost × 100

Where "Value created" is:

A worked example

Take a professional services business with a small sales team. They automate lead capture and follow-up:

12hHours saved per week across team
R250Loaded hourly cost
R156kAnnual time value saved

Time value: 12h × R250 × 52 weeks = R156,000 per year

Revenue impact: response time drops from 8 hours to 15 minutes. Based on published conversion data (5-minute response = 21× more likely to close), assume a conservative 15% improvement in close rate. If the team closes R800,000 per year in new business, that's R120,000 in extra revenue.

Error reduction: manual entry errors cause roughly R15,000 per year in rework. Automation cuts this by 80%, saving R12,000.

Total value: R156,000 + R120,000 + R12,000 = R288,000 per year

Cost: R45,000 for implementation + R18,000 per year in tool subscriptions = R63,000 in year one

ROI: (288,000 – 63,000) ÷ 63,000 × 100 = 357%, or roughly 4.6× return in year one.

The 3–5× year-one return is common for well-scoped automations. Anything below 2× usually means the wrong workflow was automated first.

Intangible benefits

Worth mentioning in business cases, even if harder to quantify:

Don't lead with these — lead with hard numbers and use these as tailwinds.

Calculating payback period

Payback period = Implementation cost ÷ monthly value created

Using the example above:

Anything under 6 months is a strong project. 6–12 months is reasonable. Over 12 months, the scope is usually too ambitious or the wrong workflow was chosen.

Setting up tracking

Before the project starts:

  1. Write down the "before" numbers. Pick one week and track every relevant metric manually.
  2. Define success criteria in writing. "Save X hours per week by doing Y" beats "improve efficiency".
  3. Set a review date. Six weeks post-launch is usually the right checkpoint.
  4. Agree on how the numbers will be measured. If you're tracking hours saved, both sides need to agree how that's counted.

After the project ends, review against those criteria — don't move the goalposts to make the numbers look better. If the ROI isn't there, that's information, not failure.

Key takeaways

  • Five metrics: hours saved, error rate, response time, conversion, cost per transaction
  • ROI = (Value – Cost) ÷ Cost × 100 — no mystery
  • 3–5× year-one return is a reasonable target for well-scoped work
  • Payback under 6 months is the real signal of a good project
  • Measure the "before" — you can't reconstruct it later

Want help scoping a project with real numbers?

We model ROI on the front end, so you know the number before you commit.

Book a scoping call →