Skip the hype.
Most AI vendors promise a big return but can't show you the maths. This is the formula I use, with three worked examples you can adapt to your own numbers.
Before you commit to AI automation, you should know what you'll get back. Not a vague promise of efficiency gains: a number you worked out yourself.
This guide gives you the formula and walks through each variable with three worked examples. By the end, you'll be able to run the numbers on any AI project yourself.
The Simple ROI Formula
AI Automation ROI Formula
ROI = (Time Saved + Revenue Gained + Cost Avoided - Total Investment) / Total Investment x 100
Time Saved
Annual value of labour hours eliminated or redirected to higher-value work
Revenue Gained
Additional revenue from faster response, better conversion, increased capacity
Cost Avoided
Money saved from fewer errors, reduced rework, eliminated penalties
Total Investment
Development + API/hosting + maintenance + training costs for year one
Four variables. The key is measuring each one honestly, with no inflated projections and no hidden costs. Each one is covered below.
Step 1: Calculate the Value of Time Saved
This is where most AI ROI lives. The process is straightforward:
- 1Identify the task. What specific process will AI handle? Be precise. Not ‘customer service’ but ‘answering tier-1 support tickets about order status, shipping and returns’.
- 2Count the hours. How many hours per week does your team spend on this task? Track it for two weeks if you don't know. Most people underestimate it.
- 3Use the fully loaded hourly cost. Salary is only part of it. Include benefits, office and equipment costs and management overhead. Call this figure R.
- 4Multiply. Hours per week x R x 52 weeks = annual value of time saved.
Example Calculation
3 people each spend 10 hours per week on manual data entry. Their fully loaded hourly cost is R.
3 people x 10 hours/week x 52 weeks = 1,560 hours x R a year
Even if AI handles only 70% of this work, that's still 1,092 hours a year back.
Step 2: Calculate the Cost of Errors You'll Eliminate
Manual processes create errors. Errors cost money. Sometimes obvious money (refunds, rework). Sometimes hidden money (customer churn, compliance risk). Calculate both.
Ask these questions:
- What is the current error rate? If you don't know, audit 100 recent outputs.
- What does each error cost to fix? Include rework time, customer communication, refunds, and any downstream impact.
- What are the indirect costs? Customer churn from repeated errors, compliance fines, reputation damage.
Example Calculation
Manual invoice processing has a 3% error rate. Your team processes 1,000 invoices per month. Each error takes 2 hours to investigate and fix, plus direct costs such as refunds, credits or late fees. Call those E.
1,000 invoices x 3% error rate = 30 errors/month
30 errors x (2 hrs x R + E) each month
Annual error cost: 360 x (2R + E)
If automated validation brings the error rate down to 0.5%, that's 25 fewer errors a month, each worth 2R + E.
Step 3: Calculate the Revenue Impact
This is the variable most people overlook. AI can also bring in revenue, in three ways:
Faster response times convert more leads.
The Lead Response Management Study, run with Professor Oldroyd of MIT, tested reply speed. The odds of qualifying a lead fell 21-fold when the first reply took 30 minutes instead of 5. If you reply in 4 hours today and AI gets that down to minutes, more leads should reach a conversation. Measure it on your own pipeline.
Better lead qualification increases close rates.
AI agents can score, qualify, and route leads in real time using the criteria your best sales reps use, applied consistently to every lead. Nothing gets cherry-picked or dropped, so your sales team spends its time on leads that are ready to buy.
Freed-up capacity means more output.
When your team isn't buried in repetitive tasks, they can take on more clients, process more orders, or focus on upselling existing accounts. Same headcount, higher revenue.
Example Calculation
A services company gets 200 inbound leads a month and replies in 4 hours, with 8% becoming a meeting. Suppose an AI reply agent lifts that to 9.7%. Treat the lift as something to test, not a promise.
Before: 200 leads x 8% = 16 meetings/month
After: 200 leads x 9.7% = 19.4 meetings/month
Additional meetings: 3.4/month x your close rate x your average deal
Additional annual revenue: 3.4 x 12 x close rate x deal value
Step 4: Calculate Total Investment (Honestly)
This is where you need to be brutally honest. Underestimating costs makes your ROI look artificially good, which leads to disappointment later. Include everything:
Development Cost
Design, build, test, deploy. Get a fixed quote, not an hourly estimate. Mine is a fixed price, quoted after a short scoping chat with me.
API and Hosting Costs
LLM API calls, cloud hosting, database, storage. Calculate based on expected volume.
Annual Maintenance
Bug fixes, model updates, prompt tuning, scaling. Budget for it every year.
Training and Onboarding
Team training, documentation, change management. Often overlooked but critical for adoption.
Year 1 Total = Development + (Monthly API/hosting x 12) + Annual Maintenance + Training. This is the denominator in your ROI formula. Get it wrong and the whole calculation falls apart.
Worked Example 1: Customer Support Chatbot
Customer Support Chatbot
An online shop with 3,000 support tickets a month
Investment
- BuildFixed price, agreed up front
- Model API and hostingMonthly, by volume
- MaintenanceAnnual
- Year 1 TotalAll of the above
Value Generated
- Assumed ticket deflection60%
- Tickets the AI handles1,800 a month
- Agent time per ticket (assumed)About 11 min
- Annual Value330 hrs x 12 x R
Payback period
Cost / value
Year-one cost divided by the monthly value, 330 x R
Worked Example 2: Document Processing Agent
Document Processing Agent
A financial firm with 50 complex documents a week
Investment
- BuildFixed price, agreed up front
- Model API and hostingMonthly, by volume
- MaintenanceAnnual
- Year 1 TotalAll of the above
Value Generated
- Processing time reduction4 hrs to 15 min
- Documents per week50
- Analyst time saved (weekly)187.5 hrs
- Annual Value187.5 hrs x 48 weeks x R
Say each document takes 4 hours of analyst review: extracting key data, cross-referencing compliance requirements and writing a summary. Suppose an AI agent cuts that to 15 minutes of human review. Then 50 documents x 3.75 hours saved x 48 working weeks comes to 9,000 analyst hours a year. Discount that for documents that still need full manual handling before you rely on it.
Payback period
Cost / value
Year-one cost divided by 9,000 x R / 12
Worked Example 3: Sales Lead Qualification
Sales Lead Qualification Agent
A B2B services company with 400 inbound leads a month
Investment
- BuildFixed price, agreed up front
- Model API and hostingMonthly, by volume
- MaintenanceAnnual
- Year 1 TotalAll of the above
Value Generated
- Leads scored automaticallyAll 400 a month
- Close rate changeMeasure it
- Additional revenueExtra closes x deal value
- Annual ValueHours saved x R, plus extra revenue
Say the sales team reviews every inbound lead by hand and spends 20 hours a week on leads that never convert. An AI qualification agent scores each lead on company size, intent signals, budget and fit. The team then spends its time on the best-matched prospects. The value is the hours saved, plus any lift in close rate you measure.
Payback period
Cost / value
Year-one cost divided by the monthly value you measure
The Payback Period: When You Break Even
ROI tells you how much you'll make. Payback period tells you how fast you'll make it.
Payback Period Formula
Payback Period (months) = Total Investment / Monthly Value Generated
Support Chatbot
Cost / value
Year-one cost / (330 x R)
Document Agent
Cost / value
Year-one cost / (9,000 x R / 12)
Lead Qualification
Cost / value
Year-one cost / monthly value
Run it on your own numbers. If the payback is longer than you can accept, shrink the scope first.
Red Flags: When AI Automation ROI Won't Work
Not every process should be automated. Here are the warning signs that an AI project will deliver poor ROI:
The process is too complex to automate reliably.
If the task requires deep judgment, nuanced context, or changes unpredictably every time, AI will struggle. Look for processes with clear rules, repeatable patterns, and structured inputs. If your best employee can't explain the decision-making process step by step, AI can't learn it either.
Your data quality is too poor.
AI is only as good as the data it works with. If your CRM is full of duplicates, your documents are inconsistently formatted, or your data lives in 15 different spreadsheets, you'll need to fix that first. Garbage in, garbage out: no amount of AI changes this.
Your team won't adopt it.
The best AI system in the world delivers zero ROI if nobody uses it. If your team is resistant to change, doesn't trust the outputs, or finds workarounds to avoid using the tool, your investment is wasted. Budget for training and change management, or don't bother building.
The volume is too low to justify the cost.
If you process 10 invoices a month, automating invoice processing won't pay for itself. AI automation ROI depends on volume. The more times a task repeats, the faster the payback. As a rule of thumb: if the task only takes a few hours a week across the whole team, a custom AI solution rarely earns back its build cost. Look for off-the-shelf tools instead.
Your Next Step: The Quick ROI Check
Before you talk to any vendor, including me, run this quick calculation on your most promising automation candidate:
30-Second ROI Check
Decision rule: If ROI is positive and payback period (D / (C / 12)) is under 6 months, the project is a strong candidate. Move forward.
This leaves out error reduction and revenue, which are upside. If the time savings alone justify the investment, the rest is a bonus.
Want to Run the Numbers With Me?
Tell me the workflow you want to automate. Fixed price, quoted after a short scoping chat with me.
Talk to meFrequently Asked Questions
How do you calculate the ROI of AI automation?+
What is the typical payback period for an AI automation project?+
What is the average ROI for AI automation in business?+
When does AI automation NOT make financial sense?+
Scope your custom AI project in 30 seconds
Three questions show the shape of your build and a realistic timeline, based on real shipped projects. After a short scoping chat with me you have a written fixed-price quote, so you know the full cost before anything starts.
Or build your own AI system piece by piece and send the design in for a written quote →