The most important calculation you can do before implementing AI is the return on investment. What will this cost? What will you save or gain? Is it worth doing? These questions need real answers, not hunches. The good news is that calculating AI ROI is straightforward. It requires three components: understanding the time saved, understanding the costs, and doing basic arithmetic. Let's walk through it.

The Three Components of AI ROI

First component: direct savings from time reduction. How much time does the current process take? How much time will it take with AI? The difference is the time saved. Multiply that by the hourly cost of the people doing the work. That's your direct savings.

Second component: indirect savings. This includes reduced errors, lower material costs, improved asset utilisation, or anything else that costs money today but would improve with AI. These are harder to quantify but often substantial.

Third component: revenue potential. Does AI improve something that helps you sell more or serve customers better? Sometimes the biggest opportunity isn't cost reduction, it's revenue improvement.

Against these benefits, you calculate implementation cost and ongoing costs. Then you have your ROI calculation.

Component One: Direct Savings From Time Reduction

Start with the current state. How much time does this process take today? Get real data. Not estimates, which are usually optimistic. Real time-tracking data. If you don't have it, track it for a week or two. Have the people doing the work log their time, or have a supervisor observe, or use time-tracking software. You need a number you can trust.

Let's say you're looking at customer onboarding. A support team member currently spends 45 minutes per customer setup. You handle 100 customers per month. That's 75 hours per month, or about 900 hours per year. At an average fully-loaded labour cost of 30 per hour (salary plus benefits plus overhead), that's 27,000 per year.

Now estimate the time with AI. With an AI-assisted system, the same task takes 15 minutes per customer because the AI handles document collection, data entry, and initial verification. The human does a 5-minute review to verify everything is correct. So instead of 45 minutes, it takes 20 minutes. That's 33 hours per month, or about 400 hours per year instead of 900.

The time saved is 500 hours per year. At 30 per hour, that's 15,000 per year in direct labour savings.

This is the basic calculation. It's simple enough that you can do it with a spreadsheet. The key is getting real numbers for current time and realistic estimates for AI-assisted time. Most people underestimate how much human oversight is still needed, so build in a realistic percentage of reduction, not an optimistic one.

Component Two: Indirect Savings

Indirect savings are real but harder to quantify. Let's use the same onboarding example. Currently, when an error happens in the setup data, it causes problems downstream. A customer's address is wrong, so their first shipment goes to the wrong place. A contact is wrong, so you can't reach them. These errors cost time and money to fix.

AI might reduce errors by 60 percent because the AI is consistent and doesn't have the attention lapses that humans do. Maybe you currently have 5 percent of customer setups with significant errors, requiring an average of 30 minutes to fix. That's about 5 hours per month of rework. If AI reduces errors by 60 percent, you save 3 hours per month of rework, or about 36 hours per year. At 30 per hour, that's another 1,080 per year.

But error costs extend beyond the immediate rework. Wrong data also harms customer experience. If your customer's first interaction is delayed because of bad setup data, you lose trust. You might lose repeat business, referrals, or positive reviews. Quantifying this is speculative, but if you can estimate the cost of a lost customer relationship, even if it's rare, you can estimate what error reduction is worth to you.

In this onboarding example, indirect savings might add 1,000 to 3,000 per year depending on how much error reduction impacts your business. Conservative estimate for our example: 1,500 per year.

Component Three: Revenue Potential

Sometimes AI's biggest value isn't cost reduction, it's revenue improvement. In the onboarding example, faster setup time means customers get access to your service sooner. Some customers might buy additional products sooner because the onboarding was smooth and quick. Some prospects who would have abandoned a slow onboarding process might complete it. Some customers might give referrals because their experience was better.

Quantifying revenue improvement is speculative but essential. You're not imagining benefits. You're estimating them conservatively. In the onboarding example, if AI-assisted setup is 60 percent faster, and you estimate that a 5 percent improvement in customer completion rate would add 15 new customers per year, and each customer has a lifetime value of 1,000, then that's 15,000 per year in additional revenue.

Or maybe it's more modest. Maybe faster setup improves customer satisfaction, leading to 2 percent higher repeat purchase rate, which on your customer base adds 8,000 per year in additional revenue. The point is to estimate conservatively based on realistic assumptions.

In our onboarding example, let's estimate conservative revenue impact at 10,000 per year.

The Full Formula

Using our onboarding example: Direct savings of 15,000, indirect savings of 1,500, revenue impact of 10,000 equals 26,500 in annual benefits. Let's say implementation cost is 8,000 (buying an AI tool, some consulting, training). Ongoing costs are 500 per month for the tool license, so 6,000 per year.

Year 1 total cost is 8,000 plus 6,000 equals 14,000. Net benefit in year 1 is 26,500 minus 14,000 equals 12,500. ROI percentage is (12,500 divided by 14,000) times 100 equals 89 percent ROI in year 1. Payback period is 14,000 divided by 26,500 equals 0.53 years, or about 6 months.

From year 2 forward, there's no implementation cost, only ongoing costs. Annual benefit stays around 26,500 (probably higher as people get better at using the system), ongoing cost stays at 6,000, so net annual benefit is about 20,500 per year.

How to Use This Framework on Your Processes

The formula is the same for every process. Identify the current time cost. Estimate the AI-assisted time cost. Calculate time saved and multiply by hourly rate. Identify any indirect cost savings. Estimate any revenue impact. Add implementation and ongoing costs. Run the formula.

Do this for each significant process you're considering. Rank them by ROI or by payback period, depending on what matters more to you. If payback is important (you need to recover investment quickly), focus on processes with fast payback. If absolute ROI matters (you want the biggest return over time), focus on highest-percentage returns. Many organisations do both and find the sweet spot where they improve high-ROI processes while building capability for longer-term wins.

Common Mistakes in ROI Calculation

First mistake: overestimating time savings. People think AI will do the work 80 percent faster. It usually does 40 to 60 percent faster because human verification still takes meaningful time. Use conservative estimates.

Second mistake: forgetting ongoing costs. The AI tool costs money every month. There are costs to maintain it, update it, handle edge cases. These ongoing costs reduce your benefit year after year. Don't just calculate year 1 benefit and call it done. Calculate year 2, 3, and beyond to see if the investment still makes sense over time.

Third mistake: ignoring implementation challenges. You estimate the tool will take 1,000 to set up. Actually, integrating it with your systems takes longer. Training takes longer. The first few months have problems you didn't anticipate. Build contingency into implementation cost, say 20 to 30 percent extra.

Fourth mistake: not accounting for change management. People don't like change. The AI tool might save time on paper, but people resist using it. You need to budget time and money for training, adjustment period, and managing resistance. This is real cost and it's often underestimated.

When ROI is Negative (and When That's Still Worth Doing)

Not every AI investment has positive ROI in year 1. Sometimes you implement something that costs more than it saves because you're building capability, learning, or positioning for future opportunity. That's fine as long as you're doing it intentionally, not by accident.

Calculate the ROI honestly. If it's negative in year 1, calculate years 2 and 3. Does it turn positive? How long until it pays for itself? Is that timeline acceptable to you? Some investments take 18 to 24 months to pay back. That's reasonable if the long-term benefit is substantial.

But if you're looking at a project that never pays back, that never generates sufficient benefit to justify the cost, then it's not worth doing. Just be honest about the calculation so you can make an informed decision.