How to calculate the ROI of an AI project before you invest

Every executive has heard the promise: AI will save time and money. What few vendors show you is the math. Before you commit $15,000, $50,000, or more, you need to know what you expect to gain, how fast, and which assumptions hold that number up. The good news is that the math is simple and fits in a spreadsheet.

The basic formula

Return on investment compares gain with cost:

ROI = (gain over the period – cost over the period) / cost over the period

Also calculate payback: how many months until cumulative gains cover the investment. For small and mid-sized companies payback is often more useful than a percentage, because cash is tight and the real question is when the money comes back.

Step 1: establish what the process costs today

Pick one process and calculate its current cost. Include:

• Monthly hours spent by everyone involved.
• Hourly cost: salary plus payroll taxes and benefits, divided by hours worked. A $55,000 salary can easily reach $70,000 fully loaded, or roughly $35 per hour.
• Cost of errors: rework, returns, penalties, discounts given for late delivery.
• Opportunity cost: sales lost because replies were slow, for example.

That total is your baseline. Without it no ROI figure can be trusted.

Step 2: estimate the gain conservatively

AI gains usually come from four sources:

1. Time freed, which only becomes money if it is redeployed to productive work or avoids a hire.
2. Fewer errors and less rework.
3. Added revenue, such as faster responses and more proposals sent.
4. Avoided costs, such as not needing another headcount as you grow.

Be skeptical. A common mistake is counting 100 percent of freed hours as savings. Apply a realization factor, typically 50 to 70 percent, because not every freed hour turns into value.

Step 3: add up every project cost

Cost goes beyond a license or a build fee. Include:

• Assessment, design, and implementation (one-time).
• Licenses, AI model usage, and infrastructure (monthly).
• Maintenance, monitoring, and continuous improvement.
• Internal team time to train, review, and adopt.
• Integration with existing systems and data cleanup.

Poorly budgeted projects almost always forget the last two lines.

A full hypothetical example

Hypothetical example: a wholesale distributor has 3 people spending a combined 90 hours a month keying in orders and matching invoices. Cost per hour: $35. Current monthly labor cost: $3,150, plus about $1,500 in errors and rework. Total: $4,650 a month.
Automation cuts effort by 70 percent, but we apply a 60 percent realization factor: gain of roughly $2,800 a month.
Costs: $28,000 to implement and $900 a month for licenses and maintenance. Net monthly gain: $1,900. Payback: 28,000 / 1,900 = about 15 months.

That number is honest, and it reveals something important: a process that is too small does not justify an expensive project. If the same project targeted a process costing $14,000 a month, payback would fall to a few months. ROI depends more on choosing the right target than on choosing the technology.

Step 4: run scenarios and test sensitivity

Build three scenarios, pessimistic, realistic, and optimistic, varying the realization factor and the implementation cost. If the project only pays off in the optimistic case, it is risky. If it pays off even in the pessimistic one, it is a strong candidate.

Step 5: reduce risk with a pilot

Instead of betting the whole budget, fund a 30 to 60 day pilot with a clear target. A pilot turns estimates into real data and lets you scale or stop with far less money at stake. Define success criteria up front, for example: cut data-entry time by 50 percent with an error rate under 2 percent.

Benefits the spreadsheet will not capture

Not everything fits the math: less dependence on key people, more trustworthy data, faster decisions, a less burned-out team. List these as qualitative bonuses, but never use them to justify a project that fails on the numbers.

Summary

• Measure the current cost of the process.
• Estimate gains with a conservative realization factor.
• Add all costs, including internal time.
• Calculate payback and run three scenarios.
• Validate with a pilot before scaling.

A note on timing

Revisit the model every quarter. Costs of AI usage tend to fall while your volumes grow, so a project that looks marginal today may look very different in a year. Keep the spreadsheet alive: update the baseline, compare forecast with actuals, and use the gap to sharpen the next estimate. Teams that do this get better at picking winners, and they stop funding projects based on enthusiasm alone.

Want to put this to work? Convertty’s senior team runs an AI assessment of your workflows and delivers an implementation plan with 30-day quick wins. Book a call.