How to Calculate AI ROI for Your Business (Worked Example)

AI ROI is the value AI adds minus its total cost, divided by that total cost, over a fixed period, usually 12 months. Value is net hours saved multiplied by their hourly cost, plus any measured extra profit; cost includes subscriptions, setup, training and upkeep. So $6,000 of value on $2,000 of cost is a 200% return.

The input owners most often get wrong is hours saved. If AI cuts a customer reply from 10 minutes to 2, but checking and correcting the draft takes another 4, the real saving is 4 minutes, not 8, and the value you can claim halves before you've counted a single subscription.

Follow me on Instagram@sagnikteaches

The same method works before you build (a forecast, using figures from a small test) and after (actual results, using measured figures). Do both: forecast to decide, then recalculate at 90 days to confirm.

Connect on LinkedInSagnik Bhattacharya

The formula and what goes into it

ROI (%)          = (Total value - Total cost) / Total cost x 100
Payback (months) = One-off costs / (Monthly value - Monthly running costs)

Each input needs a clear definition, or the answer means nothing:

Subscribe on YouTube@codingliquids

• Net hours saved: time the job took before, minus the time it takes now including checking and correcting the AI's output.
• Hourly cost: the loaded cost of whoever did the work, meaning pay plus the other costs of employing them. Take it from payroll, not the hourly wage alone.
• Extra profit: measured additional sales, converted to margin. Revenue is the wrong number, because each extra sale also carries costs.
• One-off costs: setup time, any outside help, training, and the slower first few weeks.
• Running costs: subscriptions, automation-platform charges and the time spent keeping instructions and reference files up to date.

Don't count checking time twice. It's already inside "net hours saved", so it shouldn't appear again under running costs.

The salon and its baseline

The illustrative salon has six nail technicians, a receptionist and an owner who covers the desk at busy times. It chose two AI jobs: drafting replies to booking and price enquiries that arrive by message and email, and drafting replies to online reviews. Staff check and send everything.

Before switching anything on, the salon measured two ordinary weeks. The method for this is in setting a baseline before you introduce AI; without it, the rest of the calculation is guesswork.

The "after" figures come from weeks five to eight, once the team was past the learning period. The salon also tracked how many enquiries turned into bookings: 38% before, 42% after, which the owner put down to replies now going out within the hour instead of after closing.

Working out the value

Time value. 3.6 hours a week across 50 working weeks is 180 hours a year. The receptionist's loaded cost is $20 an hour, and the saved time now goes on rebooking calls and retail sales at the desk, so it counts in full: 180 × $20 = $3,600.

Extra profit. A four-point rise in conversion on 70 enquiries a week is 2.8 extra bookings a week. Some of that rise would probably have happened anyway, so the owner counts half: 1.4 bookings. The average booking is $45, and after product costs and technician commission, say $18 of that is margin. So 1.4 × $18 × 50 weeks = $1,260.

Total value in year one: $4,860.

Some businesses have a third, simpler kind of value: money they stop paying someone else. An illustrative print and signage shop paid a freelance copywriter about $300 a month for product-page text and short client proposals. After three months of drafting with an AI assistant and having the owner edit, it cut that to one $100 job a month for the pieces that still needed a specialist. That $200 a month, $2,400 a year, is a cash saving and counts in full, with no argument about where the time went. The owner's extra editing time, about three hours a month, goes on the cost side at her hourly value. Cash savings like this are the most convincing line in any ROI figure, so look for them first.

Working out the full cost

The seat price matches the list prices of ChatGPT Business and Claude Team (both $25 a seat a month billed monthly, minimum two seats), and the automation plan matches Zapier Professional billed annually. Notice how much of the cost is people's time: $1,330 of the $2,170. Leave that out and the year-one ROI jumps from 124% to nearly 480% on paper, which is exactly why so many AI returns look spectacular in a vendor's slide and ordinary in the accounts.

The result: ROI and payback

Year one ROI  = ($4,860 - $2,170) / $2,170 = 124%

One-off costs          = $350 + $360 + $110 + $90 = $910
Monthly value          = $4,860 / 12              = $405
Monthly running costs  = ($600 + $240 + $420) / 12 = $105
Payback                = $910 / ($405 - $105)      = about 3 months

Year two ROI (no one-off costs) = ($4,860 - $1,260) / $1,260 = 286%

A return of 124% in year one and a three-month payback is a sound result for a small business. Year two looks better still, because the one-off costs don't recur, though it assumes volumes and prices stay similar.

When the cost is close to nothing

The percentage breaks down when the AI comes inside a subscription you'd pay for anyway, such as Gemini in a Google Workspace business plan or Copilot Chat in Microsoft 365. Picture a four-person design agency using Gemini in its existing Workspace plan to draft client status emails. Its only cost is five hours of setup at $30, so $150, and it saves about 40 hours a year at $30, so $1,200. That's an ROI of 700%, which is arithmetically true and practically useless: halve the saving and it's still 300%, so the figure can't guide anything. In that situation report two numbers instead: net value in dollars ($1,050) and payback in weeks (about seven). Then spend your attention on whether the time saved is real, because that's the only thing that could make the answer wrong.

Benefits that are real but shouldn't go in the sum

Some gains are genuine but too soft to price honestly. The salon's owner no longer answers messages at 10pm. The receptionist says the desk feels calmer on Saturdays. Review replies are more consistent in tone. Write these down alongside the ROI figure, as a separate list, rather than inventing a dollar value for them. They matter when the numbers are close, and they're often what staff remember, but mixing them into the formula makes the whole calculation easier to dismiss.

The same applies to risks. If the AI occasionally drafts a wrong price that a person catches, that's a cost of checking, already counted. If one slipped through and a customer was charged the lower price, record the actual cost as a line in the next recalculation.

What if the estimates are wrong?

Every input above is an estimate or a short sample. A sensitivity check asks what the answer would be if the main assumptions were off. For AI, the two that matter most are hours saved and the revenue effect.

The pessimistic case loses money in year one (though it would turn positive in year two, at about 43%). That tells the owner something useful: the whole case rests on the time saving being real. So the sensible move is a checkpoint at the end of month two. If measured savings are below two hours a week by then, fix the setup or stop before the full year's costs are spent.

At 90 days, redo the sum with measured figures in place of the forecast. Say the salon's new timed sample shows 3.1 hours saved a week rather than 3.6, because review replies needed more checking than expected. Conversion held at 42%. And one AI-drafted reply quoted last year's gel manicure price, which the salon honoured, costing a $15 discount. The recalculation:

• Time value: 3.1 × 50 × $20 = $3,100, plus the same $1,260 of profit, so $4,360 of value.
• Cost: $2,170 plus the $15 discount, so $2,185.
• Year one ROI: ($4,360 − $2,185) / $2,185, about 100%.

Still a clear yes, but a smaller one, and the review-reply job is now the weakest part of it. Two levers are worth pulling at this point. Tighten the review-reply instructions and re-time them next quarter. And if the setup is staying, switching the two seats to annual billing ($20 a seat instead of $25) saves $120 a year, which on its own lifts the year-two figure.

Six errors that inflate AI ROI

1. Counting gross minutes. "The AI writes a reply in ten seconds" ignores the minute and a half someone spends reading and fixing it.
2. Counting time that wasn't reused. If saved minutes vanish into a quieter afternoon, they don't show up in the accounts. Say where the time went, or count it at a reduced rate.
3. Using revenue instead of margin. A $45 booking isn't $45 of value when products and commission come out of it.
4. Leaving out owner time. The owner's setup and upkeep hours are often the biggest single cost.
5. A guessed baseline. "It used to take about ten minutes" is nearly always an overestimate. Measure it.
6. Crediting AI for seasonal lift. Compare the same season, or a matching period, not a quiet month against a busy one.

Error 3 is the one that most distorts small-business figures, because it sounds like good news. An illustrative florist added AI-drafted replies to wedding enquiries and counted the extra orders at their full price: 8 extra wedding orders a year at $300 each, $2,400 of "value". But flowers, sundries and delivery take a big share of each order. If the florist's accounts show a margin of about $105 on a $300 order, the real extra profit is $840. Against a year-one cost of $1,500, the claimed ROI of 60% becomes a loss of 44%. The AI may still be worth keeping for the time it saves, but the extra-orders line alone doesn't justify it.

For more on measuring the time side accurately, see how to measure time saved after rolling out AI.

Which records already hold each ROI input

Most of the inputs already exist somewhere in a small business. Knowing where saves you from estimating:

Start the setup log before you do anything else. It's the input people most often have to reconstruct from memory, and memory reliably undercounts it.

A spreadsheet layout you can copy

Put labels in column A and values in column B. The formulas below work in Excel and Google Sheets.

Row  A (label)                              B (value or formula)
2    Baseline minutes per week              352
3    New minutes per week, incl. checking   136
4    Working weeks per year                 50
5    Loaded hourly cost ($)                 20
6    Time value per year                    =(B2-B3)/60*B4*B5
7    Extra bookings per week (measured)     2.8
8    Share of uplift counted                0.5
9    Margin per booking ($)                 18
10   Extra profit per year                  =B7*B8*B9*B4
11   Total value                            =B6+B10
12   Subscriptions and platforms per year   840
13   One-off costs                          910
14   Upkeep time per year ($)               420
15   Total cost, year one                   =B12+B13+B14
16   ROI, year one                          =(B11-B15)/B15
17   Payback (months)                       =B13/((B11-B12-B14)/12)

Format B16 as a percentage. To run the pessimistic case, copy the column and change B3 and B8.

Before you rely on the sheet, have a chat assistant act as a hostile reviewer. Paste in the labels and values (no names or payroll details) with an instruction such as:

Below are the inputs to an ROI calculation for an AI tool in a
small business. Act as a sceptical accountant. List any input that
looks double-counted, unsupported by a measurement, or likely to be
overstated, and say what evidence would settle each one. Do not
suggest new benefits to add.

[paste column A and B]

A reply for the salon's sheet might say (illustrative):

1. B7 (2.8 extra bookings a week): a conversion rise from 38% to 42% over a few weeks could be seasonal. Compare with the same weeks last year. 2. B5 ($20 an hour): check this includes employer costs, not just the wage. 3. B14 (upkeep $420): 1 hour a month seems low if prices change seasonally. 4. You could also add the value of improved staff morale, perhaps $500 a year.

Points 1 to 3 are fair challenges. The first is the one to act on: pull last year's enquiry and booking counts for the same weeks before the owner defends the uplift at the 90-day review. Point 4 ignores the instruction and is exactly the soft benefit that belongs on the separate list, not in the sum, so strike it.

If you're forecasting a new automation rather than measuring a live one, calculating ROI before you build covers how to get honest inputs from a small test. And if you'd like to look for savings from the other direction, starting with your costs, finding AI savings line by line in your profit and loss is a useful companion.

Finally, compare the answer with doing nothing and with the simplest alternative. If the ROI is positive but a quick way to get the same result would be a shared template or a small change to your booking form, that may be the better choice. When you're still deciding whether to spend anything at all, working out whether AI is worth it is the quicker first step.

AI ROI questions owners ask

What counts as a good ROI for an AI project in a small business?

There's no universal benchmark. A practical bar is a positive return in year one on the expected case, a payback period well under 12 months, and a pessimistic case that doesn't lose much. Compare it with other uses of the same money and time too: if hiring a few hours of admin help returns more, that's the better investment.

Should I include my own time if I'm the owner?

Yes. Your setup hours, the time you spend updating instructions and the reviews you attend are real costs, even if no invoice arrives. Value them at what your hour is worth to the business, usually what you'd earn or protect doing your normal work. Leaving owner time out is one of the most common ways AI ROI gets overstated.

How do I value time saved for salaried staff?

Use their loaded hourly cost, but only for time that turns into something: work that previously didn't get done, overtime avoided, or capacity for sales. If the saved minutes simply disappear into a slightly quieter day, the saving is real for the person but not for the business accounts, so count it at zero or a reduced rate.

Further reads

• AI KPIs for Small Businesses: 12 Metrics Worth Tracking — The ongoing metrics that keep your ROI figure current.
• How to Measure Whether Copilot Is Paying for Itself — The same method applied to Microsoft 365 Copilot licences.
• How to Measure an AI Receptionist's Return in the First 90 Days — A 90-day return check for an AI receptionist.
• Why AI Saves Time but Not Money, and How to Capture the Gain — How to turn saved hours into money you can count.
• How to Review an AI Tool After 90 Days: Keep, Fix or Cancel — Where the ROI figure feeds a keep, fix or cancel decision.
• How Much Should a Small Business Budget for AI? — Set the cost side before you spend.
• How Much Does AI Cost a Small Business in 2026? — Itemised 2026 AI budgets for a sole trader, a six-person office and a 20-person firm, with the charges that grow after month three.
• AI or a New Hire? How to Decide Before You Recruit — Break a planned role into tasks, see which ones AI can absorb, cost hire against AI over a year, and set a clear trigger for recruiting anyway.
• Example AI Roadmap for a 12-Person Business, Month by Month — A 12-person driving school's first AI year, month by month: five workflows, about $30 a month in new tools, one idea postponed and why.
• AI Use Case Template: Score Every Idea on One Page — A one-page template with scoring anchors, knock-out questions and a worked veterinary example for ranking AI ideas before you spend anything.
• Should You Charge Clients Less When AI Speeds Up Your Work? — What to do about your prices when AI cuts the time a job takes: the rules for hourly billing, when fixed fees are fair, and how to tell clients.
• Pros and Cons of Using AI in Your Business, With Real Costs — Every pro and con of business AI with a price attached, a first-year ledger for a small law firm, and the costs that never reach an invoice.
• How Much Time Can AI Really Save a Small Business Each Week? — Realistic weekly time savings from AI: what research found, per-task estimates, a seven-person accountancy practice added up, and a two-week measuring method.
• Can AI Help a Small Business Grow Revenue, or Only Cut Costs? — When AI grows revenue and when it only trims costs: diagnose your bottleneck, pick from five revenue levers, and check two worked examples with numbers.
• Is AI Dynamic Pricing Worth It for a Small Hotel? — When AI rate-setting pays for a small hotel and when it doesn't: five deciding factors, three example properties, break-even maths and the guardrails to set.
• How Photographers Cut Editing Time With AI and Keep Their Style — Train an AI editing profile on your own finished work, test it on a job you've delivered, and review by scene so the look stays yours.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: ChatGPT Business, Claude Team and Zapier pricing pages (checked September 2026). All salon figures are illustrative.