How to Measure AI ROI Before You Invest: A One-Page Business Case for SMBs

On 24 October 2026, Apollo’s chief economist Torsten Slok wrote that outside the big technology companies AI has not yet delivered a visible financial payoff: revenue per employee and margins across the rest of the S&P 500 are not rising. Model prices keep falling, and one developer’s analysis says two major providers cut theirs by up to 80% within a few months. Cheaper AI is not the same as profitable AI, and an owner who pays for subscriptions, tokens and integrations every month needs an answer to “what did we get”. This guide shows how to measure AI ROI before you invest: full cost, three kinds of effect, a two-week baseline, a one-page business case, stop rules and a monthly report.

How to measure AI ROI: why cheap AI is not the same as profitable AI

Measuring AI ROI honestly means counting every cost line, not only the subscription. Falling token prices shrink one line of the bill. Integration, data preparation, training and the time people spend reviewing AI output stay where they were. The Apollo note frames the gap at the level of the whole economy, and the same gap shows up inside a single company.

Disappointing projects share the same patterns: nobody recorded a baseline, only the subscription was counted, saved hours never turned into money, and the pilot never ended. With prices falling, the token bill is rarely the hard part. If your data and processes are not ready, start with the AI readiness checklist before you spend anything.

Full cost of AI: subscriptions, tokens, integration, support and people’s time

Any AI ROI calculator is only as good as its inputs. Measuring AI ROI starts with total cost of ownership (TCO) for a full year, not with the price list. A pilot with ten users costs little, and the same tool rolled out to the whole company can cost many times more.

Cost line	What it includes	Often forgotten
Subscriptions and tokens	Seats, API calls, model fees	Usage growth after rollout
Integration	CRM, mailbox, databases, workflows	Change requests after launch
Data	Cleaning, access, storage	Staff time to fix documents
People	Training, review of AI output, supervision	Internal hours at a full hourly rate
Support	Quality checks, updates, incidents	Who owns it after the pilot

Cost per outcome

The cost-management team at AWS suggests dividing AI cost by a business metric to get a cost per outcome, for example the cost of one resolved request. It is a unit figure you can track every month, and it is the building block of ROI.

Which effects to count: hours, money, errors, response speed

• Time: hours of routine work the AI takes over.
• Money: extra margin or recovered leads, not extra revenue.
• Quality: fewer errors, rework and refunds, priced at their real cost.
• Speed: faster answers to customers, linked to conversion from your own history.

To measure the ROI of AI, start with a simple rule. Saved hours are not savings until they remove overtime, a hire or a contractor, or move to paid work. Count each effect once, and keep soft benefits in a separate line. Pick AI pilot success metrics from this list before the pilot starts. A process that is already automated with AI workflow automation is easier to price because each step leaves a record.

Baseline before launch: what to measure in the two weeks before the pilot

Before you learn how to calculate AI ROI, collect the numbers it needs. Nothing else counts.

1. Pick one process with a clear owner and a link to money.
2. Time ten to fifteen people on that task for one to two weeks, instead of trusting a manager’s estimate.
3. Record monthly volume, error rate and average response time.
4. Price one working hour at a full rate, including taxes and equipment.
5. Write down who will use the tool, so adoption can be tracked.

To measure the payoff before launch, treat AI as one more investment whose ROI must be proven. Keep these numbers unchanged, since the pilot will be compared with them. Someone has to measure the result of each AI task, and the ROI should be read by the same person every month.

One-page AI business case: template for the owner and the finance director

The business case fits on one page. Use it as a simple AI ROI calculator in a spreadsheet and show three scenarios: conservative (confirmed effects only), base and optimistic.

Field	Question it answers
Process and owner	What changes, and who is accountable
Baseline today	Hours, volume, errors, response time
Full yearly cost	Every line from the TCO table
Value in money	Time, money and quality effects
Result	ROI = (value minus cost) divided by cost, plus payback
Stop rule	What result ends the pilot, and on which date

Here is how to measure AI ROI on one page: fill the baseline first, estimate cost honestly, then take only the effects you could defend to a finance director. A business case for AI is a decision tool. The AI ROI framework matters less than the discipline of writing the assumptions down.

When to stop a pilot and when to scale

AI ROI measurement works best when thresholds are set before launch, with the review date. Compare AI pilot success metrics with the baseline after the agreed period, not in the first weeks.

• Stop if the main metric has not moved and adoption stays low.
• Fix if progress is visible but cost per outcome is rising.
• Scale if cost per outcome falls and quality holds.

A short business case for AI also tells you which pilot to drop first. If you want an outside view, AI consulting services can run the review and write the stop rule with you.

Monthly AI report: five numbers an owner should see

1. Total AI spend for the month, by process.
2. Cost per outcome against the previous month.
3. Adoption: who actually uses the tool.
4. Change in the main metric against the baseline.
5. Decision: keep, fix or stop.

Whichever AI ROI framework you choose, report it monthly. One page every month beats a long report once a year, and measuring AI ROI becomes routine instead of an argument.