If AI is not showing up in the P&L, it is still an experiment.

A lot of companies are measuring AI adoption when they should be measuring AI ROI. They can tell you how many employees have ChatGPT, how many people attended AI training, or how many prompts the team ran last month. Ask what changed financially, and the answer gets much less clear.

Time saved matters. Productivity matters. But those numbers become much more interesting when you follow them through the business. What happened with the five hours you gave someone back? Did the company avoid a hire? Did sales respond faster? Did outside consulting spend come down? Did the team increase customer capacity without increasing payroll?

I think every AI implementation should eventually answer a simple question: What changed in the economics of the business?

Time saved is the start of the ROI calculation.

Time savings are often the easiest place to prove value from AI because they are visible. A proposal that used to take two hours now takes 30 minutes. A weekly report that took half a day can be reviewed in 45 minutes. A leader can summarize 100 pages of material before a meeting instead of spending an evening reading it.

Those are meaningful gains, especially for a small business where one person may be carrying several jobs. But “we saved ten hours” is only the first half of the AI ROI calculation.

The second question is what the business did with those ten hours.

If the owner gets that time back and uses it to sell, meet customers, develop a new offer, or make better decisions, the value can be significant. If a team gets enough capacity back to serve 20 percent more customers without adding another employee, the result starts showing up in margin. If AI automation removes enough repetitive work to delay a $90,000 hire for a year, that is not a productivity statistic. That is a financial result.

This is why I want AI projects connected to the work and the P&L from the beginning.

THE ROI TEST

Follow the time through the business.

AI ROI shows up in more than one place.

When people hear ROI, they often jump straight to cost cutting. Cost reduction is one way AI creates value, but it is not the only one.

AI can increase capacity. A salesperson can manage more opportunities because meeting preparation, follow-up, and proposal drafting take less time. A service team can support more customers because research and administrative work happen faster. A founder can move through financial analysis, market research, and planning without waiting for three different people to assemble the information.

AI can also improve revenue. Faster response times can increase conversion. Better sales preparation can improve the quality of customer conversations. More consistent follow-up can keep good opportunities from disappearing. A stronger research process can help a company identify markets, partnerships, or product ideas it would otherwise miss.

Then there is decision quality. That one is harder to put into a spreadsheet, but it can be the most valuable. If AI helps an executive surface a contract risk, challenge an assumption, or see a problem before money gets committed, one better decision can pay for the entire AI strategy.

The point is to decide which kind of return matters before the AI workflow gets built.

Put a business metric on the workflow before you install it.

I like to baseline the work before changing it. How long does the process take today? How many people touch it? What does it cost? How often does it happen? Where does it break? What happens when it is late or wrong?

Without that baseline, AI implementation turns into storytelling. People remember that something feels faster, but nobody can say by how much. A team believes productivity improved, but nobody knows whether that improvement created additional capacity or simply filled the day with something else.

A good AI workflow should have a simple success measure attached to it. Proposal creation goes from two hours to 30 minutes. Weekly leadership reporting goes from five hours to one. Customer follow-up moves from three days to same-day. Outside research spend drops by $100,000. The owner gets ten hours back every week.

The metric does not need to be complicated. It needs to be specific enough that 30 days later, you can tell whether the AI workflow worked.

That is the difference between experimenting with AI and running an AI strategy.

Review AI like any other business investment.

I would not give an AI workflow unlimited time to prove itself. Once it has been running inside the real business, review it.

Did people actually use it? Did it save the amount of time expected? Did output quality improve? Did the workflow reduce cost, increase capacity, improve a customer experience, or help produce revenue? What still requires too much human intervention?

If the answer is yes, expand it. If the workflow is close, improve it. If it creates activity without a meaningful result, stop doing it.

This matters because AI adoption can become expensive in ways that do not show up on the software invoice. The larger cost is employee time spent testing tools, rebuilding workflows, attending training, and managing systems that never become part of how the company actually works.

A disciplined AI implementation earns its place the same way any other investment does. It produces a return.