The AI ROI Test: Can You Draw a Straight Line From AI to Revenue?

Executives are not short of AI ideas. They are short of a way to tell which ideas are worth money.

The test below is deliberately simple. It takes about twenty minutes per initiative, it can be applied by someone with no technical background, and it disqualifies most proposals before anyone writes code — which is the point.

For every proposed AI initiative, answer five questions in order.

1. What Business Problem Are We Solving?

State the problem in one sentence, in business language, without naming a technology.

Good: "A large share of the people who visit our website never contact us, and we have no idea who they were."

Not yet a problem statement: "We should be using generative AI in marketing."

If the sentence requires the word AI to make sense, you have described a solution looking for a home. Send it back.

2. What Metric Should Change?

Name one primary metric. One. It must already be tracked, or trackable within days, and it must have a current baseline.

Examples of usable metrics: qualified leads per month, consultation requests per month, quote turnaround time, share of inquiries answered within an hour, proposals produced per week, support resolution time.

Examples that are not usable as a primary metric: productivity, innovation, customer experience, efficiency. These are categories, not measures.

Write the baseline down. Undocumented baselines produce arguments later.

3. What Is That Metric Worth Financially?

Convert one unit of the metric into money.

This is the step most organizations skip, and it is the step that makes everything after it honest. If one additional qualified lead is worth a known amount to you — through your own close rate and average customer value — then the value of any change in lead volume is arithmetic rather than opinion.

Use your own numbers, conservatively. If you cannot estimate the value of one unit, you cannot evaluate the project, and no vendor's benchmark should substitute for your data.

4. Where Does AI Enter the Existing Workflow?

Describe the current workflow step by step, then mark the exact point where the AI capability acts. Be specific about the handoff: what the system produces, who receives it, in which tool, and what they do next.

If that handoff cannot be described in a sentence, the initiative will produce a demonstration rather than a result. This is the most common failure mode in AI projects — not weak models, but capabilities that never touch the workflow.

Also decide what stays human: which outputs require review, who is accountable, and what they can see and correct.

5. How Will We Measure the Result?

Define the measurement before starting: the baseline, the observation window, the comparison, and what would count as failure.

Set the window short enough to matter — often 60 to 90 days. Agree in advance what result would lead you to stop. A project with no stopping condition is not an investment; it is a commitment.

The Worked Example: AI Customer Acquisition

Run the test on customer acquisition and every question has an answer.

Problem. We pay to bring visitors to our website. Most read a page and leave without calling, booking, or identifying themselves, so we cannot follow up.

Metric. Qualified leads and consultation requests per month, against the current baseline.

Financial value. Known, from average customer value and close rate. In professional services, consulting, healthcare and dental, legal, and B2B technology, one additional acquired customer usually carries meaningful value — which is exactly why the leakage is expensive.

Where AI enters. An AI Concierge engages visitors on the existing website, answers questions from approved business knowledge, discovers intent, qualifies the prospect, captures the lead with context, and moves qualified prospects toward a consultation. Output lands in the systems already in use — scheduling, notifications, follow-up, CRM — where a person picks it up.

Measurement. The funnel is counted end to end:

Traffic → Conversations → Qualified Leads → Appointments → Customers → Revenue

Each stage has a number. Each transition has a rate. If the rates do not improve, you will know, and you will know where.

What This Framework Rules Out

Applied honestly, the test filters out the projects that consume the most attention and return the least:

• Initiatives with no owner of the metric
• Initiatives whose value depends on a benchmark from someone else's business
• Initiatives with no defined handoff into an existing workflow
• Initiatives that cannot be evaluated within a quarter
• Initiatives that exist because the technology is interesting

That last one deserves stating plainly. Interest is not a business case. A capability worth funding improves an identifiable business outcome that someone in the organization is already accountable for.

Start With One Line

Choose one number. Establish the baseline. Identify the single point in the workflow where AI could plausibly move it. Instrument the path. Review it in ninety days.

If you can draw a straight line from the work to revenue, the project is worth funding. If you cannot, the honest answer is not yet.

Learn about AI Customer Acquisition Systems, or talk with OceSha Ventures about which number to start with.