AI Strategy
The AI ROI Test: What Business Number Are You Trying to Move?
A seven-question test that takes an afternoon and settles most AI debates before anyone writes code. Question seven is the one people skip.
Executives are not short of AI proposals. They are short of a defensible way to choose between them.
The test below is deliberately unsophisticated. Seven questions, one afternoon, no technical background required. Its purpose is not to find the best AI project. It is to establish whether a proposal is a project at all.
The Seven Questions
1. What business problem are we solving? One sentence, in business language, with no technology in it. If the sentence collapses without the word *AI*, send it back.
2. What metric represents that problem? One primary metric. Not a portfolio. "Qualified leads per month" qualifies; "customer experience" does not.
3. What is the current baseline? Written down, dated, and agreed before work begins. Undocumented baselines produce arguments at review time, and the arguments are always won by whoever is most confident.
4. What would improvement be worth? Convert one unit of the metric into money using your own economics. If nobody can value one unit, the project cannot be evaluated — and that is a finding worth having early.
5. Where can AI influence the workflow? Describe the existing workflow step by step and mark the point where the capability acts. Then describe the handoff in one sentence: what it produces, who receives it, in which tool, and what they do next. Also decide what stays human, who reviews it, and who is accountable.
6. How will the result be measured? Baseline, window, comparison. Sixty to ninety days is usually long enough to learn something and short enough to matter.
7. What happens if the metric does not improve? Agree the answer now: modify, or stop. This is the question that gets skipped, and skipping it is how a pilot becomes a permanent line item that nobody can cancel because nobody ever defined failure.
Applied: Customer Acquisition
The primary example, because every question has a real answer.
Problem. We pay to bring visitors to our website; most leave without identifying themselves, so we cannot follow up.
Metric. Qualified leads and booked consultations per month.
Baseline. Last quarter's actuals, by month, from the systems already in use.
Worth. Average customer value multiplied by close rate gives the value of one additional qualified lead. Conservative inputs only.
Where AI enters. An engagement layer on the existing website: conversation, intent discovery, answers grounded in approved company knowledge, qualification, lead capture with context, and a proposed next step delivered into scheduling and follow-up.
Measurement. The stages of OceSha's AI Customer Acquisition Framework — visitors, conversations, qualified prospects, leads, appointments, customers, revenue — each with a rate.
If it does not improve. You will know which transition failed, which usually points at something fixable in the handoff rather than the model.
Applied Elsewhere
The test is not specific to acquisition. Run it on anything.
Customer support. Metric: share of inquiries resolved without a human, or first-response time. Baseline from the ticketing system. Worth: cost per contact. AI enters at triage and drafting; humans keep judgment calls and anything with risk attached.
Employee onboarding. Metric: days until a new hire is productive by an agreed definition, or hours of manager time consumed per hire. Worth: loaded salary time. AI enters as guided answers to the questions new hires ask repeatedly.
Training and enablement. Metric: completion, assessment pass rates, or time to competency in a role. Worth: the cost of the gap the training exists to close. AI enters in course production and personalized practice. This is where our AI Training & Enablement work lives.
Member engagement. Metric: renewal rate, program registrations, or active participation. Worth: annual member value. AI enters in helping members find the program, certification, or resource that applies to them.
Operational automation. Metric: cycle time or throughput on a named process. Worth: hours reclaimed, or revenue held up by the delay. AI enters at a specific step, with a specific handoff.
Notice that every example names a metric before it names a capability. That ordering is the entire test.
What the Test Protects You From
Applied honestly, it filters out proposals with no owner, no baseline, no handoff, no stopping condition, and no economics — which is to say most of them. That is not cynicism about AI. It is how you end up funding the initiatives that survive contact with a quarterly review.
Choose one number. Establish the baseline. Identify the single point in the workflow where AI could plausibly move it. Instrument the path. Review in ninety days.
If you want a worked starting point, read the AI Customer Acquisition capability or talk with OceSha Ventures about which number to pick first.
