Hook
An AI initiative rarely dies because the technology failed. It dies because nobody could say, in terms a CFO respects, what it cost and what it returned. Without that, the initiative is a believer's project: it survives on enthusiasm, and it ends the moment enthusiasm meets a budget review.
A leader driving AI adoption needs two things the believer does not: a credible business case to justify the investment going in, and honest measurement to prove the return coming out. Neither is hard. Both are routinely skipped, because AI's benefits feel obvious to the people close to it, and "obvious" does not survive contact with a finance committee.
This lesson builds both. It is deliberately unromantic. The goal is that you can walk into any room, including the skeptical ones, and account for your AI investment the way you would account for any other serious use of the organization's money and attention.
Context
The cost side, honestly
Most AI business cases understate cost, because the visible cost is small and the real cost is not.
The visible cost is subscriptions and tools: modest, easy to name. The real costs are larger and quieter. There is the time to learn: every person becoming genuinely capable with AI spends real hours doing so, and that time has a value. There is the time to build: workflows and agents do not build themselves, and the building is real work. There is the management cost: governance, oversight, the leadership attention this very track describes. And there is the cost of getting it wrong: the rework when an unverified AI output reaches a client, the cleanup when a bad process is automated.
An honest business case names all of these. Understating cost is not optimism; it is a credibility risk. A case that claims AI is nearly free invites exactly the scrutiny that finds the hidden costs and discredits the whole proposal. Name the real number and the case is defensible.
The value side, specifically
The value side fails the opposite way: it is described in terms too vague to count. "Increased productivity." "Better insights." "Efficiency." A CFO cannot bank a vague noun.
Real AI value falls into four nameable categories, and a strong case specifies which ones apply and how much. Time saved: hours returned on a recurring task, which convert directly to money. Quality improved: fewer errors, better output, measurable as reduced rework or improved outcomes. Capacity created: work that gets done that simply was not getting done before, new throughput. New capability: something the organization can now do at all that it could not do before.
The discipline is to take each claimed benefit and force it into one of these four, with a number attached. "AI makes us more productive" becomes "AI saves each of twelve analysts about four hours a week on report preparation." The second sentence has a business case in it. The first does not.
The measurement trap
Once an initiative is running, the temptation is to measure what is easy and flattering rather than what is true. Number of people with licenses. Number of prompts run. Number of workflows built. These are vanity metrics: they go up reliably, they feel like progress, and they prove nothing about value.
The trap is that vanity metrics are genuinely easy and genuinely satisfying, so they crowd out the real ones. A leader has to insist on measuring outcomes, not activity. Not "how many people have access" but "how much time is the work actually taking now versus before." Not "how many workflows exist" but "what is different in the output or the throughput." If a metric would still rise even if the AI created no value at all, it is a vanity metric, and it belongs nowhere near your business case.
When the dashboard becomes the boss, the people who create the most measurable output get rewarded - not the people who create the most valuable output.
Leading and lagging indicators
Real value often takes time to show up, which creates a gap: the initiative needs signs of life before the final results are in. The answer is to track both leading and lagging indicators.
Lagging indicators are the real outcomes: time saved, quality up, capacity created. They are what actually matter, and they arrive late. Leading indicators are the early signs that those outcomes are coming: are people actually using the tools in their real work, is usage deepening, are the pilots hitting their interim marks. Leading indicators do not prove value, but they predict it, and they tell you early whether an initiative is on track or quietly failing. Track leading indicators to steer; judge by lagging indicators.
The honest kill criterion
The final piece of a credible business case is the one believers omit: the condition under which you will stop. An initiative with no defined failure point is not a business case; it is a faith. Deciding in advance what result, by what date, would mean an initiative has not worked, and committing to act on it, is what makes every other number you present believable. A leader willing to name the kill criterion is a leader whose success claims can be trusted.
Steps
Step 1: Build the honest cost model
For any AI initiative, write the full cost: the visible tools and subscriptions, and the real costs of time to learn, time to build, ongoing management and the cost of errors. Resist understating it. The honest number is the credible number.
Step 2: Force every benefit into a countable category
Take each claimed benefit and assign it to one of the four value types: time saved, quality improved, capacity created or new capability. Attach a number and a target to each. Discard any benefit that cannot be made specific; if it cannot be counted, it cannot carry the case.
Step 3: Capture the baseline before you start
Before the initiative begins, measure the current state of whatever you intend to improve: the hours the task takes now, the current error rate, the current throughput. Without a baseline captured beforehand, you can never prove a change, only assert one. This is the single most skipped step and the most important.
Step 4: Choose real metrics and reject vanity ones
Select two or three metrics that measure outcomes, not activity. Apply the test: if the metric would rise even when AI created no value, reject it. Track leading indicators to steer the initiative and lagging indicators to judge it.
Step 5: Set the review date and the kill criterion
Fix a date, typically around 90 days, for an honest review against the baseline. And define now, in writing, the result that would mean the initiative has not worked, along with your commitment to act on that result. A business case with a defined failure point is one a skeptic can believe.
Recap
- AI initiatives die not from technology failure but from the absence of a credible business case and honest measurement. Enthusiasm does not survive a budget review.
- Name the real costs: not just subscriptions but the time to learn, the time to build, the management overhead and the cost of getting it wrong. Understating cost destroys credibility.
- Force every benefit into a countable category: time saved, quality improved, capacity created or new capability, each with a number. Vague value cannot carry a case.
- Avoid the vanity-metric trap. If a metric would rise even when AI created no value, it is worthless. Measure outcomes, not activity. Steer by leading indicators, judge by lagging ones.
- Capture the baseline before you start, set a review date, and define the kill criterion in advance. A business case with a named failure point is one a skeptic can trust.
A roadmap and a business case are leadership instruments. But adoption succeeds or fails with people. The next lesson is the hardest part of the job: getting a team to actually use what you have funded.
Continue: Change Management, Getting a Team to Actually Use AI →