Insights

The AI business case most projects never write

Hours saved is a claim. A business case is a number finance has agreed to count.

  • Ravi R. Singh
  • 4 min read

A saving nobody can find

Ask a team how their AI pilot went and you will hear a number of hours saved. Ask the finance director where those hours went and the room goes quiet. The hours were real. They were absorbed into the working week, and no cost line moved.

This gap is one of the main reasons projects stop after the pilot. Gartner predicted in July 2024 that at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025. Unclear business value was one of the four causes it named, alongside poor data quality, inadequate risk controls and escalating costs.

The Microsoft and LinkedIn Work Trend Index for 2024 found that 59% of leaders worry about quantifying the productivity gains of AI. They are right to worry. Many AI cases would not survive ten minutes with a finance team, and the fault lies in how the case was built.

Decide what counts before you count

A business case starts with an agreement about what counts as a result. There are only a few honest options.

The first is cost removed. A licence is cancelled, an outsourced task comes back in house or overtime falls. Finance can see this in the accounts. The second is capacity released. The team handles more volume with the same people, or a planned hire is no longer needed. This is real only if the volume exists and somebody commits to the hiring decision.

The third is revenue, such as faster quotes that win more work. It is the hardest to prove because many other things move revenue too. The fourth is risk, such as fewer errors reaching a customer. It needs a baseline error rate, which most teams do not have.

Pick one as the primary measure and get finance to agree to it in writing before the work starts. Hours saved feeds into these measures. It should never be reported as the result.

Count the full cost

The cost side is usually understated. The licence fee is the smallest part.

Someone has to review the output, and review time does not fall to zero. Someone has to fix the cases the tool gets wrong. Staff need time to learn the new way of working, and for a few weeks they will be slower than before. Process documents, access rights and data clean-up all take hours from people who have other jobs. If usage grows, usage-based fees grow with it.

A worked example shows the effect. Suppose a tool drafts replies to 400 routine emails a month and saves six minutes on each. That is 40 hours a month. If a person spends two minutes checking each draft, about 13 of those hours go back. If one draft in twenty needs a full rewrite at ten minutes, another three hours go.

The net gain is about 23 hours a month, a little over half the headline figure. The project may still be worth doing. The decision should be made on 23 hours.

Write down when you would stop

A good case includes the conditions under which you would stop. Set a review date, usually three months after go-live, and state the minimum result that justifies continuing. If the net gain is below that line, the project ends or changes.

This sounds negative and it has the opposite effect. Sponsors approve projects more readily when they can see the exit. Teams report results more honestly when stopping is a planned outcome and nobody's failure.

Add one sensitivity check. Halve the expected gain and double the review time. If the case still holds, proceed with confidence. If it collapses, you have learned that it depends on everything going right.

Agree the baseline first

No gain can be measured without a starting point. Before the tool is switched on, measure the task as it runs today for two weeks. Record how many items there are, how long each takes and how many come back with errors.

Teams resist this because it delays the launch. It is the cheapest part of the whole project, and without it every later number is an opinion.

One page, agreed with finance

The finished case fits on a page. It states the primary measure, the baseline, the expected gain, the full cost, the review date and the stop condition. Finance has signed the measure. The process owner has signed the baseline.

Sequence matters as well. A roadmap that lists eight use cases with no order is a wish list. Put the case with the clearest measure and the lowest cost of error first, and let its result fund the argument for the second.

This is the Think stage in the way I work. I offer it as a fixed scope called the AI Business Case and Roadmap at US$3,500, planned at 20 founder hours. Whether you buy help or do it yourself, the discipline is the same.

An AI project without this page is running on enthusiasm. Enthusiasm is useful at the start, and it does not survive the first budget review.

Sources

AI Business Case and RoadmapAll articles