Building a clean business case for AI
Short Answer
A clean business case for AI weighs four sides honestly: the measured value, the full cost including operation, a risk discount for exceptions, and the time until it pays off. The number everyone quotes first is the licence cost. It is almost always the smallest line.
A business case for AI rarely falls on the value side and almost always on the cost side nobody fully writes down. Four quantities belong side by side: measured value against a documented baseline, all costs from rollout to ongoing operation, a discount for errors and exceptions, and the time until the calculation turns positive. The most important and most often forgotten line is operation: monitoring, rework, permissions and a named owner keep costing after go-live. Skip them and a demo looks great while the real bill arrives in the second quarter. An honest case states no invented percentage but a range, a baseline and the assumptions behind it.
Why a Hunch Misses the Business Case
Most AI initiatives start with a feeling: this will surely save time, it must pay off. The feeling is often not wrong, but it is not a basis for a decision. It hides the sides that are uncomfortable and leans on the one number that is easy to name. That is exactly why so many cases look good on paper and fall apart in operation.
The most common mistake is to size the value large and the cost thin. Value gets measured against the best demo, cost against the licence. What is missing are the quiet lines: internal time for rollout and upkeep, rework on errors, permission and privacy questions, ongoing operation. A business case that does not carry these lines is not a case but a wish with a table.
A business case for AI rarely fails on value. It fails on the costs nobody wrote down, and on a number everyone celebrated too early.
The Four Sides of the Calculation
A defensible case puts four quantities side by side and treats none as a footnote. Each can be sized roughly but honestly before the build.
Value is calculated against a measured baseline, not against an ideal picture. Cost runs from rollout into operation. Risk is carried as a discount, because no system works error-free and exceptions cost work. Time says when the calculation turns positive, because value in three years is not the same as value in the first quarter.
| Side | Honest calculation | Flattered |
|---|---|---|
| Value | against a measured baseline, as a range | against the ideal case, as one number |
| Cost | rollout plus ongoing operation | licence or project price only |
| Risk | discount for errors and exceptions | assumption that all runs smoothly |
| Time | payback with a date | value without a point in time |
The right column is not malice but the default when nobody asks. The job of the business case is to force the left column.
The Cost Line Almost Everyone Forgets
When a case tips in the second quarter, it is almost never the rollout. That was planned for. It is the operation nobody sized. An AI system does not stop making work at go-live. Someone has to watch quality, approve exceptions, maintain permissions, keep an eye on cost and rework when the model gets it wrong. That work is real, recurring and belongs in the calculation.
Operation is tricky precisely because it hides as internal time and does not show up as an invoice. Half a role for upkeep and approvals appears in no licence contract but costs every month. So the most honest question in a business case is not "what does the tool cost" but "who operates this after launch, and how much time does it take".
| Cost block | When it lands | Often underestimated because |
|---|---|---|
| Rollout | one-off, before launch | integration with legacy systems takes time |
| Licence and model | ongoing, visible | the only number clearly on the invoice |
| Operation and upkeep | ongoing, hidden | it sits inside internal working time |
| Rework and exceptions | ongoing, variable | exceptions feel like edge cases but are the norm |
Before budget flows, open the full calculation: KI-Beratung services review value, cost and operation on your case. Strategy call: Book a strategy call.
Assess AI potentialThe Business Case in Six Steps
Before you write a number into a template, walk these six steps. They cost hours, not weeks, and save you a calculation that does not hold in operation.
- Measure the baseline of today's workflow before AI touches it, or you have nothing to calculate value against.
- Size the value as a range, conservative to optimistic, and write down the assumptions behind it.
- Count all costs, from rollout through licence to ongoing operation and rework.
- Set a risk discount for errors, exceptions and cases where a human takes over.
- Calculate payback with a date, not just value without a point in time.
- Name an owner who carries the case after launch and checks the numbers.
Core sentence: a business case is not a sales figure but an honest bet with visible assumptions. It may be conservative and still convince, as long as the calculation behind it is clean.
Common Misconceptions
The business case is the value number.
Licence cost is the cost driver.
A positive case proves the project pays off.
Glossary
AI ROI, Total cost of ownership (AI), Baseline, AI operation.
Questions and Responses
How do I size value without inventing a number?
Is an estimate enough for the business case, or do I need a pilot?
When is a case too good to be true?
Sources and Evidence
No external statistics or client results are asserted as verified. Concrete numbers come from the measured baseline of the specific case.
Turn your initiative into a defensible calculation: KI-Beratung services show the structured entry, or book a strategy call at our booking calendar.