The most persuasive AI business case can also be the least useful: multiply minutes saved by hourly labour cost and present the result as money the company will earn. For a Norwegian business with a small specialist team, the freed time may be valuable. It may also remain scattered across a working week, with no reduction in spending and no additional work completed.
I would ask the board to examine three separate things: capacity released, value realised and cash actually changed. Treating them as interchangeable makes weak projects look attractive and can hide the value of good projects whose main benefit is quality or resilience.
Start with the constraint, not the licence price
A hypothetical Norwegian technical-services firm processes 1,200 routine requests a month. Suppose an assistant reduces handling time by eight minutes, after including human review. That is 160 hours of potential monthly capacity: 1,200 × 8 ÷ 60. At an assumed loaded internal cost of NOK 650 per hour, the capacity has an accounting value of NOK 104,000.
None of those numbers is an observed result or a Norwegian wage benchmark. They are assumptions for testing a decision. NOK 104,000 is not a cash saving while salaries and staffing remain unchanged. To realise commercial value, management must show what that capacity enables: fewer paid overtime hours, more completed billable work, shorter queues or an improvement the company explicitly values.
The calculation also fails if eight minutes measures drafting alone. Include the checking, corrections, escalations and maintenance caused by the new process. Measure completed cases, not time spent inside the AI tool.
Price the whole operating choice
I would want the investment proposal to separate four cost groups:
- Initial work: process redesign, integration, data preparation, legal assessment and employee training.
- Recurring work: licences or usage, review capacity, support, access administration and quality checks.
- Change and exit: retesting after supplier changes, migration, retraining and maintaining an alternative process.
- Downside exposure: plausible failures, their consequences and the resources needed to contain them. Do not disguise an unsupported loss estimate as a precise forecast.
For dollar- or euro-priced services, test the effect of NOK exchange-rate movements and higher usage. A low pilot bill says little about a service that will run on every customer interaction.
Waiting is an option with a price
Statistics Norway reported on 25 September 2026 that 48% of enterprises with at least ten employees in the covered industries, excluding finance, used AI. Adoption is not evidence of profitability, and that figure does not describe every Norwegian small business. Its relevance is that management should investigate changing expectations rather than assume the market is standing still.
Waiting can avoid premature integration, supplier dependence and wasted training. It can also preserve an expensive queue, postpone learning or leave employees using unapproved alternatives. Ask management to identify which effects actually apply to this company and how they would be observed. Fear of missing out is not a quantified cost of delay.
Fund the next piece of evidence
A board does not always have to choose between a company-wide rollout and doing nothing. It can authorise a bounded experiment with a spending limit, a named owner and a decision date. The experiment should resolve the uncertainty most likely to change the investment decision: review effort, customer acceptance, technical integration or demand for the released capacity.
Before it starts, define a downside case. If quality stays acceptable but review work removes most time savings, would management still proceed? If the answer depends on a different benefit, name and measure that benefit now.
The leader's guide to turning time into results explains the operating work behind those benefits. The board's task is to approve a defensible use of capital and make it possible to stop when the evidence disappoints.
Sources and scope
Sources checked on 11 October 2026. The financial illustration is hypothetical, excludes tax and financing effects, and is not an ROI forecast. The investment approach and interpretation of the adoption statistic are my analysis.