The order book is growing, but the same experienced employees keep approving every difficult case. One manager proposes a new hire. Another wants to buy external support. A third believes AI can absorb the extra work. All three may be addressing different parts of the problem.
Before comparing prices, define the work that must be completed, the quality required and the constraint that prevents it. A new tool will not solve a shortage of authorized professional judgment if every additional draft still waits for the same reviewer.
Describe the shortfall as work, not headcount
Consider a hypothetical Norwegian service business with 40 employees. Its management team estimates that next quarter requires another 120 productive hours per month. Sixty concern routine document preparation, forty concern specialist review and twenty concern customer coordination.
These are planning assumptions, not measured demand. Validate them against actual cases, including peaks, absences and difficult work. Ask employees which tasks can be removed or simplified before adding capacity. If duplicate reporting consumes part of the workload, eliminating it may be more durable than automating it.
NHO's 2025 competence survey, published in February 2026, reports unmet competence needs among six in ten member businesses. That is relevant context for planning, not evidence that your particular vacancy can be filled by AI.
Compare four responses to the same need
Recruitment can build lasting capacity, customer knowledge and succession. Include the time to recruit and train, supervision, equipment, absence coverage and the risk that demand changes. A new colleague may also bring expertise the current process lacks. Do not value that contribution only as hours.
Buying an external service can provide specialist capability or temporary capacity. Define the delivery and acceptance criteria, internal coordination, access to information, rework and continuity. Buying a service and hiring agency workers are different arrangements; assess the actual relationship and applicable rules before choosing a contract form.
AI-assisted work may reduce repeatable preparation but adds evaluation, review, support and supplier dependence. Check whether the work it removes is the work limiting delivery. If there is no reliable source against which to check the output, the promised saving deserves particular scrutiny.
Process improvement may remove approvals, duplicate entry or unnecessary service variations. Some checks are essential, so establish why they exist before removing them. A combination can be best: simplify the process, assist routine preparation and recruit the expertise that remains scarce.
Put the AI option through a capacity calculation
In the example, assume a trial reduces the 60 preparation hours to 30. Additional checking and maintenance require 10 hours. Net capacity released is therefore 20 hours per month, leaving 100 of the original 120 hours unresolved. The 40 hours of specialist review and 20 hours of coordination have not disappeared.
If the tool costs NOK 6,000 per month, its tool cost alone is NOK 300 per net hour released. That excludes setup, training and other incremental costs. It is not directly comparable with an employee's hourly salary, and it is not NOK 6,000 in savings. The result may still be worthwhile if it removes a real delay or improves quality. It does not support a claim that one whole role is no longer needed.
Test a downside case too: if extra review consumes the entire saving, what remains valuable? If nothing does, stop or redesign the trial. Use the guide to measuring business results before treating capacity as an economic gain.
Involve the people who must make the new work function
Discuss the changed tasks, review burden, training and escalation with employees and their representatives as appropriate. Arbeidstilsynet's guidance on organizational change emphasizes information, participation and competence development. Treat their observations as evidence about whether the proposed work design is realistic.
Preserve routes for junior colleagues to learn the work they will later review. Assign responsibility for exceptional cases and keep someone able to operate when the tool or supplier is unavailable. Do not quietly transfer all difficult work to one senior employee while describing the change as a productivity improvement.
Make a staged commitment
For this business, I would first validate the workload, remove avoidable tasks and test one preparation workflow. In parallel, establish what expertise must be secured regardless of the trial. Waiting for perfect AI evidence should not delay a clearly necessary specialist hire.
The CEO owns the operating proposal and workforce plan. The board should assess material investment, long-term commitments and resilience implications at the appropriate level. Revisit the choice when demand, quality or review effort changes, not simply when a subscription renews.
Bring a one-page comparison to the management meeting: capacity delivered by task, time to usefulness, total incremental cost, quality risks and reversibility. If an AI trial is justified, choose its first workflow deliberately.
Sources and scope
Sources checked on 12 October 2026. The business, hours and costs are hypothetical. The comparison and calculations are my analysis. Workforce changes, procurement and employment arrangements require assessment of the applicable agreements and rules.