AI use cases: how to estimate value before a pilot
An AI use case is a specific application of AI in a single process, with a defined user, data and a measurable effect. Estimating its value makes it possible to compare ideas with one another and reject those that do not justify the investment. The article sets out formulas for three types of use case, a worked example, a sensitivity analysis, a conversion to a 1–5 scale and typical mistakes.
Amounts and values in the examples are illustrative and show how the calculation works. In projects, we use the client's data.
Dominik Dębski · Lemur Media · · reading time: 4 min
Reviewed by Dominik Dębski, 1 November 2026
What is an AI use case and how do you measure its value?
An AI use case is a description of one application of AI: the process, user, input data, output and success measure; its value is the annual change in the company's results less the annual cost of the solution. To keep the portfolio comparable, we use three types of formula, depending on where the effect arises.
| Use case type | Example | Annual value |
|---|---|---|
| Time savings | Ticket handling, preparing proposals | Value of time saved × adoption − cost |
| Revenue | Product recommendations, lead prioritisation | Revenue × change in conversion × margin − cost |
| Risk | Fraud detection, document checks | Expected value of loss avoided − cost |
Which formula do we use for time savings?
We calculate the annual value as the value of time saved, adjusted for the adoption rate and less the annual cost of the solution.
The adoption rate reflects the fact that not all employees will use the solution, and not in every case. Leaving out this factor is the most common cause of inflated estimates. Time saved only has monetary value if the company uses it differently: it handles a larger volume without new hires or moves people to other tasks. Before the calculation, the process owner specifies which of these two scenarios applies to the team and records it together with the estimate.
What does a worked example look like?
The customer service department receives 10,000 tickets a month, and handling one takes 6 minutes on average; under the assumptions in the table, the annual value is PLN 141,600.
| Component | Calculation | Result |
|---|---|---|
| Annual volume | 10 000 × 12 | 120,000 tickets |
| Working time | 120 000 × 0,1 h | 12 000 h |
| Cost of working time | 12,000 h × PLN 80 | PLN 960,000 |
| Time saving of 30% | PLN 960,000 × 0.30 | PLN 288,000 |
| Adoption of 70% | PLN 288,000 × 0.70 | PLN 201,600 |
| Annual cost of the solution | licences, integration, maintenance | − PLN 60,000 |
| Annual value | PLN 141,600 |
Every zloty of the solution's annual cost yields PLN 2.36 of net value (141,600 ÷ 60,000). The 30% time-saving assumption falls within a 2023 estimate by McKinsey & Company (a productivity gain from generative AI in customer service equivalent to 30–45% of that function's costs), but it is higher than the result of a 2023 NBER field study of 5,179 agents, which found an average productivity gain of around 14%. A conservative scenario should therefore start from a value closer to 14%. This is a macro estimate for the function as a whole, so it does not replace measurement in a pilot.
How do you test the sensitivity of the estimate?
The estimate is recalculated for a conservative and an optimistic scenario, because the share of time saved is usually the least certain assumption. In the example, changing the saving from 30% to 50% raises the annual value from PLN 141,600 to PLN 276,000 (960,000 × 0.5 × 0.7 − 60,000).
| Time savings | Adoption | Annual value |
|---|---|---|
| 20% | 70% | PLN 74,400 |
| 30% | 70% | PLN 141,600 |
| 50% | 70% | PLN 276,000 |
| 30% | 50% | PLN 84,000 |
The break-even point at 70% adoption is a time saving of 8.9% (60,000 ÷ 672,000). If the pilot decision depends on this assumption, the purpose of the pilot is precisely to measure it.
How do you convert the value to a 1–5 scale?
We set the scale thresholds with the board, in proportion to the size of the company's revenue and costs. Example thresholds for a company with 50–249 employees:
| Score | Annual value |
|---|---|
| 1 | below PLN 50,000 |
| 2 | PLN 50,000–149,999 |
| 3 | PLN 150,000–299,999 |
| 4 | PLN 300,000–599,999 |
| 5 | PLN 600,000 or more |
The customer service example scores 2 in the base scenario and 3 in the optimistic scenario. With a pilot qualification threshold of at least 4, the use case does not make the candidate list unless the company combines it with another process with a similar profile.
What are the typical mistakes and limitations of the estimate?
The estimate is a tool for comparing use cases with one another and loses its value when the assumptions have no source or owner. Together with the result, record the source of each figure, the person who confirmed the assumptions and the assumptions to be verified in the pilot.
- Ignoring the adoption rate and calculating savings for the entire team.
- Ignoring internal costs, e.g. IT team time, legal review and training.
- Counting the value of time that the company will not use in any other way.
- Comparing use cases calculated with different formulas without a common scale.
- No conservative scenario when deciding on the budget.
Sources
- McKinsey: The economic potential of generative AI, the next productivity frontier (14.06.2023)
- NBER Working Paper 31161: Generative AI at Work (E. Brynjolfsson, D. Li, L. R. Raymond, 2023)
- Harvard Business School Working Paper 24-013: Navigating the Jagged Technological Frontier (F. Dell'Acqua et al., 2023)