Practical example
AI use-case and business-case practice pack
A worked, fictional first-year business case for one AI use case: baseline, benefits, full costs, ROI with its assumptions, a downside case, risks and a pilot with success and stop conditions.
Illustrative example. The organisation, figures and outcomes are fictional and were written by LSBUK for practice. They are assumptions in a planning exercise, not results, and no AI investment has a guaranteed return.
The case
A customer service team answers about 4,000 policy questions a year, averaging 6 minutes each. The proposal: an assistant drafts answers from approved policy documents, and staff check and send them.
First-year numbers
| Item | Amount | Note |
|---|---|---|
| Expected annual benefit | £20,000 | Assumes the time saving is realised and redeployed to work of equal value |
| One-time implementation | £8,000 | Integration, set-up, staff training |
| Annual operating cost | £4,000 | Licence and usage, monitoring |
| First-year total cost | £12,000 | £8,000 + £4,000 |
First-year ROI = (benefit − cost) ÷ cost = (£20,000 − £12,000) ÷ £12,000 = 66.7%
That figure holds only if the benefit is actually realised and the cost list is complete. It leaves out internal staff time for the project, any extra compliance review and the cost of fixing errors. Add those before deciding.
A downside case
| Scenario | Annual benefit | First-year cost | First-year ROI |
|---|---|---|---|
| Expected | £20,000 | £12,000 | 66.7% |
| Downside: half the expected saving | £10,000 | £12,000 | −16.7% |
The downside case shows the decision depends heavily on adoption. That is why the next step is a pilot, not a full roll-out.
Risks that are not in the numbers
Incorrect drafts sent without checking, outdated policy documents, data-protection obligations, supplier lock-in and staff who do not use the tool. Each needs an owner and a control.
The pilot decision
- Success condition: handling time down at least 30% over six weeks, with no increase in complaints or errors.
- Stop condition: an error rate above today's manual rate, or adoption under 50% after training.
- Owner: head of customer operations, reporting to the executive sponsor.
Decide in advance what result would make you scale up, and what would make you stop.
Field guide
| Field | What to record |
|---|---|
| Business problem | The problem in business terms, without mentioning AI. |
| Baseline | How the work is done now, with the measure you will compare against. |
| Use case | What the AI system would do, in one sentence. |
| Data readiness | Whether the data exists, its quality, who owns it and any personal-data issues. |
| Benefits assumption | The benefit you expect and every assumption behind it. |
| One-time cost GBP | Implementation, integration, training and change costs. |
| Annual operating cost GBP | Licences, usage, monitoring and support each year. |
| Annual benefit GBP | The expected yearly benefit if the assumptions hold. |
| Risks | Accuracy, bias, privacy, security, supplier and adoption risks. |
| Pilot success condition | The measurable result that would justify scaling up. |
| Stop condition | The result or event that would make you stop. |
| Owner | The named executive accountable for the outcome. |
Download
- Blank template (CSV, opens in Excel or Google Sheets)
- Completed illustrative example (CSV)
- Template with field guidance (Markdown text)
Everything here is illustrative practice material created by LSBUK. It is not real client, learner or cohort data.
AI strategy
The full guide: building, documenting, measuring and presenting an AI strategy.
Open the guide →Related articleAI Strategy Document Template: What to Include
What an AI strategy document should contain, with a filled example.
Read the article →CourseAI Strategy for Executives
Build and challenge business cases like this for your own organisation's use cases.
View the course →