Prompt 1“We want to give all the employees of our company (80 people, in administration, sales and production) a generative artificial intelligence assistant. Prepare a project plan for the rollout, with phases, timing, an indicative budget and the KPIs to measure the return on investment.”
I propose a 12-week project, with an initial pilot and a gradual extension to all 80 employees. The goal is to reduce the time spent on specific tasks while maintaining the quality of the work.
The timing and amounts below are planning estimates: they assume a standard business assistant, with no software development or complex integrations.
Tasks to start with
| Area | First uses to evaluate | Benefit to measure |
|---|---|---|
| Administration | Drafts of communications, document summaries, report preparation | Time per case and reduction in rework |
| Sales | Meeting preparation, drafts of quotes and emails, summaries of sales information | Preparation time and speed of response |
| Production | Looking up approved procedures, drafts of shift reports, summaries of anomalies | Time spent searching for information and completeness of reports |
For production, check the availability of devices, access during shifts and the relevance of the uses. Answers that concern operating procedures must be checked against the approved documentation.
Phases, timing and expected results
| Phase | Period | Main activities | Result |
|---|---|---|---|
| 1. Kick-off and analysis | Weeks 1–2 | Appoint owners, interview the three areas, select the tasks and measure current time and quality | Shared goals and starting point |
| 2. Selection and configuration | Weeks 3–4 | Compare solutions, check data terms, configure accounts and access, define usage rules | Environment ready for the pilot |
| 3. Pilot | Weeks 5–8 | Involve 12–15 people from the three areas, train them and measure results weekly | Evidence on usefulness, quality and costs |
| 4. Decision | Week 9 | Compare results with goals, update the business case | Extension, adjustment of the pilot, or suspension |
| 5. General rollout | Weeks 10–12 | Gradually enable the other employees, provide role-based training and set up support | Access and readiness for all 80 employees |
| 6. Optimisation | From month four | Review usage, benefits and licences; improve the use cases | Ongoing control of the return |
An internal project lead coordinates the project; management approves the budget and the extension. IT handles configuration and access, and the heads of the three areas validate the uses and the results. A qualified contact person reviews the contractual and data protection aspects.
Before the pilot, simple rules are needed: authorised tools, permitted data, human review of outputs, and how to report problems.
Indicative first-year budget
Reference scenario with 80 licences for 12 months: this is a conservative forecast, even though activation will be gradual.
| Item | Assumption | Amount |
|---|---|---|
| Licences | 80 users × CHF 30/month × 12 months | CHF 28,800 |
| Kick-off, configuration and project support | External services | CHF 8,000 |
| Training for the three areas | Preparation of materials and sessions | CHF 4,000 |
| Organisational support and maintenance | External assistance during the year | CHF 3,600 |
| Contingency | About 10% of the items above | CHF 4,400 |
| Budget for purchases and services | CHF 48,800 | |
| Internal time | 300 hours in total × CHF 35/hour | CHF 10,500 |
| Total economic cost for the first year | CHF 59,300 |
The CHF 30 fee is a calculation assumption, not a verified price. The 300 internal hours include attending training, coordination and measurement.
Any connections to management systems, document archives or production systems require a separate assessment. After the pilot, adjust the number of licences to actual needs.
KPIs and conditions for going ahead
The proposed thresholds should be agreed before the pilot starts.
| KPI | How to measure it | Proposed initial target |
|---|---|---|
| Net time saved | Comparison on equivalent tasks, including review and corrections | At least 15% on the selected tasks |
| Quality | Assessment with shared criteria, before and after | At least equivalent to work without AI |
| Useful adoption | Users who complete agreed tasks with the assistant every week | At least 70% of pilot participants |
| Rework | Outputs needing correction out of the total assessed | No increase compared with the starting point |
| Correct use of data | Checks and reports | No serious unresolved issues |
| Economic return | Measured benefits compared with total cost | Payback within a period approved by management |
The extension goes ahead only for the uses that show a benefit. Logins and number of prompts, on their own, do not measure the return.
Example ROI calculation
For illustration, let us assume the following for the first year:
- 60 active users;
- 20 net minutes saved per day;
- 150 days of use per person, taking the gradual start into account;
- an average hourly cost of CHF 35;
- 50% of the time saved can be put to productive use.
The result would be: 3,000 hours saved × CHF 35 × 50% = CHF 52,500 of capacity value.
With a total cost of CHF 59,300, the first-year ROI would be about −11%. With the same assumptions, breaking even would require about 23 net minutes saved per day per user.
This benefit represents available working capacity. It becomes a financial saving when it reduces an actual expense; any additional revenue must be demonstrated separately.
The first investment decision should therefore cover the kick-off and the pilot, leaving the extension dependent on the results observed.