Prompt 2The professional's prompt: the role of a senior digital transformation consultant, eight requirements, limits and format, and the context of an SME in Ticino. The full text is in the AI projects tab of the experts' prompts.
Draft adoption plan for an AI assistant — SME in Ticino, 80 employees
The goal is to make an AI assistant available to employees through a controlled process, first showing which tasks improve and under what conditions it makes sense to extend its use.
The only information available is: 80 employees, based in Ticino, no AI policy, and some staff using personal accounts. Industry, organisation, existing tools, budget and data processed are [to be verified]. The timing, pilot size and thresholds below are proposals for approval, not company data.
1. Start from the problems to solve
Before choosing the product, identify frequent tasks that take up time or cause rework.
| Possible problem to verify | Candidate use of AI | Measure of the result |
|---|---|---|
| A lot of time spent preparing recurring texts | Prepare a first draft from authorised information | Total time until the approved version |
| Long documents to read and compare | Summarise content with references to the sources | Time taken and omissions found |
| Repetitive work on non-confidential information | Classify or reformat content | Time per case and error rate |
Whether these problems exist and how relevant they are is [to be verified]. Select two or three use cases based on frequency, expected benefit, how verifiable the result is, and data risk.
Searching internal documents requires a further check of permissions and access before connecting company archives.
2. Bring current use under control
Propose a temporary rule, communicated to everyone:
- Do not enter non-public company information or personal data into personal AI accounts.
- In the pilot, use only company accounts and approved tools.
- Check outputs before using or sharing them.
- Report errors or possible data exposure to a designated contact person.
Carry out a review of the tools already in use, the tasks performed and the categories of data that may have been uploaded. The aim is to identify exposures and offer a practical alternative. Any past incidents are [to be verified].
3. Governance and responsibilities
Appoint a senior management champion and an operational project lead. These roles can be filled by existing staff or by external support: availability and names [to be verified].
In the matrix: A approves and is accountable for the decision; R carries it out; C is consulted; I is informed.
| Activity | Management | Project lead | IT/security | Privacy/legal contact | Department heads |
|---|---|---|---|---|---|
| Approve goals and budget | A | R | C | C | C |
| Define policy and data rules | A | R | C | R | C |
| Assess vendor and configuration | I | A | R | C | C |
| Select use cases and measure results | I | R | C | C | A |
| Organise training and support | I | A/R | C | C | R |
| Handle technical or data incidents | I | A | R | R | C |
| Decide go/no-go and extension | A | R | C | C | C |
Each employee remains responsible for checking the content they use in their own work; the department head approves uses within the process.
4. Data rules and choice of tool
The Swiss Federal Act on Data Protection (FADP) also applies to processing personal data with AI. Whether a data protection impact assessment is needed depends on the risk of the processing; applicability to the project [to be verified]. (edoeb.admin.ch)
For the pilot, the following classification is proposed:
| Category | Proposed rule |
|---|---|
| Public information | May be used in the approved tool |
| Internal, non-confidential information | May be used only for authorised use cases |
| Confidential information, client documents or personal data | Excluded from the initial pilot; admission subject to a specific review |
| Passwords, credentials, keys and access secrets | Must never be entered in prompts |
Before purchasing, check: use of data for training, retention and deletion, data location and transfers, sub-processors, access management, authentication, contract terms and export options.
The actual guarantees depend on the product and the contract: [to be verified]. The label “business” alone does not replace this check.
5. Starting point and pilot
Proposed duration: 12 weeks, depending on people's availability and the vendor checks.
| Phase | Proposed period | Required result |
|---|---|---|
| Review and initial safeguards | Weeks 1–2 | Current use mapped, temporary rules in place and owners appointed |
| Baseline measurement and preparation | Weeks 3–4 | Use cases, baseline data, tool and evaluation criteria approved |
| Pilot | Weeks 5–10 | Weekly measurements, problems logged and user support |
| Evaluation | Weeks 11–12 | Updated business case and documented decision |
Involve 10–12 employees, as an initial proposal, who represent the selected tasks and have different levels of digital familiarity.
Before the rollout, measure on a sample of tasks:
- Volume and frequency of the work.
- Total time, including review and corrections.
- Quality, errors and rework.
- Relevant company hourly cost: [to be verified].
During the pilot, compare tasks of similar complexity, and include in the time with AI the preparation of prompts and the checking of the output. Define a shared quality grid beforehand, to limit purely subjective assessments.
6. Go/no-go criteria
The following thresholds are proposals to be validated before the pilot:
| Criterion | Proposed condition for going ahead |
|---|---|
| Efficiency | Median reduction in total time of at least 15% in the use cases to be extended |
| Quality | Quality at least equivalent to the current process, with no increase in serious errors |
| Adoption | At least 70% of participants use the assistant regularly in the agreed tasks |
| Data and security | No serious unresolved issues; mandatory checks passed |
| Economic sustainability | Conservative benefit higher than costs, and payback within the company threshold [to be verified] |
The decision can be:
- Go: extend the use cases that pass the checks.
- Targeted extension of the pilot: gather more evidence when the sample is insufficient.
- No-go: suspend the extension if quality, security or sustainability remain insufficient.
Success in one task does not automatically justify extending to others.
7. Business case and costs
With the information available, it is not possible to give a reliable investment figure or ROI. The business case must be completed with quotes and pilot results.
| Item | Calculation method | Amount |
|---|---|---|
| Licences | Enabled users × fee × months | [to be verified] |
| Configuration and initial checks | Internal hours and external support | [to be verified] |
| Initial training | Hours of participants and trainers | [to be verified] |
| Recurring management and support | Monthly hours × hourly cost | [to be verified] |
| Any integrations needed | Specific quote | [to be verified] |
| Exit or change of vendor | Export, reconfiguration and transition | [to be verified] |
Do not assume that all 80 employees need a licence immediately: needs, frequency of use and access arrangements are [to be verified]. Everyone must still receive appropriate rules and training.
The benefit model is:
Annual hours saved = eligible annual tasks × share actually done with AI × net minutes saved per task ÷ 60.
Value of capacity saved = hours saved × company hourly cost × share of time that can actually be reused.
The time freed up represents available capacity. It becomes a monetary saving only when it reduces a demonstrable expense, for example overtime or external services. Avoid counting the same benefit twice.
For the first year, include start-up, training, recurring costs and gradual adoption. Calculate the steady-state cost separately.
First-year ROI = (first-year benefits valued − total first-year costs) ÷ total first-year costs.
Present a conservative, a central and a favourable scenario, varying actual usage, net time saved and the ability to reuse it. All values are [to be verified].
8. Change, risks and control after the extension
Provide hands-on training on real tasks, examples of permitted use, exercises in spotting errors, and a support channel. Department heads must identify how to use the time saved and collect operational difficulties.
| Risk | Proposed measure |
|---|---|
| Wrong but convincing answers | Human review, source checking and clear limits on uses |
| Data exposure | Managed accounts, classification and access controls |
| Continued use of personal accounts | An accessible company alternative and understandable rules |
| Low adoption or extra workload | Task-based training, support and measurement of the full time |
| Overestimated economic benefits | Observed measurements, conservative scenarios and verification of the capacity saved |
| Vendor dependence | Exportable procedures, exit terms and periodic review |
After the go, extend in successive groups and review usage, quality, costs and incidents every month. Any new connection to data or systems requires a check before activation.
The initial decision proposed to management is to fund a review phase and a pilot with a maximum budget of [to be verified]. The financial commitment for the extension is decided on the measured results.