The 4-week AI pilot: how Danish companies get started
22 April 2026
Want to go from AI idea to working prototype? Here's how we run a 4-week AI pilot at Indee — week by week, with real numbers and a real budget.
An AI pilot is a time-boxed prototype that tests one concrete AI use case inside your own organisation — typically over 4–6 weeks, on a budget of DKK 75,000–150,000. At Indee we run pilots that are focused and measurable: you go from hypothesis to a working solution with real users, real data and a clear decision about the next step. This article shows you exactly how — week by week, so you can judge for yourself whether your organisation is ready.
What is an AI pilot — and why not just a POC?
The difference between a proof of concept and a pilot is decisive. A POC proves itself in a demo environment and often dies there. A pilot runs on real data, with 5–10 selected users, and ends with a clear go/no-go criterion.
According to McKinsey's 2025 State of AI analysis, 60–70% of POCs never make it into production. The main reason is that they never met real workflows. That's why we at Indee always take it all the way — and measure ourselves on whether the solution is still running reliably 6 months after launch, not just on delivery day.
Week by week: how we run a 4-week AI pilot
Week 1 — Use case and data assessment
- Two workshops (4 hours each) with your team
- Clarifying the use case, success criteria and available data
- GDPR screening: which data can we use, and how?
- Choosing the model stack (e.g. Claude, GPT, self-hosted Llama)
Deliverable: a 2–4 page pilot specification.
Week 2 — Build the first version
- Model integration, prompt design, and RAG set-up against your documents where relevant
- Internal testing with synthetic data
- First preview with your product owner
Deliverable: a working prototype in a closed test environment.
Week 3 — Real users, real data
- 5–10 employees use the solution in their daily work
- Daily logging of output, errors and feedback
- At least one mid-point iteration with adjustments
Deliverable: usage data, an error overview and improvement proposals.
Week 4 — Evaluation and decision
- Before/after KPI comparison (time spent, error rate, satisfaction)
- An honest go/no-go recommendation
- A scaling plan if the decision is go
Deliverable: an evaluation report plus a concrete quote for full implementation — or an honest recommendation to stop.
What does an AI pilot actually cost?
At Indee we work with fixed prices — no open-ended hourly billing. Typical ranges in 2026:
| Pilot type | Price | Duration |
|---|---|---|
| Small (e.g. one chatbot or classification task) | DKK 75,000–100,000 | 4 weeks |
| Medium (RAG on a document archive) | DKK 100,000–150,000 | 4–6 weeks |
| Complex (integration with several systems) | DKK 150,000–200,000 | 6–8 weeks |
The price covers everything from workshops to running the solution through the test period. Licence and API costs at the model provider (OpenAI, Anthropic, Google and others) typically come to DKK 2,000–8,000 over the pilot period, depending on traffic volume.
Which use cases make the best first pilot?
From the projects we've run since 2023 — including our own product Validi, where AI generates journal notes and care plans in production at Danish clinics — we see three patterns that almost always deliver positive ROI:
- Document search and question answering (RAG on internal knowledge)
- Drafts and summaries (emails, reports, notes)
- Categorisation and triage (incoming enquiries, tickets, documents)
More ambitious use cases — fully autonomous AI agents or decision support in safety-critical flows — can certainly be run, but shouldn't be your first pilot.
Frequently asked questions
What does an AI pilot cost?
An AI pilot at Indee typically costs DKK 75,000–150,000 including workshops, development, testing and evaluation. The price is fixed and agreed up front — we don't work with open-ended hourly billing. On top of that come licence and API costs at the model provider, which over the pilot period run to DKK 2,000–8,000 depending on model and traffic volume. After the pilot you receive a concrete quote for full implementation based on real data from the test period.
How long does it take to implement AI in our organisation?
A pilot takes 4–6 weeks from kick-off to evaluation. Full implementation after a successful pilot typically takes a further 2–4 months depending on integrations and compliance requirements. Realistically you should plan for 4–6 months from first workshop to stable operation — not 4 weeks. We measure success on whether the solution still delivers value 6 months after going live.
Should we build in-house or hire an AI partner?
If you have in-house ML engineers and LLM experience, you can build it yourselves. Most Danish companies we talk to don't have that set-up yet — and the learning curve is steep. An AI partner like Indee typically saves you 3–6 months on your first pilot and reduces the risk of ending up with a POC that never reaches production. We can also help build up your internal team over time, if the goal is to bring the capability in-house.
Want a no-obligation conversation about where an AI pilot would make the most sense for you? Contact Indee or see our AI implementation services.