Skip to content
AI

AI pilot in 4–6 weeks: what you get, what it costs, and what you skip

Adrian WollumFounder, WOLLUM8 min read

TL;DR

An AI pilot in 4–6 weeks is a productised offering with a fixed scope and fixed price (typically NOK 80,000–250,000). You get one specific process automated in production — not a demo, not a report. Prerequisites: one clearly defined process, accessible data, and one decision-maker. The exclusions are deliberate: no scope creep, no open hours, no delivery without a real user testing it. 70 percent of Norwegian SMBs that start this way go on to phase 2.

An AI pilot in 4–6 weeks is a scoped delivery where a business gets one concrete AI solution from concept to production in under two months, with fixed pricing and a fixed scope. The purpose is to validate both the technology and the process before any further investment — not to produce a final solution covering the whole business.

Most AI projects do not fail on technology. They fail on scope. The business has ten good ideas, the vendor has 200 hours on an open estimate, and six months later there is still not a single production-deployed use case. The AI pilot in 4–6 weeks is WOLLUM's answer: one specific process, fixed price, fixed delivery date. The demo is replaced by a pilot that is actually in production when you open your laptop on the first day of week 7.

What you actually get

The pilot is a finished solution for one concrete process — not a strategic report, not a demo that only shows one scenario. You get actual software running in production, used by actual employees, connected to the systems you already use.

  • One specific work process automated in production — typically email sorting, invoice processing, document classification, or customer service routing.
  • Integration with one existing system (Microsoft 365, Tripletex, Visma, or similar). More systems are possible, but then it is no longer a pilot.
  • Pilot in operation with 5–20 real users — not a test environment, not a demo with hand-picked data.
  • Logging and monitoring in place from day one. You can see what the AI does, and roll back.
  • GDPR assessment delivered as a document, not an afterthought.
  • Handover with documentation, training, and full source code in your hands.

What we deliberately exclude

What is not delivered matters as much as what is. Every time we have delivered a pilot on hourly billing instead of fixed scope, it has become 3 times as expensive and 2 times as late. So these exclusions are explicit:

  • No additional processes. If you want to automate both email and invoices, that is two pilots — not one extended pilot.
  • No design variation or A/B testing. That comes in phase 2 with data from the pilot.
  • No custom UI. We use standard interfaces where possible — often directly inside Microsoft 365 or the system you already have.
  • No complex approval flow. The pilot version has one approver. Roles and hierarchies come later.
  • No model fine-tuning. We use GPT, Claude, or Gemini with good prompt engineering. Fine-tuning is rarely worth it at the pilot stage.

What it costs

Fixed price ranges for Norwegian SMBs in 2026:

  • Simple pilot (email sorting, document classification): NOK 80,000–120,000.
  • Standard pilot (invoice processing, customer service routing with one system integration): NOK 120,000–180,000.
  • Advanced pilot (RAG against your own documents, AI agent with 1–2 tools): NOK 180,000–250,000.

Running cost in production is typically NOK 500–3,000 per month for the AI model itself, depending on volume. No surprise costs: scope and price are locked before we start.

Who it fits

The pilot delivery suits businesses that meet four criteria: you can pick one process and stick to it, you have data accessible (not spread across 12 SharePoints), one person is the decision-maker, and you are willing to set up larger scope and budgets after the pilot has proven value.

It fits less well for businesses that want to change scope mid-flight, that have data scattered and unstructured on paper forms, where no one is empowered to make decisions across departments, or who expect to solve the whole business's AI needs in a single delivery.

Timeline week by week

  1. Week 1: Discovery, scope locking, goal definition, GDPR assessment, and baseline measurement of time spent on the process before AI.
  2. Week 2: Design and prototype. Access to test data. Model selection. First technical sketch shown to you.
  3. Weeks 3–4: Build and integration. We set up logging, approval flow, and fallback from the start.
  4. Week 5: Test with 2–3 real users. Adjustments based on the failure patterns we see.
  5. Week 6: Expand to 5–20 users. Measure against baseline. Handover with documentation, training, and source code.

Example: an accounting firm in Rogaland

An accounting firm in Rogaland with 1,200 incoming receipts a month spent two days a week on coding and review. Pilot scope: an AI that reads receipts, suggests account code, VAT code, and approver based on history. Price: NOK 145,000 fixed. Delivery week 6: AI in production, the accountant still clicks "approve" but typically 80 percent of cases go through without manual adjustment. Time on receipt review fell from two days to four hours per week. Paid back in under five months.

What happens after the pilot

The pilot delivery stands on its own. You are not obligated to continue with us. If the pilot delivers as expected, three common next steps are: extended automation (more processes, more complex logic), AI agent covering a wider workflow, or integration with more systems. If the pilot does not deliver as expected, you are only NOK 80,000–250,000 worse off — not six months behind schedule and a million in debt to an open hourly project.

Considering an AI pilot? Contact WOLLUM for a no-obligation conversation. We will find one process that is mature for piloting and deliver fixed pricing and timeline in the first conversation. Read also our guide to AI integration for businesses or what AI integration costs.

Frequently asked questions

What is the difference between an AI pilot and a full AI project?
A pilot is deliberately narrow: one process, fixed price, 4–6 weeks. A full project covers multiple processes, has complex approval flows, and takes 3–6 months. The pilot validates that you should invest in the full project — it does not replace it.
What if we are not satisfied with the pilot result?
We measure the result against the baseline we set in week 1. If the pilot does not deliver as promised, we fix it inside the fixed price. If the process turns out to be unsuitable for AI (rare, but happens), the pilot cost is all you have risked — not a half-finished large project.
Can we change the scope mid-flight?
No, not within the pilot delivery. That is the whole point. Scope changes are handled in phase 2 after the pilot is delivered and measured. This keeps the pilot predictable in both time and price.
Do we need an in-house developer or IT department?
No. We deliver the entire solution and hand over complete documentation. You need someone who can answer process questions and approve access to test data. That is typically 4–8 hours from one person on your side over the 6 weeks.
What about GDPR and sensitive data?
A GDPR assessment is part of the pilot delivery and is provided as a document in week 1. We set up in EU regions with data processing agreements. For particularly sensitive data, we can use models that run locally or in Norwegian cloud — this is discussed in the discovery conversation.
No commitment · 30 min · Free

We build what you need. Let's figure out what that is.

Book 30 minutes with us and get an honest picture of what the right technology can do for your business.

You can change these choices at any time. Necessary is always on because it is required for the site to work.

Necessary

Required for basic functionality, such as remembering your consent choice. No tracking.

Statistics

Anonymised visit statistics via Google Analytics 4. Helps us understand which pages people read.

Product insights

Detailed behavioural analysis via PostHog, including session recording with masked input fields. Used to improve the user experience.