If you want to put AI into one specific workflow and don’t yet know whether it will pay off, a five-day audit is the cheapest way to find out. Our AI Production Audit is built for Nordic companies of roughly 50–250 employees that already use ChatGPT or Copilot but haven’t connected AI to their own systems and data. It takes five business days, costs €3,500, and ends with a written report that either proposes a fixed price for the build or tells you not to build. The fee is credited in full against the AI Production Launch if you start within 60 days.
An AI readiness assessment is a short review that answers whether a specific process should be automated with AI, what it would take and what could go wrong. Many assessments on the market are questionnaires that score your “AI maturity” or end in a strategy deck. Ours is deliberately narrower: one workflow, your real data, your systems, and a number at the end.
Why audit before you build anything?
Most AI projects that stall don’t fail because the model is bad. In 2024 Gartner predicted that at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025, and the reasons it gave were poor data quality, inadequate risk controls, escalating costs and unclear business value.
Every one of those can be checked before anyone writes production code. That is what the audit is for. The rest of the road from a working demo to a system people rely on is covered in our playbook for taking a GenAI proof of concept to production.
What happens on each of the five days?
| Day | What we do | Your time |
|---|---|---|
| Day 1 | A 2–3 hour workshop with the process owner and IT. We walk through the workflow as it runs today, using real examples: what comes in, who handles it, which systems are involved, how long it takes and where mistakes are expensive. | 2–3 hours from the process owner and an IT contact |
| Days 2–4 | We map the systems and data the workflow touches, check data quality and access on real samples, review the data flows for security and GDPR, and choose the model and hosting. | Granting read access or exports, answering follow-up questions |
| Day 5 | We deliver the written report and walk you through it. | The people who will decide on the launch |
Days 2–4 are where most of the findings come from:
- Systems. Which systems the workflow reads from and writes to (an ERP such as Fortnox, Visma or Business Central, a CRM, SharePoint, a shared mailbox), whether they have usable APIs, and what access we’d need.
- Data quality. Whether the examples are complete and consistent, and whether the “right answer” is recorded anywhere. If the correct outcome only lives in someone’s head, that’s a finding in itself.
- Security and GDPR. What personal data moves where, and which agreements would be needed.
- Model and hosting. Hosting in an EU region, preferably in your own cloud account; model APIs that don’t train on your data; or a self-hosted model when data must not leave your environment.
What do you need to prepare?
- A process owner who knows the workflow in detail and can join the day-1 workshop.
- An IT contact who can grant read access to, or exports from, the systems involved.
- A sample of real cases, such as recent emails, orders, invoices or customer questions, and the correct outcome where you have it.
- Today’s numbers, even rough ones: volume per month, time per case, error or rework rate. These become the baseline.
- Any constraints: data that must stay in your environment, existing cloud or Microsoft agreements, internal AI policies.
What’s in the written report?
| Section | What it tells you |
|---|---|
| Recommended workflow and scope | Exactly what the first version will do, and what it won’t. |
| Architecture | Where it runs, what it connects to, which model and why, and where a person reviews the output. |
| Success metric and baseline | The number that defines success, such as minutes per invoice or the share of questions answered correctly, and its value today. |
| Draft evaluation set | The real cases and scoring rules the system will be measured against. The launch completes the set in weeks 1–2. |
| Acceptance criteria | What the sprint-1 demo must show, in writing. |
| Risks | Data, security, adoption, cost per task and dependence on a single model provider, each with a mitigation. |
| GDPR and AI Act notes | Personal data flows, agreements needed, whether a DPIA is likely to be required, and how the AI Act treats the use case. |
| Fixed-price launch proposal | A plan of up to 6 weeks at a fixed price between €20,000 and €35,000, depending on scope, or a smaller first step or a longer fixed-price plan if 6 weeks won’t fit. |
Two of these deserve a definition. An evaluation set is a fixed list of real inputs with known correct outputs; the system is scored against it before launch and after every change, so “it seems better” becomes a number. The acceptance criteria are the basis of our first-sprint guarantee: we agree them in writing during the audit, and if the launch’s sprint-1 demo doesn’t meet them, you can stop the launch and don’t pay for that sprint.
What do the GDPR and AI Act notes cover?
On personal data, the report says what personal data the workflow touches, where it would be stored and processed, which model terms apply, and whether the processing is likely to need a data protection impact assessment (DPIA). If our engineers, who work from Ukraine, could come into contact with personal data, a data processing agreement with the EU Standard Contractual Clauses is signed first. Our security and data protection page describes the setup.
On the AI Act, most workflows we see, such as answering questions from documents, routing orders and invoices into an ERP or drafting quotes, are not high-risk. The use cases that are (recruitment and employee management, creditworthiness, access to essential services and the other areas listed in Annex III) carry heavier obligations, and the AI Omnibus has postponed those for stand-alone systems to 2 December 2027, as summarized by White & Case. The transparency rules in Article 50, such as making clear to people that they’re dealing with an AI system, apply from 2 August 2026. The report states which category your workflow falls into and what that means in practice. It’s an engineering assessment to brief your legal counsel or data protection officer, not legal advice.
When does the audit say “don’t build”?
“Don’t build it” is a valid outcome, not a failed audit. If we don’t believe the workflow will pay off, the report says so and explains why.
| What we find | What we recommend instead |
|---|---|
| The data isn’t available or isn’t good enough | Fix the data first. Sometimes the right next step is a clean-up, not a launch. |
| The volume is too low to pay back | Keep it manual, or use a feature in a tool you already pay for. |
| An off-the-shelf product already does it well | Buy it. Our guide to building versus buying AI covers how to decide. |
| Errors are expensive and hard to catch | Narrow the scope to a step where a person reviews every output, or leave it manual. |
| Nobody can describe the process the same way twice | Define the process first. Automating an undefined process gives you faster confusion. |
| It won’t fit in 6 weeks | A smaller first step, or a longer fixed-price plan. |
A rough illustration of the volume test, not a client result: take a launch at the top of the range, €35,000. If the workflow saves 20 hours a month of work that costs you €50 an hour, that’s €1,000 a month, and payback takes almost three years before running costs. If it saves 150 hours a month, payback takes under five months. The audit replaces these guesses with your own numbers.
If the answer is no, you’ve spent €3,500 rather than the €20,000–35,000 of a launch, and the findings on your data and systems stay useful whoever builds something later.
What does the audit cost?
| Item | Price | Terms |
|---|---|---|
| AI Production Audit | €3,500 | Fixed price, 5 business days |
| Credit against the launch | −€3,500 | In full, if the launch starts within 60 days |
| AI Production Launch | €20,000–35,000 | Fixed price set after the audit, depending on scope, up to 6 weeks, paid 30/40/30: at signing, at the mid-point demo and at production acceptance |
| Audit and launch together | €20,000–35,000 | When the credit applies |
Prices exclude VAT and are invoiced by Smoother Development AB in Sweden. After the launch you can keep the workflow running with Care Lite, our fixed-price maintenance plan from €4,200 a month, extend it with a larger team, or take it in-house: you own the code.
Who does the work?
A founder is accountable for every audit, and a senior engineer does the technical review. Our engineers work from Ukraine; your contract, the data processing agreement and the accountability sit with our Swedish company. AI we’ve built includes UpliftIQ, a decision intelligence platform with role-adapted AI agents and RAG (retrieval-augmented generation, where the model answers from your own documents and data instead of from memory), and CyberResilient’s AI-driven NIS2 maturity assessment, developed with support from MSB and NCC-SE.
What to do next
Pick the one workflow you’d most like to see in production and write down which systems it touches and roughly how much time it takes today. Then book the AI Production Audit: €3,500, five business days, credited in full against the launch if you start within 60 days. A founder replies, usually within one business day, to agree the scope and dates.
