AI consulting for small companies in Finland
AI consulting at a small company comes down to one question: which parts of your weekly work are worth handing to a machine, and which are not. We answer it with one measured pilot, a written rule for your company data, and a straight answer when AI is the wrong tool.
- No minimum contract: book a few hours, stop whenever you want
- Scope agreed in writing before work starts
- Independent advice: no reseller commissions
The honest answer is shorter than the sales pitches suggest, but it is real and it can be measured in hours.
Who this is for
Founders and teams in Finland whose working language is English, and EU companies working with us remotely. Typically a company with a few people, a lot of repeated written work, and no time to test ten tools to find the one that helps.
What AI does well today
| Use | What it means in practice |
|---|---|
| Drafting | Quotes, job ads, product descriptions, social posts, as a first draft a person finishes |
| Summarizing | Long email threads and meeting notes |
| Customer service assistant | Answers the most common questions using your own information |
| Document handling | Reads an invoice or an order and picks out the fields |
| Automation glue | Moves data from one system to another and sorts it |
| Translation | English, Finnish and Swedish drafts for customer communication |
Notice what is missing: forecasting, sales optimization and decision making. They are often sold under the same name, but they need more data than a small company usually has.
What AI does not do
- It does not judge for you. A model produces convincing text even when it is wrong. Judgment stays human work.
- It does not know your prices. Not your price list, delivery times or terms, unless you give them to it. Without them, an assistant invents something that sounds right. This is where most customer service bots fail.
- It cannot be left alone with a customer. A bot that promises a price or a delivery date without oversight makes commitments for your company. The limits are set in advance: what it may answer, when it hands over to a person, and what it never says.
- It does not fix a broken process. If quotes go unsent because nobody follows them up, AI writes them faster and more of them go unsent.
Every setup we build has the same shape: your input, the model, a human check, then the output. The check can be light. In a customer service assistant it means the conversation moves to a person as soon as a question falls outside the agreed scope. That step is never removed.
One rule for your company data
Do not paste customer names, contracts or personal data into a service whose terms nobody at your company has read. Free tools may use what you type to improve their models; paid business versions usually do not. The terms are the only place that tells you which applies. In practice the rule means three things: pick one official tool and tell everyone, remove names and identifiers when the task does not need them, and write one page on what may be entered and what may not.
If an AI tool really does process personal data, the GDPR requires a data processing agreement with the provider (Article 28) and an entry in your record of processing activities (Article 30). It is not a big job, but it is done before use, not after. The Finnish Data Protection Ombudsman publishes guidance in English if you want to read the rules yourself.
What a pilot is
A pilot is one bounded use, tested on real work, followed by a decision. It is not a strategy or a review of the whole company. After the free first call you can choose a light review of two or three possible uses, a small pilot with one use built and tested, or a customer service assistant built from your material, quoted in writing first. Prices are on the pricing page. On top of any project there is usually a monthly subscription per user for the tool and a usage fee for the model. Ask for both separately from anyone you buy from, because they are rarely in the project price.
For a customer service assistant, the key step is testing: pick 30 genuine customer questions from last month's messages and run them through it. The questions it could not answer also show what your website is missing.
Measure before you start
Most AI projects do not fail; nobody knows whether they work, because nothing was measured first. Before we build, we write down one number: questions a person had to answer, minutes to write one quote, rows typed by hand each week. A month later we take the same number again and count how often the output had to be corrected. If there is no difference, the use is stopped. We agree that rule before building, because afterwards it is hard to agree on.
When AI is the wrong tool
| Situation | Why |
|---|---|
| The process is broken | AI makes it broken faster and at larger volume |
| You cannot name work that repeats every week | Without repetition there is no saving |
| Your information is in three places and contradicts itself | The assistant answers wrong because the material is wrong |
| Nobody has time to check the output | Unchecked output reaches a customer sooner or later |
| You want AI because a competitor has it | That produces a cost, not a measurable benefit |
The first row matters most. If invoicing runs late because hours are never logged, the problem is that the data does not exist, and the right purchase is simple custom software for logging work, not AI. We say this on the first call, because it is cheaper than saying it after a pilot.
How we work
- Half an hour on the phone, free. We go through your weekly work and look for the task that repeats and can be checked.
- One use, not ten. We pick the one where the saving can be measured in hours. The rest go on a list.
- The data rule in writing. One page: what may be entered, into which tool, and what never.
- Pilot and test with real questions. Thirty genuine questions before anything reaches a customer.
- Measure after a month. If no time was saved, we stop rather than expand.
Who does the work
DT Nova Tmi, Business ID 3595452-5, Helsinki. The founder, Mustafa Tarabya, has a background in business development and digital transformation, and does the work himself. We build what we recommend, which keeps the advice narrow. DT Nova has published 22 apps on the App Store (see them on the App Store, counted 25.10.2026). If the calendar is full, we say so on the first call. Related: IT consulting and app development. This page in Finnish: tekoälykonsultointi. To start, book a free discovery call.
Questions we get
Can we put customer data into an AI tool?
Only into a service whose terms someone has read and with which you have a data processing agreement. The practical rule: do not paste a customer's name or contract into a tool whose terms nobody at your company knows. Remove names whenever the task does not need them, which is more often than people expect.
Will an AI assistant give our customers wrong answers?
It will if it is not given your information and clear limits. That is why we build it from your own material, test it with real customer questions before launch, and hand any question outside its scope to a person. Every answer it gives has a route to a human.
Will AI replace an employee?
At a small company, no. It removes repeated work such as first drafts, summaries and moving data between systems. Judgment, sales and responsibility stay with people. If someone promises you a headcount saving, ask them which weekly tasks disappear and how they will measure it.
Does it work in Finnish as well as English?
Yes. Current models write and understand Finnish well enough for customer service and drafting, and English even more so. Industry terms and product names need to be given as material, otherwise they come out wrong. We test both languages with real questions before anything goes live.
Who owns the result?
You do. The material is yours, and the source code of anything we build is handed over to you. You can change supplier without starting from scratch, and the accounts for the AI tools themselves are opened in your company name, not ours.
What if the pilot does not deliver?
Then it is not expanded. We measure after a month, and if no time was saved, we say so and stop. That is agreed in writing before anything is built, because stopping is cheaper than continuing to pay for something that does not work.