Look for an AI consultant and the market looks uniform, but it is not. One pile holds large firms with nobody here, solo trainers who run good workshops, agencies that install a chatbot and call it transformation, and a small number of teams that build a system and keep it running. All of them describe themselves in the same words. This guide is about getting what you thought you were buying. We write it as a team that builds here and lives with the results of its own work.
Four kinds of AI provider you will meet
Four different businesses sell under the AI label: trainers who teach your team the tools it already has, resellers who configure one platform and invoice monthly, large consultancies with method and governance but nobody local, and builders who build the system and keep running it. Each is an honest business. The trouble starts when you buy one and believe you bought another. Our page on AI consultancy in Estonia covers how to tell the four kinds apart, so this guide starts where that one stops: what to check before you sign.
Checklist: six things a good AI partner does
If a provider does all six, it has probably done this work for real. If two or three are missing, ask why.
- Looks at your work before the price. A price given before anyone has looked at your business is a guess.
- Names the exact tools and the reason. Vague technology talk covers thin work.
- Tests in your customers' language. Have the system answer your customers' real questions. A recorded demo proves nothing.
- Leaves the code, accounts and data in your name. Then switching providers is your call, not theirs.
- Says what happens when the model is wrong. The system hands the request to a person, logs it or tries again, but never drops it quietly.
- Puts the scope in writing before work starts. Anything outside the scope is a new line with its own price.
Red flags in a pitch or an offer
None of these alone makes a provider bad. Several at once is a reason to walk away.
- There is a price, but nobody has seen your work. The number is a guess, and whatever it left out arrives later as an extra invoice.
- Every problem gets the same product. You are talking to a reseller, and in its world there is no other option.
- The accounts and keys stay in the agency's name. Switching providers later is costly or impossible.
- You only see a recorded demo. You do not know how it answers your customers in their language.
- Asked about errors, they quote an accuracy percentage. There is no plan for errors, and a request can go missing without anyone noticing.
- The answer to everything is yes. You are being sold software, not judgement.
Seven questions to ask before you sign
These are the questions we would ask if we were the buyer. Send them in writing and compare the answers.
What did you find before you named a price? A number given before anyone looks at your business is a guess in nice packaging. The work starts with an audit that names where money leaks: data retyped between two systems, quotes that go out three days late, enquiries nobody answers in time.
Which tools exactly, and why those? A clear answer names the platform and the reason: Make when you want a hosted setup your own people can read, n8n on your own server when the data has to stay under your control or volume makes per-step pricing hurt. Both are good answers. No answer is not.
Who owns the code and the accounts? The answer decides whether you can ever switch providers. We go through it below.
What happens when the model is wrong? Every model is wrong sometimes. The design question is what the system does next: hand over to a person, log it, try again, or fail without a sound and lose the contact. Someone who has shipped answers at once, because they have been burned. Someone who has not talks about an accuracy percentage.
What should not be automated? The most useful thing a consultant can say is often no. A broken process has to be fixed first, because automating it only makes the mess faster. If everything you mention gets an excited yes, you are talking to a salesperson.
Who actually does the work? Ask who you will talk to after the contract is signed. The person presenting in the sales meeting and the person building the system are not always the same.
What does month three look like? A system ages with your prices, offers and ways of working. Ask who watches and fixes it after launch, and how you will hear about it.
Code, accounts and data: how you leave
If the accounts and keys are in the agency's name, your systems stop the day the work ends.
Before you sign, ask three things: whose name the service accounts are in, where the code lives, and who gets the documentation.
A good answer is short: everything is in your name, and the handover is written into the contract.
What we build stays in your name, so you can fire us without breaking your business. Our trust center describes how we hold access, and our technical partnership page covers documentation and handover.
How to compare written offers
Send every provider the same brief and have them price the same unit of work. Then read the lines that decide the real total: running cost kept apart from the build, whose name the accounts are in, a written acceptance test, and what happens when volume grows. The full comparison sheet is in our guide on how to compare AI automation quotes.
What the free audit decides
Half an hour with Karl
- where your work loses time, money or information, and what that costs you today;
- whether a build is worth it at all;
- where to start, if it is;
- what your systems, volumes and data look like, because that is what the price is made of.
If a build is not worth it, we say so. The findings stay with you even if you do not go further with us. You can book the audit on our contact page.
How the price is built
There is no single price, because every job is different. We build the price from the work the audit found: the effort each part takes, plus an allowance for coordination and risk. Then we check it against what the problem costs you today.
The price depends mainly on:
- how many workflows we change;
- how many systems we connect, and whether they have a ready interface;
- what state your data is in;
- how many languages the system works in;
- who maintains the system after launch.
The build is a one-off fee, written down before work starts. Watching and fixing the system is a separate monthly line, if you want it. Third-party costs, such as AI model usage, hosting and licences, are shown on their own lines. Why we do not publish prices explains the choice, and what an AI consultant costs covers the wider question.
Your customers' language and local systems
Pick a local partner when your customers write in Estonian or your systems touch Estonian infrastructure. Otherwise a remote team can work, as long as it starts with an audit and leaves the accounts in your name.
Estonian is a small language. An AI assistant that answers fluently in an English demo can get case endings wrong in Estonian, misspell a product name or misread the question. You will not see that on a slide. You see it when a customer writes.
So ask for a live test in the language your customers use, not a recording. With our systems you see for yourself how it answers before launch, and you can steer it before it reaches your customers.
The same goes for your systems. An Estonian company runs on Estonian infrastructure: the business register, local accounting software, e-invoices, and for some owners a company set up through e-Residency. A partner who has connected these before does not have to discover them during your project.
Tallinn, Tartu and the rest of Estonia
Location matters less than it used to, but it still matters. The build is remote work. The audit gains from someone walking through your building and seeing where the hours go, which is easier within an hour's drive. We visit on site in Tallinn and Harju County and work remotely elsewhere.
EU rules that come with the build
Estonia is in the EU, so the EU AI Act applies to what you build. The European Commission's description (October 2026) lists four risk levels: unacceptable risk, high risk, transparency risk, and minimal or no risk. Transparency risk covers a chatbot, for example: people must know they are talking to a machine. The Commission expects the great majority of AI systems now used in the EU to fall into the last group. Recruitment, credit decisions and biometric identification are high-risk uses. You do not need to learn the regulation. You need a partner who tells you your risk level before the build, not after a complaint.
On the governance side, the NIST AI Risk Management Framework is a good starting point. It is voluntary and comes from the US, but it lays out clearly how to think about AI risk. A consultant who can name a framework has at least read one.
When not to hire anyone yet
An AI partner is worth it when the problem costs more than the fix. Otherwise you do not need to hire anyone yet, and a good partner will tell you so.
- If the process is broken, fix it first.
- If the problem is that people use the tools they already have in their own ways, training helps more than a build.
Whether outside help pays off in your case is the first question the free audit answers.
Check the work, not the pitch
A portfolio is easy to write and hard to prove. Three checks need no technical knowledge from you.
First, ask for the live thing, not a picture. If they sell an AI assistant, type one of your own customers' real questions into it yourself, right there at the table. If they sell automation, have them show the workflow on screen and ask what happens when one step fails. A working thing either exists or it does not. Anyone can make a slide.
Second, ask for a client's name and permission to call them. Not an anonymous quote from a leading company, but a name and a number. Some clients do not want their name public, and that is fair. Then the consultant has to be able to show the work itself.
Third, ask what went wrong last time. Real projects go wrong now and then. Anyone who says everything has always gone smoothly either has not delivered enough or is not telling the truth, and either way that is your risk.
And one thing you do not need to ask for: certificates. A certificate shows someone finished a course. A working system shows they can do the job. Our own work, with named systems and public links, is on our case studies page, so you can run the same checks on us.
Two of our public examples. For a tyre shop in Tartu we added 29,364 new products to its existing WooCommerce shop, plus search by number plate through the shop's own link to the state vehicle register. Stock and prices come in from the wholesaler every two hours (the RehvidTartus24 case study). For a corporate finance firm we set up a cold-email system in the firm's own accounts. The emails go out from sending domains in the firm's name (case study).
How we work
Audit first. In a free 30-minute audit, Karl looks with you at where your work loses time, money or information. The findings are yours even if you do not go further with us. If a build is justified, we write down the scope and build the specific fix: an automation, a systems integration, an AI assistant or a website, depending on what the audit points to. Then we run it with you, or hand it over with the code, accounts and documentation. The code, accounts and data stay with you. How this looks for companies here is on our page about AI consultancy in Estonia.
Frequently asked questions
How do I choose an AI agency?
Choose the one that looks at how your work runs before it names a price, puts the scope in writing, and leaves the code, accounts and data in your name. Ask which tools it uses and why, ask for a live test in your customers' language, and ask what happens when the model is wrong. An agency that says yes to everything is selling software, not judgement.
What should I ask an AI partner before signing?
Seven things: what it found before naming a price, which tools it uses and why, who owns the code and accounts, what happens when the model is wrong, what should not be automated, who actually does the work, and what month three after launch looks like.
What are the red flags of a bad AI provider?
A price before anyone has seen your work, the same product for every problem, accounts in the agency's name, only a recorded demo, and an accuracy percentage where an error plan should be. One of these is a reason to ask more. Several at once is a reason to walk away.
How much does an AI partner cost?
There is no single price. It depends on how many workflows change, which systems get connected, the state of your data, the languages involved and who maintains the system afterwards. Our audit is free. We write down the build scope and price after the audit, before any work starts.
What does the free AI audit decide?
It shows where your work loses time, money or information and what that costs you today, whether a build is worth it, and where to start. The findings stay with you even if you do not go further with us.
Should I pick a local partner or a remote team?
Location matters less than accountability, but in two cases it really matters. If your customers write and call in Estonian, a remote team will not notice where an answer breaks in the language. And if your systems touch Estonian invoicing, the business register or local accounting software, local experience saves time. Decide the rest by whether the partner starts with an audit and leaves the accounts in your name.
Does the EU AI Act apply to my company?
If you operate in the EU, yes, and the practical question is which risk level your use falls into. The European Commission's description (October 2026) lists four: unacceptable risk, high risk, transparency risk, and minimal or no risk. The Commission expects the great majority of AI systems now used in the EU to fall into the last group. Recruitment, credit decisions and biometric identification are high risk. A good partner tells you your level before the build.
