Short answer: what separates vendors isn't the technology, and not even the price. It's four things: whether they commit to a specific number before starting, whether they agree to start with one small process, whether they clearly say what the monthly fee includes, and whether they're able to say “this isn't worth doing for you” at all. A vendor who never says no is selling to you, not advising you.
The market is young, everyone uses the same vocabulary, and the proposals all look alike. So below you won't find general advice, but specific questions, along with what a good and a bad answer sound like.
Seven questions
- 1. What number does the system have to hit?
Good answer: “We process 85% of documents without a person, and response time drops to five minutes.” Bad answer: “We'll increase your efficiency.” If a vendor can't name a number, they can't prove it worked either, and you can't say it didn't. - 2. What happens if it doesn't work?
A good answer is specific: money back, fixed for free, contract ended without a penalty. A bad one: “That's never happened.” It has. To everyone. - 3. Which of my systems will it connect to, and how?
Ask them to name the systems and the method: through an API, through files, or by working on screen like an employee would. Both the price and the reliability depend on it. If the answer is vague at the proposal stage, it won't get clearer during the build. - 4. What exactly does the monthly fee include?
Servers, monitoring, updates, bug fixes, changes. Ask separately: are changes included, or billed by the hour? Cheap maintenance with expensive fixes often costs more than a pricier “everything included”. - 5. Who keeps the system and the data if we end the contract?
Rarely asked, and one of the most important questions. Can the data be exported in a normal format? Do you get the configuration and documentation so another vendor could take over? If the answer is foggy, you're building a dependency whose price you'll only learn when you want to leave. - 6. How fast do you fix it when it breaks?
Everything breaks. What matters isn't a promise that it won't, but the response time and who's on call. One person who goes on vacation in August isn't an answer. - 7. Where will I see what the system did?
There has to be an action log that you can see, not just the vendor. Without it, you won't know what's happening in your business, and you won't be able to check either the result or the bill.
The best test is simple: ask when automation isn't worth it in your case. Whoever has no answer hasn't worked with real processes yet.
Red flags
- A fixed price before looking at your processes. A number given on the first call is either a guess or has a buffer hidden in it, at your expense.
- Promises of 100% accuracy. With real documents and live customers, that doesn't happen. A serious vendor talks about how the system flags what it isn't sure about.
- The main selling point is the model's name. Which model is used will change within six months. The question isn't what it runs on, but what it does and with which of your systems.
- Won't agree to start small. Whoever demands the whole budget and the whole process at once shifts all the risk onto you.
- Doesn't ask about your exceptions. Whoever talks more than they ask on a call doesn't know yet what they're offering.
What you don't need to ask
A few things people usually ask that I'd consider secondary:
- Which model you use. It changes every few months and barely affects your result. What matters more is what happens to your data, and that's a GDPR question, not a technology one.
- How many clients you have. Thirty clients in other niches tell you nothing about your case. It's more useful to ask for one example that's similar in process, not in industry.
- Whether you've worked in my industry. Helpful, but not decisive. Entering invoices at a factory and at a beauty salon differ less than you'd think.
Local, foreign or a solo specialist
Three paths, and each has its price:
- A Lithuanian vendor. Same language, same time zone, understands local systems and how things are done here. The most practical option when the solution connects to what's used in Lithuania.
- A large foreign company. More resources and more expensive, but a small client there often ends up low priority. Worth it when the project is large and international.
- A solo specialist. The cheapest, and sometimes completely enough. But the key question for them isn't price, it's who maintains the system six months from now, when they have new projects or are simply on vacation.
How to compare proposals that can't be compared
The biggest practical headache: one quotes a fixed price, another an hourly rate, a third a monthly package. To compare them, bring everything down to one number: the cost per year.
Add the build, plus twelve months of maintenance, plus a rough estimate of changes over the year. Then the numbers fall into place, and it often turns out the cheapest proposal wasn't. How to calculate ROI from that number is in the ROI article, and market price ranges are in the pricing article.
And now about us
It wouldn't be fair to write this list and not check myself against it. So, briefly, here's how we answer these questions: we agree on the number before we start and work through a 2-week pilot; if we miss the target, I refund all the money; the price is up to $10 per hour of agent work, and you see the hours in the action log in your own system. Maintenance is included in that price, with no separate support fee.
And yes, we say “not worth it”. There are whole sections on that in the ROI and hiring comparison articles. If the numbers don't work in your case, you'll hear it on the first call, not three months in.
Questions about choosing an AI vendor
How do I choose an AI automation agency?
Choose based on whether the vendor commits to a specific number before starting, agrees to start with one small process, clearly explains what the monthly fee includes, and tells you when automation isn't worth it. A vendor who never says no is selling to you, not advising you. Price is secondary: the cheapest proposal without a measurable target usually ends up the most expensive.
What should I ask an AI automation vendor before signing?
Seven things: what number the system has to hit; what happens if it doesn't work; which of your systems it will connect to; what exactly the monthly fee includes; who keeps the system and the data if the contract ends; how fast it gets fixed when it breaks; and where you'll see what the system did.
Who keeps the system and data if I end the contract with an AI vendor?
That has to be written into the contract, and often isn't. Ask specifically: can the data be exported in a usable format, do you get the system's configuration and documentation, and can you hand them over to another vendor. If the answer is unclear, you're building a dependency whose price you'll only find out when you want to leave.
Should I choose a Lithuanian or a foreign AI automation vendor?
A Lithuanian vendor is more practical when the solution connects to local systems and when the same language and time zone matter. Large foreign companies are often more expensive, and a small client there ends up low priority. A solo specialist is the cheapest, but the key question for them isn't price, it's who maintains the system six months from now.