The short answer
In a small company, AI helps most in four places: drafting and summarising text, answering questions from your own documents and data, sorting incoming emails and requests, and reading data out of documents such as invoices and orders. It helps least where the company's data is scattered across disconnected tools, because an assistant can only be as good as the data it can reach.
Most companies have not started yet. In 2025, 12% of Latvian enterprises with ten or more employees used at least one AI technology, against 20% across the EU and 42% in Denmark (Eurostat). Among Latvian companies with 10 to 49 people it was under 10%. Here is how to start in a way that pays off and stays within the rules.
Start from tasks, not tools
Do not start by choosing an AI product. Start by listing the work where people read or write a lot: answering similar customer questions, writing quotes and summaries, searching old emails and documents, typing data from PDFs. Those are the tasks where today's AI is reliably useful.
The same scoring as for any automation applies: how often, how many minutes, how costly a mistake is. What should a small company automate first? shows the method.
Four uses that work
Drafting and summarising. First drafts of emails, quotes, job adverts and meeting notes, always checked by a person. In a study of 5,179 customer support agents (Brynjolfsson, Li and Raymond, NBER), an AI assistant raised the number of issues resolved per hour by 14% on average and by 34% for new and less experienced staff.
Answers from your own data. An assistant that answers “which customers haven't ordered since spring?” or “what did we agree with this supplier?” from your CRM, accounting and documents, with links to the sources. This is where the value is, and it needs the data connected first.
Sorting incoming work. Emails and form requests are read, labelled (order, complaint, quote request) and routed to the right person or system.
Reading documents. Invoices, orders and delivery notes are turned into structured data and checked against what you expected, instead of being retyped.
Get the data in order first
Most AI projects that stall do so on data, not on the AI. Across the EU, companies that considered AI but did not use it most often named a lack of expertise (71%), unclear legal consequences (53%) and data protection concerns (49%), according to Eurostat. Latvian companies also often cite data quality and incompatibility with their existing systems.
The order that works: map the process, connect the systems so the data is in one place, then put a dashboard or an AI assistant on top. Skipping the first two steps gives you a confident assistant that answers from half the facts.
Set rules for company and customer data
Latvia's Data State Inspectorate (DVI) recommends, as reported by iFinanses, first asking whether personal data is needed for the task at all, using anonymised or general information where possible, choosing providers you can trust and checking their data terms, and setting internal rules and training staff. Do not paste employment contracts, personal ID codes, client contact lists, passport copies or trade secrets into public AI tools.
In practice, three rules cover most of it. Use business accounts with a data processing agreement, not personal ones. Prefer providers that keep data in the EU. And write a one-page AI policy that says which tools are allowed and which data never goes into them.
Know what the EU AI Act asks of you
A company that uses AI tools is a “deployer” under the EU AI Act. For a typical small company the obligations are light, but there are some, according to the Future of Privacy Forum's timeline of the act as amended in 2026.
Simplified. Check the official texts or a lawyer before using AI in decisions about people.
Keep a person in the loop and measure
AI makes mistakes, and people trust it too easily. In KPMG's 2025 study on trust in AI, 53% of respondents in Latvia admitted making mistakes at work because of AI, and only 24% were aware of existing AI rules. Keep a person responsible for every output that reaches a customer or the books, and measure the hours saved after a month, as you would with any other automation.
If the project is larger, the LIAA digitalisation programme has offered co-funding of up to €200,000 for AI solution projects. Check whether a call is open before you plan around it.
Checklist: AI in a small company
The tasks with the most reading and writing are listed and scored.
The data an assistant would need is connected and in one place.
Only business accounts with a data processing agreement are used.
A one-page AI policy says which tools are allowed and which data stays out.
Nobody uses AI to make decisions about hiring or credit without legal advice.
A person checks every AI output that reaches a customer or the books.
Hours saved are measured after the first month.
