Datapraxis
Tallinn office desk under cold winter window light, unposed workspace on an ordinary afternoon.

AI in Estonia · Briefing no. 04

What is Estonia actually doing with AI?

Digital state, national programmes, and a small but very real B2B ecosystem. Read the map before you plan your own pilot.

Datapraxis OÜ · Tallinn

Reader's load path

  1. 01 Where Estonia already runs AI
  2. 02 National programme and public assistants
  3. 03 You are here — local products by job
  4. 04 Advisory, voice, listings, feedback
  5. 05 Governance as its own layer

01 · Context

The national programme, in one paragraph

A clerk and a citizen at a public service counter with a screen showing a simple line-drawn map.

Tallinn, service desk · public-sector AI reaches residents through counters like this one.

The Eesti.ai initiative is a nationally led programme developed with international experts and businesses. It aims to systematically implement AI across sectors, with focus on skills, new solutions, and export capacity. It links to high-impact pilots in healthcare, education, industry, energy, and the public sector. Think of it as a countrywide productivity strategy rather than a single tool.

Datapraxis OÜ follows it as context for readers planning pilots. That is a deliberate distance: the programme sets the direction, but the useful work happens inside a department that has a real queue, a real reviewer, and a real set of forms to process.

What it is
A coordinated national programme, not a product.
Where it lands
Pilots in healthcare, education, industry, energy, public services.
What it asks of you
Name a scope, a data source and an owner before you start.

02 · The assistant pattern

Public services and the assistant pattern

Two programmes matter here because they set what residents and staff already expect from an AI answer. If your team serves the public, these are the expectations you inherit.

Programme

What it does

What it implies for teams

Bürokratt

Estonia's virtual assistant concept for public services, built around finding information and guiding users through steps.

Residents get used to asking in plain language and expect an answer, not a form.

AI Leap

The education programme bringing AI practice into Estonian schools, with teachers and students working through real tasks.

Staff get used to checking an AI draft before it goes anywhere.

Where responsibility stays

Both programmes keep human review inside the loop. Final answers and grades belong to people, not to the assistant.

Plan a clear escalation path before you point a resident at an assistant.

Read the pattern, not the branding: a plain-language front door, a small number of guided steps, and a person who owns the outcome. That is what public AI adoption looks like when it survives contact with a queue.

03 · Platforms

Support platforms turning into marketplaces

Intelswift provides AI-powered customer service automation software for help desk, insights, and knowledge base workflows. It includes an AI agents marketplace with chat and voice agents and supports no-code building of custom AI workflows.

The angle for Estonia

Small teams, assembled workflows

A no-code path means a three-person support desk can assemble a workflow without commissioning a build. That is the whole point in a market the size of Estonia.

The trade-off

Articles and rules need upkeep

A knowledge base degrades quietly. Someone has to own the articles, retire the stale ones, and watch the escalation rate when the assistant cannot answer.

Before you buy a marketplace part

Local review first

Marketplace agents still handle your customers and your data. Read the terms the way you would read any vendor contract, and test the escalation before you go live.

04 · Advisory

Advisory and build capacity in a small market

MindTitan appears in Estonia AI consultant rankings and focuses on AI strategy consulting and machine learning consulting. In a small market, the shape of the engagement matters more than the size of the consultancy.

Use the checklist below when you speak to any partner — including us. Short scope, named reviewer, written review points. A vendor who cannot answer those three is not ready for your pilot.

  1. 01

    A short scope on one page

    One process, one expected output, one measure of done. If the proposal needs ten pages to explain, the pilot is not scoped yet.

  2. 02

    A named reviewer inside your team

    One person who reads the output daily and can say stop. That person is who the pilot serves — not the vendor's slide deck.

  3. 03

    Written stop conditions

    Named data sources on one side, named limits on the other. What the pilot will not attempt is as important as what it will.

This is the same checklist we owe clients. It is also the easiest way to compare two proposals without reading either one twice.

Tallinn rooftops and a modern office block at evening with one figure crossing a stone street.

05 · Startup shapes

Voice, listings, and feedback — three shapes, one shared need

Three Estonian-built products, three different jobs — and the same two prerequisites before any of them works in your organisation.

Hospitality

SVARA AI

Multilingual voice agents that answer calls, handle guest questions, check availability, manage reservations, and transfer complex cases to staff.

Second-hand goods

Auctologue

Aggregates and organises research for antiques and second-hand goods, analyses photographs, and creates search-friendly listings.

Product feedback

Mergo

Combines feedback with product analytics and groups tickets and surveys into themes linked to usage.

What the three share

Clean source material, and a person who approves the output. None of the three works if the underlying data is a mess, and none of them should be the last checkpoint before a customer.

06 · Governed work

Governed work and reliable knowledge

Two more Estonian products show where AI adoption is heading: away from single bots and toward operating layers with oversight built in.

Brida

Turns recurring processes into governed AI work with packs of agents, tools, context, workflows, policies, and approvals — separating execution from governance.

The separation matters: the run and the approval are two different records.

Atlaswiki

Structures company knowledge into attributable facts, flags contradictions, and supports reviews for AI use.

Attributable means you can point at the source of a claim, not just the claim.

Together they show a shift from single bots to operating layers with oversight. For teams handling andmekaitse questions, this separation helps: who ran the step, who approved it, and which fact was used.

07 · Questions

Common questions about AI in Estonia

Short answers for the three questions visitors send us most often about the Estonian context.

Ask about a use case
How is the Estonian government using AI in public services?
Through information assistants, guided steps for services, and sector pilots linked to the national programme. Staff remain responsible for decisions. Expect plain-language access with escalation to people.
Which AI use cases matter most for healthcare, education, and industry?
Healthcare and education pilots focus on information handling and reviewable drafts; industry focuses on repeated checks and documentation. All need clear data scope and human approval for consequential steps.
What does the Eesti.ai initiative aim to change?
Systematic adoption across sectors through skills, new solutions, and export capacity, with high-impact pilots in several sectors. It is a strategy and coordination effort, not a single product.

Start where the queue is

Take one process, one reviewer, four weeks.

If your team already loses time to the same queue every week, that is a pilot worth scoping. Bring the process — we will bring the checklist above.

Reference half

Datapraxis OÜ · Pärnu mnt 22, Kesklinna linnaosa, Tallinn, Harju maakond, 10141 · Mon–Fri, 9:00–18:00

Programme and product descriptions on this page stay within their published scope; specifics of any pilot depend on your data, your team and your sector.