Finland’s Public-Sector AI Plan Puts Replacement Before Public Value
AI could improve public services and help Finland respond to labour shortages and pressure on public finances. But replacing employees should not become the goal before the country defines the public value it expects AI to create.
Text: Dr Umair Ali Khan, 23.7.2026 | Photo by Adobe Stock Photos
“AI will essentially replace humans.”
That is how Juha Majanen, Permanent Secretary at Finland’s Ministry of Finance, described the direction he expects Finland’s public sector to take. In an interview with Helsingin Sanomat in June 2026, as reported by Helsinki Times and Yle, Majanen said that the state, municipalities, and wellbeing services counties would operate on an AI basis by 2031 (Helsingin Sanomat, 2026; Helsinki Times, 2026; Yle News, 2026).
The Ministry estimates that the wider reform could improve public-sector productivity by at least 20 percent. Some staff reductions would happen through natural attrition, as retiring employees would not be replaced. Majanen also acknowledged that many public-sector employees could lose their jobs during the transition.
There are good reasons for Finland to explore greater use of AI. The population is ageing, public finances are under pressure, and some public services face labour shortages. AI could reduce administrative work, improve access to information, and help employees complete routine tasks more efficiently.
However, the current proposal remains a high-level ambition rather than a detailed implementation programme. It is not yet clear which services would use AI, which decisions would remain with human experts, how performance would be measured, or how citizens could challenge incorrect decisions.
This article examines both the opportunities and risks in Finland’s plan. It considers why replacing employees is the wrong starting point, where AI can create real value, why some public-sector decisions cannot be safely delegated, and why saving time or money does not automatically improve public services.
Human Replacement Could Be the Wrong Starting Point
The main question should not be how many public employees AI can replace. It should be: what public problem are we trying to solve?
Only after answering that question should the government decide whether the best solution is automation, better digital tools for employees, redesigned processes, additional training, or some combination of these options.
Beginning with human replacement reverses this process. It makes staff reduction appear to be the objective before the technical feasibility, total cost, service quality, legal responsibility, and effects on citizens have been properly assessed.
The timing also makes this language more sensitive. Statistics Finland reported 376,000 unemployed people in May 2026, which was 68,000 more than one year earlier. The non-seasonally adjusted unemployment rate was 12.7 percent, the highest recorded since 1998 (Statistics Finland, 2026). May is often a difficult month for employment figures because students and recent graduates enter the labour market, but the figures still show the weakness of the employment situation.
Statements by senior officials influence how employees, companies, and citizens understand AI adoption. When a senior official says that AI will replace humans, public employees may see the reform as a threat rather than an effort to improve services. Private companies may also begin to treat layoffs as proof that their AI investments are working, even when they have not established whether the technology improves the full workflow or creates measurable value.
That is the opposite of the discipline required for a serious AI transition.
AI Adoption Does Not Automatically Create Value
Evidence from Nordic organizations suggests that AI use does not automatically produce broader value. Many organizations achieve limited efficiency gains from individual applications but struggle to scale them or translate them into broader organizational and financial outcomes (Deloitte, 2026; McKinsey & Company, 2025).
Our earlier analysis of Finnish organizations found the same gap between adoption and value. Many companies are experimenting with AI, but local productivity improvements often remain isolated. Real value requires process redesign, capable employees, suitable data, clear ownership, integration with existing systems, and measures that connect AI use with organizational outcomes (Khan & Asikainen, 2026a, 2026b).
Deploying AI does not automatically make an organization better. An AI system may save time in one department but create more checking work in another. It may speed up document production while increasing complaints and corrections. It may reduce the workload of one team while creating additional legal, technical, or administrative responsibilities elsewhere.
A public body that treats staff reductions as the main measure of success may later discover that the expertise it removed is more difficult to replace than the software it purchased.
Where AI Will Create Real Public Value
This is not an argument against public-sector AI.
In our AI consultancy work with Finnish organizations through the Finnish AI Region project, we have seen that the strongest AI projects do not begin with a target for reducing staff. They begin with a clearly defined problem, suitable data, realistic expectations, and measurable value.
In many cases, the most useful systems support employees with repetitive, deterministic, and information-heavy work while keeping human expertise and accountability in place.
AI can help citizens find the correct service using information from approved public sources. It can already point citizens to the right service using approved public information, transcribe meetings and prepare summaries, and draft long-form routine reports. It can pull the required information out of documents, invoices, applications, and supporting documents, flag when something required is missing, classify incoming cases, and check documents against organizational policies, to name a few.
These are generally good starting points because the tasks are bounded, repetitive, and relatively easy to verify. Errors can often be corrected before they cause serious harm.
In these cases, AI supports public employees rather than replacing their judgment. The risk increases when AI moves from supporting administrative work to influencing decisions about people’s rights, income, healthcare, housing, or access to essential services.
Where AI Will Become Risky
AI can process large amounts of information and identify patterns, but access to information is not the same as understanding a real-life situation.
Consider a customer-service conversation. A language model knows recognized communication practices, but an experienced professional also considers the customer’s earlier complaints, current emotional state, level of knowledge, financial limitations, tone, hesitation, cultural background, previous commitments, and possible long-term consequences.
AI may produce a plausible answer, but a plausible answer is not always the right answer.
A social-benefit application may appear routine until one sentence in an attachment reveals a medical emergency, domestic abuse, an unusual family situation, or an immediate risk of homelessness. An AI system may apply the visible rules correctly but still reach the wrong conclusion because an important fact is unclear, hidden, contradictory, or missing.
This is why the EU AI Act classifies certain systems used to assess access to essential public benefits and services as high-risk (European Union, 2024). These decisions can affect livelihoods, legal rights, and fundamental freedoms.
Keeping humans involved appears to offer a solution. However, human oversight introduces another question: does the combined AI-human workflow actually improve the service?
Human Oversight Can Reduce Risk, but Also Reduce Value
Keeping a human in the loop to validate and correct AI output is a reasonable approach. However, it works only when that person has the time, knowledge, authority, and confidence to challenge the system.
Suppose AI prepares a draft decision for a benefits application. A caseworker may still need to reread the application, inspect every attachment, verify the extracted facts, repeat the eligibility calculation, check the legal reasoning, and correct the final decision.
If this review takes almost as long as preparing the decision without AI, the system has not created meaningful efficiency. It has added software, monitoring, and integration costs while keeping most of the original human workload.
AI can also be used to check another AI system’s output. This is commonly described as the “LLM-as-a-judge” approach. A second AI system can identify contradictions, flag unsupported claims, estimate confidence, and send uncertain cases to human experts.
However, AI can also make mistakes when checking AI. The second system may repeat the first system’s error, overlook missing information, or confidently approve an incorrect result.
Another option is to let AI process simple cases and send complex cases to humans. But this is easier said than done. The system must first recognize that a case is genuinely complex. The decision about whether a case is simple or complex can therefore be as important as the final decision itself.
Human oversight can reduce risk, but it does not automatically make an AI workflow accurate, efficient, or economical.
Public Value Must Be the Measure
Time savings, cost savings, efficiency, and improved outcomes are different measures.
A system may save employee time while increasing technology and review costs. It may reduce staffing costs while slowing the overall service because exceptions, complaints, and appeals accumulate. It may save both time and money while producing more incorrect or unfair decisions.
Even if an AI-human workflow reduces processing time by, let’s say, 50 percent, the public organization must still pay for procurement, cloud and model use, software integration, data preparation, employee training, privacy protection, cybersecurity, legal compliance, monitoring, auditing, maintenance, error correction, and appeals.
The correct measure is therefore the total cost of producing an accurate, fair, timely, accessible, and legally defensible result.
Public value is broader than financial savings. It includes trust, equal treatment, service accessibility, citizen wellbeing, employee workload, and the ability to identify and correct harmful decisions.
Headcount may be one budget indicator, but it should never become the definition of progress.
A Better AI Adoption Path
Public sector AI adoption requires a very careful and responsible approach. Each proposed use should clearly state the problem it solves, the value it is expected to create, the risks it introduces, and the evidence that will show whether it works.
Employees, unions, service users, legal experts, and technical specialists should be involved before deployment, not consulted after the main decisions have already been made.
Bounded, low-risk, and easily checked tasks should come first. Decisions affecting rights and welfare should retain real human authority, not simply a human signature added after the system has already shaped the outcome.
Automate what AI can do reliably, support employees where technology can improve their work, and keep people where judgment and accountability still matter.
The goal should not be an AI-based public sector with fewer humans. It should be a better public sector that uses AI where it creates proven public value.
Author
References
Deloitte. (2026, April 8). State of AI in the Nordics 2026. https://www.deloitte.com/fi/fi/Industries/technology/research/state-of-ai-in-the-nordics.html
European Union. (2024). Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence. Official Journal of the European Union. https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng
Helsingin Sanomat. (2026, June 26). Tämä on painajaisunta, ajatteli Suomen vaikutusvaltaisin virkamies talvella. https://www.hs.fi/politiikka/art-2000012081842.html
Helsinki Times. (2026, June 26). HS: Finland plans AI overhaul of public sector by 2031. https://www.helsinkitimes.fi/finland/finland-news/domestic/28980-hs-finland-plans-ai-overhaul-of-public-sector-by-2031.html
Khan, U. A., & Asikainen, M. (2026a, May 27). Finland has moved beyond AI curiosity. Now it must prove it can scale. Finnish AI Region. https://www.fairedih.fi/en/2026/05/27/finland-has-moved-beyond-ai-curiosity-now-it-must-prove-it-can-scale/
Khan, U. A., & Asikainen, M. (2026b, May 27). Why AI adoption in Finland isn’t becoming business value. Finnish AI Region. https://www.fairedih.fi/en/2026/05/27/why-ai-adoption-in-finland-isnt-becoming-business-value/
McKinsey & Company. (2025, November 5). The state of AI in 2025: Agents, innovation, and transformation. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
Statistics Finland. (2026, June 24). Fewer employed persons and more unemployed persons in May 2026 compared to one year ago. https://stat.fi/en/publication/cmfp8w31ba0n707vyezbdyaz4
Yle News. (2026, June 26). Finnish public sector’s move to AI will mean job losses, top civil servant tells HS. https://yle.fi/a/74-20233641
