Sofokus AI Maturity Model

Sofokus AI Maturity Model

The Sofokus AI Maturity Model measures an organisation’s level of AI maturity on a four-level scale. The levels of the model are ungoverned, individual, process and organisation. The model is based on observation, because research shows that people’s perception of their own behaviour is systematically inaccurate (chapter 4.2).^[1]

The model also sets out to solve a second challenge that is characteristic of AI. The market situation and AI technology move so fast that an assessment tied to the market situation would be out of date within months. Concrete observations about your own organisation stay true regardless of the market situation. The assessment can be repeated whenever you want, so that you can follow how AI develops.

The model looks at the organisation from several different perspectives to narrow assessment bias, though such bias naturally cannot be eliminated entirely. Honest observation of the indicators, grounded in the real situation, matters more in the end.

An assessment model that rests on concrete observations has one further benefit. If for example the owners, the directors or the executive team hold differing views about their own organisation’s AI maturity, reality can be checked against the indicators.

Change begins when an organisation’s areas for development take concrete shape.

Principles of the maturity model

The levels of the model are based on observable indicators: documented guidelines, completed purchases, named responsible individuals and reported metrics. Every indicator can be verified in everyday work without ambiguous scoring or staff surveys.

The model does not measure technological capability or the number of AI tools in use. Technology dates quickly, and processes, the way work is organised and clear responsibilities are critical in turning AI into added value.

In the model, the maturity level rises as the use of AI moves from the actions of individual people into the structures of the organisation: from an individual’s choice to a shared guideline, from a guideline to a process, and from a process to part of the management system.

What kind of organisation is the model for?

The model can in principle be applied to any organisation, but in practice it is not necessarily optimal for the very smallest and the very largest. In smaller organisations, processes and guidance are often incomplete simply because the work gets done without them. In the largest organisations, headcount and the complexity of structures have grown so great that a broader set of questions is needed, and McKinsey’s maturity model, for example, is suited to that.

Most organisations, however, fall in between. They have substantial operations, staff, processes and other structures, and for them the maturity model offers useful information on their own AI transformation journey.

AI maturity levels

The AI maturity model is presented below as a diagram.

A zero level has deliberately been left out of the model. Employees start using AI on their own initiative regardless of what the organisation has decided on the matter.^[2] An organisation that believes it is at level 0 is more likely already at level 1 without knowing it.

The level descriptions in the maturity model set out the characteristics, risks, benefits and costs.

Level 1: Ungoverned

Area	Typical indicator
AI leadership	No one responsible. No documented information on the use of AI.
Ground rules and documentation	No written guidance.
Tools and accounts	Personal or free accounts are in use. The company pays for individual pilot licences at most.
Skills development	The development of AI skills is individual-led; the organisation does not steer it.
Processes	The work steps are unchanged, because there is not yet even a plan to make them AI-based.
Agents and automation	Agents are not used, or they are used on people’s own initiative without centralised access to data and systems.
Customer relationship	The organisation neither knows nor limits which tools are used to process customer material.
Budgeting	At most a pilot budget has been set aside for AI tools.
Measurement and value creation	No measurement, and value is visible at most in an individual’s own work, not to the organisation.
Brief summary	AI is used, but the organisation has no reliable information on how and with which tools it is used. Use rests mostly on the initiative of individual employees on their own or free accounts. Company and customer data may have ended up in public tools.

An organisation at this level already uses AI, but an organisation-wide picture is missing. No one knows for certain in which tasks AI is used, who uses it and with which tools. The benefits stay at the level of the individual.

This maturity level is a typical starting point for Finnish companies. According to Statistics Finland’s official November 2025 statistics, 38% of companies employing at least 10 people used AI in spring 2025, but only 15% had documented guidelines.^[3] Among companies with 50–99 employees, use was 47%, but documented guidelines only 25%.^[3]

In an international survey by the University of Melbourne and KPMG, 56% had used AI in their work without knowing whether it was permitted, and 48% had uploaded company information to a public AI service.^[4]

AI may be used on free or personal accounts, and there is no centralised management. At this level the risk is permanent, because the organisation does not yet have a clear, shared line on what information may be fed into AI applications. AI vendors may use the data entered by staff to train their models, because consumer accounts are in use.

Costs are mainly licence and working-time costs. Budgeting for AI tools is at pilot level.

It is not worth staying at this level. The target should be at least the next level.

Level 2: Individual

Area	Typical indicator
AI leadership	The objective has been written down, and a named representative of the Owners, Directors or Executives oversees progress at recurring points in time.
Ground rules and documentation	Written guidance is part of induction.
Tools and accounts	Licences for AI tools have been budgeted for a whole role group (e.g. sales), and the accounts are under the organisation’s control.
Skills development	The development of AI skills is organisation-led, recurring and part of a regular training programme.
Processes	AI tools are in use, but the work steps or their order have not yet been changed.
Agents and automation	Agents work as an individual’s aids on company accounts, and a person is responsible for the output.
Customer relationship	There is an internal policy on using customer material in AI tools, but the use has not been agreed in writing with customers.
Budgeting	The budget takes licence costs into account. Training, measurement or process development for the use of AI have not been taken into account.
Measurement and value creation	Surveys or feedback are collected, but value creation is measured neither before deployment nor after it. Value is visible, for example, as estimates of time saved in staff surveys, but not in the organisation’s result figures.
Brief summary	The organisation governs the responsible use of AI with tools, guidance and training. The benefit remains time saved by individuals, because the work is still done with the old processes.

At the second level, an organisation has taken control of how AI is used. Tool choices have been made, shared ground rules have been documented and staff are trained to use AI. The objective has been written down, and a named representative of the Owners, Directors or Executives oversees progress at recurring points in time.

Rising to this level reduces the risks described at the previous level, because there is an active effort to guide, support and encourage staff in the responsible use of AI. On the other hand, use at the individual level is likely to increase, which raises the need for management resourcing. Suitable tool licences manage the previous level’s risk of model training, but as regards customer data the risks typically still remain.

The benefits materialise mostly as time saved by individuals, because the work is largely done with much the same processes as before. In the terms used in chapter 3.1, the applications in use are probably closer to task automation than to AI transformation. The laborious phase of change, the redesign of processes, is still undone. Two things are likely to follow from this: the time individuals save is not systematically directed to other use, and any time savings do not create organisationally measurable value (compare EBIT, for example).

A specialist can adopt a generative AI tool on their own initiative in the morning and get benefit from it the same day. The low threshold for adoption is at the same time a challenge. When adoption does not force a change in the way work is done, the workflow stays as it was.

In a Procter & Gamble field experiment, 776 R&D professionals worked on the company’s real product development tasks in the following ways: alone, alone with AI, as a human pair, or as a pair with AI. An individual working with AI performed better (+0.37 standard deviations) than a pair working without AI (+0.24).^[5]

The key finding was that the individuals using AI were able to cross their accustomed boundaries: commercial specialists produced technically usable solutions and vice versa.^[5] AI therefore did not merely speed things up, but made part of the coordination work unnecessary.

The best result in the experiment came from a pair that used AI.^[5] That study therefore does not support the reading that AI would replace a human team. Instead, it appears to show that AI changes what is required of a team and may make it possible to improve quality.

Costs are mainly licence and working-time costs, and they have been budgeted for.

This level can be regarded as a kind of starting level for an organisation’s AI transformation. The foundation has been laid and the actual work of AI transformation can begin. The organisation has the chance to develop into an effective user of AI.

Level 3: Process

Area	Typical indicator
AI leadership	The owners of AI-transformed processes are responsible for use, measurement and the outcome.
Ground rules and documentation	The AI step is documented in the process description or in the induction of a new employee.
Tools and accounts	AI tools are a standard part of the process (cf. an individual’s choice).
Skills development	In addition to general AI skills, the organisation develops the skills of the people who take part in AI-transformed processes.
Processes	At least two processes contain an AI step whose omission would cause a deviation. At least one of the processes generates customer billing or serves the target group.
Agents and automation	Agents routinely carry out work steps in production, and the process owner is responsible for how they operate.
Customer relationship	The use of AI tools on customer material has been agreed in writing in the standard terms of customer agreements.
Budgeting	In addition to licences, the budget separately takes into account training, measurement or process development for the use of AI.
Measurement and value creation	The baseline of the process is recorded before deployment. After completion, the improvement against the target is measured, as is the fact that the outcome did not suffer in other respects. Value is visible in the measurements of AI-transformed processes (cf. lead time, error counts and so on).
Brief summary	AI has been built into at least a few processes in such a way that bypassing its use would be an error. The impact is monitored with process metrics, and agents carry out work steps in production. In AI-transformed processes, growth requires less human work than before. The impact on results cannot yet be demonstrated at the level of the financial statements.

At the third level, an organisation has made AI an integral part of its business processes. The organisation already has a good grasp of what AI transformation means in practice and is able to redesign its processes with AI in the lead. Customer agreements contain a standard clause for the data processing that AI requires.

Across a US company dataset, business success correlated with how extensively AI had been integrated into the organisation’s operations.^[2]

AI-transformed process stages cannot be bypassed, because they would cause a deviation. The effects of AI are measured and there are agentic work steps in production. Note that the starting point is not to automate processes completely with agents, but their use has at least been carefully assessed.

A process change typically brings several agents into the workflow, along with fixed rules and people who review and handle exceptions. ABC Legal describes running over fifty agents in production and most of them have been built by the business’s own specialists without a software development background.^[6]

A checking agent reached the same outcome as the company’s own team of specialists in 98% of cases, the company says. The company also tracks the value the agents produce against their costs. The firm says the agents were unprofitable at first, and that profitability followed only once suitable guidance had been found.^[6] Note that the above is the company’s own account and no independent measurement of it has been made.

In another experiment, 43 companies delegated the job interview to a voice-based agent. There were 12% more accepted job offers and employee retention improved by 17–18% with no decline in quality.^[7]

Applicants were asked the same things as before. The change was in the AI transformation of the process. Human bias fell away and the hiring manager’s calendar was no longer a bottleneck. The researchers’ view was that the key mechanism of impact was the reduction in process variation.^[7]

Few organisations reach the process level. The reason is that the AI transformation of an existing organisation realistically takes years. In addition, taking agents into production creates a new kind of work in practice, and calls for oversight and new skills from executives and staff.^[8]

At this level the cost structure moves from licences towards oversight and process development work. The focus of risk management moves from data leaks to making sure the agents operate correctly.

Research on complementarity shows that the full package produces more than its parts separately,^[9] but it also warns that changing a few elements at a time can fall far short of the total benefit.^[10] Nor does the benefit show immediately. Adoption follows a J-curve in which performance first falls and rises only after organisational investments.^[11]

Although the benefits of this level arise from the combined effect of AI, the reorganisation of work and measurement, in practice a successful change in organisational culture is a requirement.

Reaching this level, then, requires enough time and above all a cultural change in the organisation.

Level 4: Organisation

Area	Typical indicator
AI leadership	Executives steer the use of AI as part of their investment portfolio.
Ground rules and documentation	Every AI initiative includes a business rationale and stopping criteria agreed in advance.
Tools and accounts	Costs are allocated separately to each process or initiative (cf. a single cost item).
Skills development	The majority of investment is directed at people and at developing and governing processes, not at tools.
Processes	The organisation’s workflows have been thoroughly redesigned to make use of AI.
Agents and automation	Agents are scaled systematically. Value and costs are monitored.
Customer relationship	The organisation’s AI transformation has changed what the customer pays for, not just how the work is done.
Budgeting	AI is an investment area of its own, and its size is decided as a strategic commitment relative to revenue or an equivalent total budget.
Measurement and value creation	Measurement is a permanent practice and part of standard reporting. Value is visible in ordinary business metrics (cf. margin or customer retention, for example).
Brief summary	Growth requires proportionally less human work than before the AI transformation. The reference point is the organisation’s own starting position (vs. the scalability of a SaaS product, for example). The impact of AI is reported as a business figure as part of standard reporting. Executives steer AI as an investment portfolio, and AI has changed the revenue model of at least one service. The level is a state to be sustained, not an endpoint.

At the organisation level, the added value produced by AI runs through the whole operation. Every AI initiative includes a business rationale and stopping criteria agreed in advance. The impact of AI is monitored with business metrics as part of normal reporting.

One of the indicators of this level is that the revenue model of at least one service has changed because of AI. Billing may, for example, be based on the value produced rather than on hours.

The organisation’s output capacity does not grow in direct proportion to headcount alone. In other words, an organisation that has reached a high level of AI maturity can probably scale its operations by increasing the number of agents alongside recruitment. The benefit this brings depends on the type of organisation: for a SaaS company it is more common than for a people-led service business.

Measuring the value gained from AI is a permanent practice and part of standard business reporting. Every AI solution has an owner and a budget within the organisation, and costs are allocated separately to each initiative.

In February 2024, Klarna’s AI assistant handled two thirds of customer service contacts during its first month. Resolution time fell from 11 minutes to under two minutes. Repeat contacts fell by 25%.^[12]

In May 2025, Klarna’s CEO Siemiatkowski told Bloomberg that the quality of the service had suffered because of an excessive focus on costs. The company ended up recruiting more people into customer service, even though use of the AI assistant continued.^[13] The lesson of the story was that, alongside measuring time, measuring quality is central to AI transformation.

For an organisation operating at this level, the emphasis in risk management shifts to business continuity. Service outages and variations in quality show up directly in revenue once the revenue model is tied to the use of AI. In risk management, control of supplier dependencies, among other things, becomes central.

An organisation operating at this level has a well-informed understanding of the use of AI, particularly at the level of value creation. Does AI expand the business model, or does it make current operations more efficient? Does profitability improve by scaling an AI solution?^[14]

The level is highly demanding. In McKinsey’s data, only about 6% of companies achieve an impact of at least five per cent on their result.^[8] Companies expect AI investment to grow from about 0.8 per cent to 1.7 per cent of revenue per year.^[15] This is an international sample in which more than a third of respondents represent organisations with revenue of over one billion dollars, so the level describes the international leading edge.^[8]

Choosing a suitable target level for your organisation

For most organisations, the highest level is therefore likely to be more of a direction than a destination. The majority of Finnish organisations will probably settle at level one or two. In the digital maturity model published by the author in 2020, the target level is set organisation by organisation. A suitable target level is determined by strategy, industry and business goals.^[16] This guide recommends the same approach for AI transformation.

In McKinsey’s AI survey published in August 2026, almost ¾ of high-performing organisations said they had thoroughly changed their workflows with the help of AI. A year earlier the share was 55%. In other organisations the corresponding share was a quarter.^[8]

According to the source, workflows are specifically redesigned (AI transformation) rather than merely adopted.^[8] The survey does not prove a causal relationship, but it does show that simply introducing an AI tool into an old process does not yet build a lead.

The lower levels are not a sign that an organisation has fallen behind, as the reference point of the maturity model is the international leading edge. In practice, choosing this level also commits the organisation to continuous upkeep: overseeing agents, following metrics and updating processes.

In ServiceNow’s index, the average maturity score of companies fell in a year from 44 to 35, because the assessment criteria tightened as the field developed.^[17] It is worth noting here that ServiceNow’s index points fell because its index is tied to the market.

The AI transformation of an organisation is a learning journey with no end point. There are several significant milestones along the way.

From AI user to AI developer

There is a small group of organisations in the world whose use of AI is in a class of its own. In them, agents do not just run processes but develop the product and even the organisation itself. How would a café sound where an AI agent carried out the recruitment and handled the rest of running the café as well? The Swedish Andon Café says it operates under the direction of an autonomous agent called Mona.^[28]

Google DeepMind’s AlphaEvolve is a coding agent that runs on Gemini models. The agent sped up a computation kernel used in Gemini’s training by 23% and cut training time by one per cent. In other words, Gemini improves its organisation’s ability to produce the next versions of itself.^[18] In the same way, Fin (formerly Intercom) trained its customer service agents on the billions of customer service conversations its own AI had handled.^[19]

Why is a self-developing organisation not level 5 in the maturity model?

In the publication “Digitalize Your Operations”, the highest maturity level was based on network effects. A good user experience attracts more users, the users generate more data, and the data keeps improving the service.^[16] In Fin’s case the loop is similar to the one platform economy players have.^[19] At DeepMind the loop is shorter, because it does not pass through the customer. When the product can develop its owner organisation itself, the lead grows even if the competitive field keeps developing.

A “self-developing organisation” does not, however, form a fifth level in the maturity model. The reason is that levels 1–4 of the model are cumulative and that developing your own AI model requires none of the things those levels call for. Put bluntly, an AI lab can be developing a frontier model at the same time as its own governance sits at the level of ungoverned use. This is a strategic “buy or build” decision, which an organisation can make regardless of its maturity level.

AI-nativeness

An organisation can be an AI-native organisation by origin. That means its operations and processes are designed on AI’s terms from the very start (“AI-first”).

In AI transformation, an organisation that is already up and running changes its ways of working step by step (chapter 3.1). The AI maturity level describes how far an organisation has got on its own AI transformation journey.

Note that an AI-native organisation may be at the second maturity level and a 25-year-old company at the fourth, even though the latter was not an AI-native organisation to begin with. The starting point therefore does not automatically determine which maturity level an organisation is given when it is measured.

This measurement paradox arises because an AI-native organisation has no transition phase through which it would AI-transform. An AI-native organisation is by its nature already AI-transformed and may use, for example, the autonomy of agents and the direction of freed-up capacity.^{[20][21][22][23]} Despite an advanced level of AI use, private and commercial AI service accounts may be in use without the organisation yet governing them.

BCG describes AI-nativeness as a state in which an organisation’s processes have been completely redesigned with the help of AI. PwC depicts a state in which AI runs through operations, decision-making and architecture. Deloitte emphasises the redesign of work.^[24][25][26]

The large consultancies name AI-nativeness as the goal, but this guide is a little more cautious about recommending it. Not because AI-nativeness would not in principle be worth pursuing, but because there is not enough measured evidence and organisations differ greatly.

To draw a parallel with digitalisation: in principle every organisation should digitalise its operations, but in practice an investment-side holding company may not even need websites. Equally, the “investment pipeline” may already be full of high-quality targets, so agentic AI would not deliver enough added value. Another variable is the time span. In Finland, banking was digitalised decades ago, retail considerably later.^[27]

For these reasons this guide encourages organisations to examine the benefits available from AI-nativeness and to form their own view and timetable on the matter.

Summary for decision-makers on AI maturity

The maturity model in this guide measures an organisation’s level of AI maturity on the basis of concrete observations rather than a subjective assessment. The model describes the benefits, risks and indicators of each level, which gives grounds for choosing a target. Setting the target level must always start from the organisation’s own strategy.

Here are some questions a decision-maker can ask in their own organisation:

• Does our strategy take a position on a target AI maturity level?
• What maturity level would succeeding in our strategy require?
• Which process’s starting and end level do we measure as part of AI transformation?
• Where should the time freed up by AI be directed?
• Do we have a picture of what the maturity level we are aiming for costs?
• How is responsibility for raising the maturity level assigned in our organisation?

Sources

1. METR, Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity, 10.7.2025. arXiv:2507.09089.
2. Bonney, Breaux, Dinlersoz, Foster, Haltiwanger & Pande, U.S. Census Bureau: BTOS AI supplement survey, CES Working Paper 26-25, 4/2026.
3. Tilastokeskus, Tietotekniikan käyttö yrityksissä, 27.11.2025. Suomen virallinen tilasto (SVT).
4. Gillespie, Lockey, Ward, Macdade & Hassed, Trust, attitudes and use of artificial intelligence: A global study 2025. University of Melbourne and KPMG, 2025.
5. Dell’Acqua, McFowland III, Mollick et al., The Cybernetic Teammate: A Field Experiment on Generative AI Reshaping Teamwork and Expertise. NBER Working Paper 33641, 2025.
6. Anthropic, How ABC Legal turned every employee into a builder with Claude Managed Agents, 17.8.2026.
7. Jabarian & Henkel, Voice AI in Firms: A Natural Field Experiment on Automated Job Interviews. arXiv:2607.28222, 7/2026.
8. McKinsey & Company, The state of AI in 2026: On the road to ROI, 25.8.2026.
9. Aral, Brynjolfsson & Wu, Three-Way Complementarities: Performance Pay, Human Resource Analytics, and Information Technology.
10. Milgrom & Roberts, Complementarities and fit: Strategy, structure, and organizational change in manufacturing.
11. McElheran, Yang, Kroff & Brynjolfsson, The Rise of Industrial AI in America: Microfoundations of the Productivity J-curve(s).
12. Klarna, press release 27.2.2024.
13. Interview with Sebastian Siemiatkowski, Bloomberg 8.5.2025.
14. McKinsey QuantumBlack, Where AI agents pay off: A practical guide to the economics of agentic workflows, 24.8.2026.
15. BCG, AI Radar 2026.
16. Malinen, Boost Your Digital Maturity — Digitalize Your Operations. Sofokus, 2020. Digital business maturity model.
17. ServiceNow and Oxford Economics, Enterprise AI Maturity Index 2026.
18. Novikov, Vũ, Eisenberger et al., AlphaEvolve: A coding agent for scientific and algorithmic discovery.
19. Fin (formerly Intercom), Announcing Fin Apex: the age of vertical models is here, 26.3.2026; background: VentureBeat interview 26.3.2026.
20. McKinsey Technology (Tavakoli, Goodman, Rowshankish, Kumar, Soller, Barreto, Parekh & Litta Modignani), Building the foundations for agentic AI at scale.
21. Humlum & Vestergaard, Still Waters, Rapid Currents (formerly Large Language Models, Small Labor Market Effects). NBER Working Paper 33777, 5/2025, revision 3/2026.
22. Suh & Oh, Generative AI and the Reallocation of Time: Productivity, Leisure, and Fulfilling Work. arXiv:2602.12695, 13.2.2026.
23. Parasuraman & Riley, Humans and Automation: Use, Misuse, Disuse, Abuse. Human Factors 39(2), 1997 · Parasuraman & Manzey, Complacency and Bias in Human Use of Automation: An Attentional Integration. Human Factors 52(3), 2010.
24. Budget allocation principle. Boston Consulting Group, Closing the AI Impact Gap, 2025.
25. PwC, Two futures for jobs in an AI era: 2026 Global AI Jobs Barometer, June 2026.
26. Deloitte, AI maturity model, AI maturity and digital value materials, 2024–2026.
27. Tilastokeskus, Väestön tieto- ja viestintätekniikan käyttö (SUTIVI). Suomen virallinen tilasto (SVT). StatFin table 13ud. Early level of banking services use Tieto- ja viestintätekniikan käyttö 2008 — Internetin käytön muutokset, 27.4.2009.
28. Andon Labs Cafe.

About the author

Founder & Chairman

Teemu Malinen

Teemu writes about digital business trends, modern company culture and startup investments.

Read the full bioShow less

Read more articles (157)