Cookie Preferences

We use cookies to provide a better user experience and personalised service. By consenting to the use of cookies, we can develop an even better service and will be able to provide content that is interesting to you. You are in control of your cookie preferences, and you may change them at any time. Read more about our cookies.

Skip to content

Director’s Cut: 5. Good AI stands on top of good management – how do you get there?

Director’s Cut is a no-fluff blog series. Mika Tarhala, Arter’s Managing Director, shares the reality of leadership. What he’s learned the hard way, where he’s succeeded, and why structures are everything.

AI models get talked about a lot. At Arter we found a new gear in AI-driven work when we put the AI knowledge base in order. A single AI model solves very little by itself, because what matters more is the kind of information the model uses in its work. In this blog I explain why AI-driven work requires putting the knowledge base in order. You will also hear what we have done about it at Arter.

Scattered information does not move your organisation forward

When we mapped out this autumn how AI-driven work actually operates at Arter, the first diagram showed six functions and their systems. Sales has its CRM and finance its ERP. Between the functions and AI there was empty space.

Source systems such as CRM and ERP know what has happened. But the rules, definitions and decisions that give meaning to the work were not generally available to everyone. That information was in individual PowerPoint presentations and in people’s heads.

In a situation like this, every person and every agent, meaning AI that carries out tasks independently, builds its own context, a little differently each time.

Scattered information therefore rarely helps you reach the desired outcome, and scaling the work becomes difficult, because the risk of doing things wrongly grows.

AI accuracy can rise to as much as 95% when a shared knowledge base is in order and in use

The strongest evidence for using a shared knowledge base came from Anthropic in the summer of 2026.

The company reported on its own analytics agents, whose accuracy rose in internal tests from 21% to over 95% when the whole organisation’s shared and up-to-date information was used in the background.

The biggest difference was that the business definitions had an owner, and were approved and up to date. These figures are the company’s own and have not been verified externally. The direction is nevertheless the same as the one we have seen in everyday work at Arter.

According to a rule of thumb from Boston Consulting Group:

  • 10% of the effort in an AI transformation goes into algorithms,
  • 20% into technology and data,
  • and the remaining 70% into people and processes. It is precisely this largest share that the management system carries.

Take a closer look at Boston Consulting Group’s AI Maturity Matrix 2026 here.

AI works well in software development:

  • Code has owners, versions, review and tests, and all of it lives in a shared repository.
  • The rest of the company has no such repository.
  • The management system is the repository for how the company operates.

The management system is a layer between the functions and AI

A management system is a description of how your organisation operates. It contains, among other things, strategy, processes, roles, instructions, decisions and metrics.

Many people know the management system through the ISO 9001, ISO 14001, ISO 45001 or ISO 27001 standards.

Source systems such as CRM and ERP hold the transactions themselves, while the management system holds what gives meaning to the other systems in your organisation’s operations.

At the core of Arter’s management system are the fact layers. The layer shared by the whole company is at the top, and below it are the management’s and the functions’ own levels. Each layer is owned by someone and its information is kept up to date; in the event of a conflict, the higher layer wins.

People with their AI tools, and agents, read the same approved content directly from the source through an open interface (MCP).

AI writes only drafts into the management system, which a human approves.

Because the rules are in the management system and not in an individual vendor’s tool, the AI model is easy to swap.

The management system tells you how things are done and how things relate to each other

In general, the folders on your computer or on a server contain individual files. They do not tell you what principles govern those files, or what dependencies a given piece of information has with any given whole.

A management system, in contrast, tells you about the relationships between the parts of your organisation’s operations and gives a picture of why your organisation operates the way it does.

In a management system every connection is visible, versioned and approved.

  • For a person, the management system is a kind of visual map that makes it easy to understand their own role in a process and why things are done in a particular way.
  • For AI, the management system provides a structured knowledge base on how things are done in your organisation, instead of individual files.

If you describe a process in the management system, it carries with it everything it is connected to. This includes, for example, the process owner, instructions, risks, information systems, metrics and goals.

The benefit of a management system shows in how quickly a change becomes visible in everyday work.

When in October we decided to talk about “AI-driven work” at Arter, the term was changed once in the fact layer, and from the next task onwards everyone used it, people and AI alike.

When strategy changes, the connections described through the management system show everything the change affects.

With AI, the effort of maintaining a management system can fall by as much as 80%

A common objection is that describing the operations of a whole company takes years and goes out of date immediately. That used to be true. Updating felt like a separate undertaking, and instructions were left in folders nobody opened.

Now AI drafts the management system on the basis of the information your organisation already has. In my own experience, the effort of building and maintaining a management system falls by about 80%. This has not yet been measured with customers.

Maintenance stays light because the work itself produces the updates.

In my own daily work this shows up in small things. When I correct AI once, the correction is recorded in the instruction, and I don’t need to repeat it in the next conversation.

This is quality management applied to AI:

  • The rules are agreed in advance,
  • AI and agents do the work,
  • deviations are compared against the rules, and the rules are refined when necessary.
  • A quality professional recognises the cycle immediately, and for a leader, continual development is the goal state.

Overseeing agents requires a digital twin of your organisation’s operations

Agent platforms collect logs, monitor runs and cap costs. All of that is needed. Yet logs only answer what the agent is doing.

An auditor asks something else. On the basis of which approved instruction did the agent act, who was responsible, and when was the instruction last reviewed? The organisation answers those questions. If operations have been modelled in a way AI agents can use, quality management extends to AI-originated work as well.

The ISO standards have arrived at a similar solution: ISO/IEC 42001 is a management system standard for AI, and it has been designed to be attached to an existing management system.

Case Arter: structure first, then agents

Agents are not yet widely in production at Arter:

  • The first agents will start in customer support, because its data and rhythm are ready.
  • This order is a deliberate choice for us.
  • First the structure and people’s AI work, then agents join the recurring processes.

During 2026, Arter colleagues have reported over 500 AI experiments across all functions. The enthusiasm for using AI is genuine, but the value from individual experiments stays with individuals, and the functions pulled the use of AI in different directions.

That is why Arter’s focus this autumn of 2026 has been on standardising the use of AI. When a practice works, it is described in the management system. After that, everyone can follow it, and the work can be scaled. And this is how we get, across the whole company:

All tracks pointing the same way. 🎿

Get to know ARTER, the management system for the age of AI.

Author of the Director’s Cut blog series: Mika Tarhala

Mika Tarhala is Managing Director at Arter Oy. With over 15 years of experience in the software and consulting industries, he has built a SaaS business from scratch to market leadership, leading it from Finland, with a distributed team, and for two years from Mexico. Today, he is building Arter as part of the European TSS group.

Mika’s leadership philosophy is captured in a single sentence:

Structure enables clarity, clarity builds trust, and trust drives growth.