How Norges Bank Investment Management integrated AI into 171 processes across the fund

Norges Bank Investment Management, which manages the world's largest sovereign wealth fund, conducted not a point experiment but a large-scale AI restructuring of the entire organization over two years. Instead of seeking one perfect scenario, the team embedded AI directly into 171 processes — from internal work tasks to critically important investment decisions.

Scale of AI Implementation

The key idea of this case is simple: AI did not become a separate "innovation lab" at NBIM where beautiful demos are stored. The fund moved in the opposite direction and began systematically searching for areas where models could save time, reduce costs, or improve decision quality. As a result, the implementation spread throughout the company rather than remaining within a few enthusiast teams.

For a large financial organization, this is particularly telling: typically any changes there move slowly due to requirements for risk management, approvals, and accuracy. This approach transforms the very logic of digital transformation. Instead of betting on one "golden case" that should prove the value of the technology, NBIM distributed the effect across dozens and hundreds of small applications.

This reduces dependence on a single project and yields more sustainable results: even if some experiments fail, the overall effect continues to accumulate. For companies with a large number of internal processes, this may be a more realistic strategy than waiting for one breakthrough product.

How They Restructured Processes

The most rigorous step was mandatory training for employees, including those who initially did not want to work with AI tools. The logic is clear: if the technology is to become an everyday part of work, it cannot be left only to a narrow circle of specialists. In parallel, the fund abandoned the familiar Scrum approach in favor of micro-teams, where two developers and one business representative work side by side.

This reduces the distance between idea, implementation, and verification of benefit, and also accelerates the launch of changes. This scheme was not limited to abstract process changes on paper. The fund simultaneously changed training, team structure, and how tasks were formulated, so that AI did not depend on individual enthusiasts.

Through this, new practices became embedded in daily work rather than existing as pilots that are shown at internal presentations but never reach large-scale application. This is why implementation more easily survives shifts in priorities and does not collapse after the first quarter.

• mandatory training in AI tools for all employees;
• transition from large development processes to micro-teams of three people;
• implementation of AI not in one product, but in 171 separate processes;
• creation of agent architecture for tasks that affect investment decisions.

Agent architecture is particularly important. When dealing with sensitive financial tasks, one model with one answer is not enough. Companies need a system where roles are divided: one agent collects data, another verifies calculations, a third helps formulate conclusions, and humans retain control over the final decision. For the capital management field, this is much closer to working reality than the image of a "magic button" that solves everything on its own.

What They Got in the End

The results look not like cosmetic improvement but like a shift in everyday work culture. More than 50% of fund employees now write code. This does not mean that half the company became full-fledged engineers, but it shows something else: programming and automation cease to be the monopoly of the IT department.

People from business functions begin to assemble scripts, prototypes, and internal tools themselves if it helps them solve their tasks faster. This gradually changes the requirements for roles within the fund. There is also a more direct business effect.

NBIM reports savings on trading costs and a sharp reduction in time spent preparing for meetings — approximately 80%. For an investment organization, this is a very practical metric: less time is spent gathering materials, checking data, and preparing brief conclusions for discussions. If such improvements are scaled across the entire fund, AI ceases to be an expense item for experiments and becomes infrastructure that daily impacts operational efficiency.

What It Means

The NBIM case is interesting because it shows a mature corporate scenario: not "implement a chatbot for the sake of a checkbox," but restructure training, teams, and decision-making around AI. For large companies, the takeaway is obvious: the greatest effect can come not from one flashy pilot, but from dozens of embedded applications that bit by bit speed up work, reduce costs, and make employees significantly more independent. Especially where the cost of slow processes is measured in millions.