Companies are increasingly adopting artificial intelligence, but scaling the technology has not yet led to a comparable increase in financial performance. One reason may lie not in the capabilities of the models themselves, but in how well they understand the organization’s internal processes: who makes decisions, how employees hand off tasks to one another, which rules allow exceptions, and when a problem needs to be escalated. This organizational context is becoming a distinct resource that determines the practical value of AI.

AI adoption is growing, but the financial impact is not

According to McKinsey’s 2026 The State of AI study, 37% of respondents said that AI contributes at least somewhat to the company’s earnings before interest and taxes (EBIT). This figure has remained virtually unchanged compared with the previous year.

Even more telling is the share of companies that McKinsey classifies as the most successful AI users. This category includes organizations where the technology accounts for at least 5% of EBIT and its impact is considered significant. There were only 6% of such companies — and that share also remained unchanged.

These figures point to a gap between the spread of AI and its measurable financial return. Simply implementing new tools does not guarantee that they will improve the performance of the organization as a whole.

AI needs to know not only the rules, but also the real processes

One approach to solving this problem is to create what is known as institutional intelligence. This term refers to systematized knowledge about how a company actually operates, rather than only how its processes are described in official documents.

Corporate knowledge bases may store instructions, regulations, and employee information. But many important details usually remain undocumented: which exceptions are acceptable in practice, who has the authority to override a standard procedure, why some issues require immediate approval, and why others are resolved at the team level.

These details often determine how successfully a task moves from one employee to another. If an AI agent has access only to formal instructions, it may fail to account for real constraints, approval specifics, and the reasons why employees do not always follow the standard scenario.

As a result, a gap arises between management’s intentions and how work is actually carried out across different departments. In Smartsheet’s material, this phenomenon is called execution drift — a divergence between intended and actual process execution.

Why it is not enough for an AI agent to simply be smart

A new employee usually learns a company’s internal processes gradually — through training, practice, and communication with colleagues. An external contractor may never gain access to all the necessary information.

An AI agent is potentially capable of using accumulated knowledge immediately after deployment. But this advantage works only if the company has already collected the necessary context, organized it into a clear structure, and provided secure access to it.

If that foundation does not exist, every new AI tool risks starting almost from scratch. A more powerful model alone will not reconstruct the history of decisions, learn unwritten rules, or understand why a particular process is designed the way it is.

If organizational knowledge is systematized, different agents can use it. As they work, they may also potentially enrich this knowledge base, provided the company has put in place mechanisms for verifying and updating the information.

How companies can prepare organizational knowledge for AI

To do this, it is not enough to simply gather all documents in one repository. Information about tasks, employees, workflows, and the systems in use must be connected so that AI can take it into account in the context of specific work.

Workflow management platforms, including Smartsheet, propose bringing together tasks, data, and team interactions in a single environment. This approach can help organizations:

· align the work of departments with shared goals;

· connect data from different corporate systems and applications;

· adapt processes as projects and priorities change;

· monitor the actions of AI agents and track their execution;

· gradually systematize knowledge about how work is actually carried out.

At the same time, it is important not to confuse the accumulation of data with the formation of institutional intelligence. A large archive of documents still does not guarantee that AI will be able to correctly interpret their content or determine which information is relevant. This requires a high-quality structure, up-to-date information, access controls, and oversight of agent actions.

The next stage of AI adoption is organizational

The thesis about the importance of organizational context lies at the core of Smartsheet’s promotional material, so its conclusions should be viewed with the company’s commercial interests in mind. However, the issue itself deserves attention: corporate tasks are rarely reduced to the application of universal knowledge. Their solution depends on specific rules, constraints, and relationships within the organization.

In the early stages of AI adoption, companies often focus on choosing a model and evaluating its capabilities. But as agents become more widespread, increasing importance may be placed on what information is available to these systems, how reliable it is, and how well it is embedded in real workflows.

AI may suggest a way to solve a task. To carry it out in accordance with the rules of a specific company, it needs to understand how that company is structured. Therefore, further growth in returns from artificial intelligence may depend not only on improving models, but also on how well organizations learn to describe their own work in a form suitable for machine use.