From ERP to Context Management: How the CIO's Role is Changing in the Age of Artificial Intelligence

Alexander Semenov, CEO of KORUS Consulting Group

Over the past decades, companies have invested heavily in digitalization: implementing ERP systems, automating key processes, building corporate data warehouses and application integrations. This foundation has enabled businesses to transition from manual management to transparent, data-driven processes. But the development of generative AI shows that this level of digitalization is no longer sufficient. The next challenge is teaching AI to work with business logic: understanding processes, taking constraints into account, and making decisions within established frameworks. For CIOs, this means expanding their scope of responsibility. It's not enough to simply ensure systems function; they must help businesses create an environment in which new technologies deliver measurable results in a short period of time.

From Experiments to Business Impact

According to McKinsey , 78% of companies are already using generative AI in at least one business function. The situation is similar in Russia, but most are still at the stage of implementing specific use cases: employee assistants, document automation, and analytics and development testing. To achieve company-wide impact, simply implementing a new model or launching a new service is not enough.

The key question is how to measure results. According to our data, only one in five companies has reached the level of systematically managing AI implementation with a specific strategy, KPIs, and performance analysis. For me, the key metric is labor productivity—how much more a company can earn with the same number of employees.

In some processes, the impact is measured directly. For example, in mass hiring of blue-collar workers, the speed and cost of recruitment can be compared before and after automation. If the process has become faster and cheaper, the economic impact is clear and measurable.

In software development, evaluation is more complex. AI accelerates coding, documentation, and testing—but the final impact depends on the maturity of development processes, architectural standards, and the team's proficiency in using new tools. For example, we implemented our own AI platform that consolidated client, project, document, and expertise data into a single, secure environment. In certain scenarios, the time to prepare standard documents and design materials decreased by 25–30%, and the time to find the necessary information decreased by up to 40%. On average, new employee onboarding time was reduced from five to three working days.

Why data doesn't become knowledge on its own

Large companies have vast amounts of information, but data volume doesn't equal knowledge. Information is scattered across ERP, CRM, MES, analytics platforms, archives, and document databases. Individual departments understand their processes, but the company rarely has a unified view of how the organization as a whole operates.

For classic IT systems, this level of description was sufficient. For an AI agent, it's not: it requires context, which humans carry unconsciously in their heads. A company employee understands that the downtime of a specific unit can halt production, that a certain supplier is critical, and that changing a single parameter will affect several departments. A model lacks this understanding by default: connect it to corporate data, and it will have access to the information, but it won't understand the significance of objects and the relationships between them. You'll have to constantly set additional constraints and manually check the agent's actions.

Thus, the key question for the company is changing. Previously, clients asked, "How do we connect AI to our data?" Now they ask, "How do we define our business so that AI can work within it?"

For example, an "equipment" object at an industrial enterprise would be a unit of account with an inventory number for an ERP system. For manufacturing, it would be part of a technological process. For the operations department, it would be an object with a maintenance schedule and risk of failure. For the financial department, it would be an asset that impacts the company's economics. Four descriptions of the same object—and none of them gives the agent a complete picture.

This is where the concept of ontology comes in: a description of a company's subject area—what objects exist in the business, what properties they possess, how they are related to each other, and what rules govern their interactions. This is not a new idea for IT architecture: in object-oriented programming, a model of objects and relationships is defined before building logic on top of it. The same principle applies to corporate AI—only it's the business, not the code, that needs to be modeled.

An ontology links data, processes, and business rules, enabling agents to work with company logic, not just information. A procurement agent considers production plans, delivery times, alternative suppliers, budget constraints, and the business implications of a decision. A logistics agent considers routes, shipping costs, customer priorities, warehouse constraints, and the risk of missed deadlines. For example, we have a yard management solution in which AI agents manage all stages of vehicle movement within a distribution center. If multiple vehicles with different priority levels arrive simultaneously, the agents consider the urgency of orders, dock occupancy, processing time forecasts, and the current situation on the premises. Instead of processing vehicles sequentially, the system dynamically reorganizes the queue, reassigns docks, and routes to avoid congestion. As a result, yard throughput increases by up to 25%.

Therefore, ontology is not a standalone technological initiative, but rather a part of corporate architecture: a connecting link between transaction systems and next-generation AI platforms. Without it, companies repeatedly hit the same wall: data is accumulated, models are accessible, experiments are underway, but sustainable economies of scale are lacking.

You may also be interested in the following material from the IT Leaders Club Compass CIO

ERP in a new outline

ERP remains a key element of the IT landscape, but its role is shifting. Historically, ERP helped manage operations—orders, procurement, production, finance, inventory—and its goal was to provide a unified accounting framework. As intelligent systems develop, this is no longer enough: businesses need to not only record transactions but also use the accumulated information for decision-making.

A manufacturing company may have data on orders, inventory, and production plans in its ERP. But a broader context is crucial for decision-making: equipment status, production constraints, resource availability, delivery times, and the economic impact of changes. This is the job of the ontology layer—it integrates data from ERP, CRM, MES, and other systems with descriptions of objects, processes, and rules. ERP continues to handle the operational framework, while a mechanism for managing business context and decision-making is built on top of it.

This also changes the approach to implementing new systems: it is important to think through in advance not only the process logic within the ERP, but also what data, connections, and rules should be available to the automation tools that will appear on top of it.

How the CIO agenda is changing

Traditionally, the CIO's area of ​​responsibility rested on three pillars: infrastructure to meet business needs, stable operation of corporate systems, and information security. These tasks are still relevant, but AI adds a new dimension: the CIO becomes a participant in changes to the company's key business processes.

Connecting the model to equipment data is only half the battle. The real challenge is determining what decisions the agent can recommend, what actions it can perform automatically, who is responsible for the results, and what constraints are built into its operation.

The approach to architecture is also changing. Previously, the focus was on system integration and unified access to data. Now, it's about building an architecture in which AI can securely interact with the corporate environment: managing corporate knowledge, describing processes, monitoring data quality, managing agent access, and delineating human and system responsibilities.

Essentially, the CIO is moving from the role of IT systems owner to that of change architect—someone who helps the company understand which processes can be reimagined, what knowledge needs to be preserved, and where technology actually creates business value.

Five Questions Before Scaling AI

1. What process are we changing and what effect are we measuring?

The first step isn't choosing a technology, but identifying the process where the company is wasting time, resources, or degrading the quality of decisions. Each AI implementation scenario should have measurable targets: reduced operation time, reduced costs, increased productivity, and improved quality.

2. Is the process described in such a way that it can be transferred to an AI agent?

Most company processes rely on employee experience. Formal regulations often fail to reflect the actual decision-making rules, exceptions, and limitations. Before implementing AI, it's essential to define the essence of the process: what actions are performed and according to what rules.

3. What data is the source of truth?

AI alone won't solve the problem of disparate and inconsistent data. It's necessary to determine which systems own the information, who is responsible for its quality, and what data can be used to make decisions.

4. What can be delegated to AI?

Not all operations should be performed by the agent independently. The boundaries of autonomy must be defined in advance: what the system does automatically, where employee confirmation is required, and what decisions are left to the human.

5. Is the architecture ready for agents?

Agents operate on top of existing systems, requiring unified access rules, integration between platforms, action monitoring, and decision auditing.

2030: What will determine company success?

In the coming years, companies that solve three practical problems will gain an advantage: defining their own processes, creating a unified context for working with data, and defining the rules for interaction between humans and intelligent systems. This requires changes not only in technology but also in management: identifying owners of business knowledge, rethinking approaches to data quality, formalizing critical processes, and integrating controls into agent work.

The models themselves will change rapidly—today's market leader will give way to another tomorrow. Building a strategy around a specific model is short-sighted. Sustainable advantage is built around what's harder to copy: an understanding of one's own business, the quality of corporate data, and the ability to quickly redesign processes.