Your company invested in AI. Is it getting the full return?

The challenge is no longer adopting Artificial Intelligence

The first two series of articles I published recently aimed to understand the transformation that modern IT operations are going through. We began by investigating why so many organizations continue to struggle to respond quickly to incidents, even when surrounded by monitoring, automation, and observability tools.

Throughout that reflection, we concluded that reducing resolution time depends not only on the speed of the teams or the quality of the technologies employed, but mainly on how quickly the operation can build a consistent understanding of the problem. This idea was developed in more depth in the article “Why reducing MTTR does not mean resolving incidents faster”, in which we argued that the greatest consumption of time occurs even before technical recovery, during the process of investigating and understanding the incident.

In the second series, we expanded this discussion to the role of Artificial Intelligence. We argued that specialized agents represent an important evolution, but that their ability to produce consistent decisions depends on the existence of a shared operational context. It was precisely this conclusion that led us to the article “The next generation of AIOps will be built on operational context platforms”, in which we proposed that the real differentiator of intelligent operations will not be the number of agents available, but the ability to build and maintain an integrated representation of the organization’s operational reality.

These two ideas lay the foundation for the new stage of our reflection.

If we understand that reducing MTTR depends on the speed with which we build understanding, and that this understanding comes to depend on a shared operational context, a new question naturally arises. After all, what will be the impact of this architecture on the investments companies have already started making in Artificial Intelligence?

This question gains importance because the market has changed quickly. A few years ago, the main discussion revolved around whether or not to adopt AI-based solutions. Today, that stage has already been passed by most large organizations. Specialized agents, copilots, intelligent features built into corporate platforms, and in-house AI projects have become part of the reality of companies in the most diverse sectors. Investment stopped being a future possibility and became part of the strategic agenda of the executives responsible for digital transformation.

This movement is confirmed by recent surveys on corporate adoption of Artificial Intelligence. The study The state of AI: How organizations are rewiring to capture value (McKinsey, 2025) shows that more than three quarters of organizations already use AI in at least one business function and that adoption of generative AI continues to grow rapidly. At the same time, the survey itself highlights that most companies are still in the early stages of capturing value at scale, indicating that the challenge is no longer simply experimenting with the technology but turning isolated initiatives into sustainable business results.

It is exactly at this moment that a new concern begins to emerge. After investing in Artificial Intelligence, how do you ensure that this investment continues to produce value over time?

The answer to this question will hardly lie in the continuous acquisition of new models or new agents. As these technologies evolve, the decisive factor stops being the number of existing initiatives and becomes the architecture capable of connecting them, allowing all of them to work from the same understanding of the operation.

It is this change of perspective that will guide this third series of articles. If before we discussed the challenges of modern operations and, later, the architecture needed for Artificial Intelligence to produce more consistent decisions, we will now turn our attention to an essentially strategic question:

How do you turn investments in Artificial Intelligence into a permanent operational capability, continuously preserving and expanding the value generated for the organization?

Return on investment does not grow at the same speed as investment

When an organization begins its journey of adopting Artificial Intelligence, it is natural for the first results to appear quickly. Automating repetitive tasks, speeding up content production, improving the user experience, or supporting incident investigation usually produce noticeable gains in a short time. This initial phase tends to reinforce the perception that expanding investment in AI will produce, in the same proportion, new gains in productivity and efficiency.

In practice, however, this behavior tends to change as new initiatives begin to coexist within the same organization.

The first project normally solves a specific problem. Then other areas start developing their own agents, embed intelligent features into already existing platforms, or adopt specialized solutions for different business needs. Soon after, the company is living with a growing set of initiatives that evolve relatively independently, each one producing value within its own domain.

This evolution undoubtedly represents a sign of maturity in the adoption of Artificial Intelligence. At the same time, it introduces a new challenge, very different from the one faced at the beginning of the journey. The problem stops being identifying opportunities to apply AI and becomes ensuring that all these investments contribute, in a coordinated way, to expanding the organization’s operational capability.

It is at this moment that many companies begin to notice a seemingly contradictory behavior. Although the number of Artificial Intelligence-based initiatives keeps increasing, the marginal return of each new investment tends to decrease. New agents keep adding important functionality, new platforms keep offering more sophisticated features, and new models start performing tasks previously considered unfeasible. Even so, gains start to grow at a slower pace than expected.

This perception also appears in recent studies on AI maturity in companies. In the report Seizing the agentic AI advantage (McKinsey, 2025), the authors describe the so-called “generative AI paradox”. Although adoption of the technology has grown rapidly and most organizations already report initiatives in production, few manage to demonstrate relevant financial impact at corporate scale. According to the study, the challenge is no longer only technological and now involves architecture, integration, process redesign, and the ability to scale the value generated by the different initiatives.

This slowdown is rarely related to the quality of the technologies adopted. Nor does it stem from a lack of competence of the teams responsible for the AI projects. In most cases, it reflects an architectural limitation. Each initiative continues to produce value within the space for which it was designed, but the organization finds it difficult to turn this set of individual capabilities into an integrated operational competence.

This distinction is important because it completely changes the nature of the discussion. For a long time, increasing the return meant adding new intelligent features to the environment. From a certain point on, however, the decisive factor stops being the amount of Artificial Intelligence available and becomes the ability to make all of this intelligence work in a coordinated way, sharing the same understanding of the operation.

It is at this point that many organizations begin to discover that the greatest return on their future investments may not depend on the next AI solution they will acquire, but on the architecture that will make it possible to leverage all those they already have.

The greatest return from AI is not in the next agent

It is natural to imagine that expanding investments in Artificial Intelligence means adding new capabilities to the operational environment. As more specialized agents, more sophisticated models, and platforms able to perform increasingly complex tasks appear, the tendency is to believe that the increase in productivity will occur as a direct consequence of this technological evolution. From this perspective, there will always be a next agent able to solve a problem that remains unsolved.

This logic makes sense as long as we analyze each initiative in isolation. One specialized agent can speed up incident investigation, another can support change analysis, a third can assist development teams, and a fourth can produce technical documentation in seconds. Individually, all of them represent important advances and justify the investments made.

However, corporate operations rarely work as a set of independent activities. A critical incident hardly remains restricted to infrastructure, applications, the network, or the database. Likewise, an operational decision almost never depends only on the information produced by a single tool or a single agent. As environments become more distributed and processes more interdependent, the need also grows to relate information from different technology domains to understand what is really happening.

And in this context, many organizations begin to notice an important difference between adding intelligence and increasing the intelligence of the operation.

Adding intelligence means incorporating new agents able to perform specific tasks more efficiently. Increasing the intelligence of the operation means creating conditions for all the knowledge produced by those agents to be interpreted in an integrated way, considering the existing relationships between applications, infrastructure, services, business processes, and changes made to the environment. They are complementary goals, but profoundly different.

This distinction helps explain why organizations with the same set of technologies often obtain such different results. It is not uncommon to find companies using similar observability, monitoring, ITSM, and Artificial Intelligence platforms but showing very different capabilities to diagnose incidents, coordinate teams, or make decisions during critical situations. The factor that normally differentiates these organizations is not the amount of technology available, but the architecture built to connect these technologies and turn scattered information into shared operational knowledge.

From this perspective, the greatest return on the next AI investment may not lie in the next agent to be added to the architecture, but in the ability to make all existing agents work from the same understanding of the operation. When that happens, the gain stops being incremental. Instead of producing just one more isolated capability, the organization starts to leverage all the capabilities it has built throughout its Artificial Intelligence adoption journey.

This change seems subtle, but it deeply alters how we should think about future investments. Instead of asking which will be the next technology able to add value to operations, it may be more important to ask which architecture will make it possible to extract more value from all the technologies already adopted. In many cases, this second question tends to produce much greater impacts on productivity, decision quality, and speed of response to incidents than simply acquiring one more Artificial Intelligence-based resource.

This perception marks an important change in the maturity of organizations.

During the initial phase of AI adoption, the main challenge was to experiment with new capabilities and identify application opportunities. As these initiatives mature, the challenge stops being technological and becomes architectural. Value no longer depends only on the quality of the available agents, but on the ability to turn them into a coherent system, able to produce a common understanding of operational reality.

And this architecture begins to determine how much additional value Artificial Intelligence can still deliver in the coming years.

Operational context turns initiatives into operational capability

When we observe the evolution of Artificial Intelligence in companies, we notice there is an important difference between having several AI-based initiatives and developing a true operational capability supported by Artificial Intelligence. At first glance, this distinction may seem merely a matter of scale. However, it represents something much deeper, because it involves how the knowledge produced by different systems comes to contribute to a common understanding of the operation.

During the initial phase of AI adoption, it is natural for each project to be developed to solve a specific problem. One agent supports the infrastructure team, another helps the Service Desk, a third improves developer productivity, and a fourth starts analyzing observability events. Each initiative generates real benefits and can usually justify the investment made. However, as this set of solutions grows, the organization begins to realize that the mere coexistence of different agents does not, by itself, produce a smarter operation.

Each agent continues to observe reality from the domain for which it was designed. Its specialized knowledge remains extremely valuable, but it is still limited to the universe of information it can access. If the context used by each one remains fragmented, decisions also tend to remain fragmented. The organization comes to have excellent specialists, human and artificial, without necessarily building an integrated understanding of what happens in its own operation.

Operational context then stops being merely a technical requirement and takes on a strategic role. When the organization manages to consolidate information from different platforms, preserve the relationships between applications, infrastructure, services, changes, integrations, and business processes, and make this representation available to people and AI agents, it no longer leverages just one specific initiative. It leverages all of them simultaneously.

This change produces an effect that usually receives little attention in discussions about Artificial Intelligence. The return on a new investment stops depending exclusively on the capabilities incorporated by the newly acquired agent and becomes influenced by the quality of the operational context it will find when integrated into the environment. An extremely sophisticated agent will produce limited results if it works from fragmented information. In contrast, a relatively simple agent may generate far more relevant contributions when it has access to a consistent, up-to-date, shared context.

This logic helps explain why organizations following quite different technological paths can reach similar levels of operational maturity. Some will opt for commercial models, others will develop their own agents, while others will combine different technologies over time. These choices will certainly influence costs, functionality, and speed of adoption. However, all of them will depend on the same element to produce sustainable results: an architecture able to preserve and share the organization’s operational context.

This shift in focus has also been observed by market analysts. The guide Enterprise Guide to Generative AI: Expert Insights on ROI, Use Cases, and Cost Management (Gartner) highlights that obtaining consistent return from AI requires more than choosing good models or use cases. Organizations need to evolve their architecture, establish governance mechanisms, measure the value delivered, and manage the total cost of the initiatives. In other words, as AI matures, the discussion moves from technology to the organizational capacity to integrate it sustainably.

From this perspective, context stops being only a component of the AIOps architecture and becomes a multiplier of the investments made in Artificial Intelligence. Each new agent added to the environment no longer starts its work from a limited view of the problem and instead takes advantage of a contextual asset that the organization has already built. The benefit lies not only in that agent’s individual productivity, but in the ability to make the entire ecosystem of intelligent solutions evolve in a coordinated way, preserving a common understanding of the operation.

Operational context comes to be understood as a strategic asset: it not only increases the efficiency of a specific technology, but continuously expands the value generated by all the technologies the organization decides to incorporate throughout its digital transformation journey.

The biggest AI investment may not have been made yet

When we look at the speed at which Artificial Intelligence has evolved in recent years, it is natural that organizations’ attention has focused on choosing the most promising technologies. New models appear at ever shorter intervals, specialized agents become more capable with each new generation, and practically all the major platforms embed intelligent features in their solutions. In this scenario, it is understandable that many companies tie their AI strategy to the ability to keep up with this technological evolution.

However, as the market matures, a different perception begins to emerge. The factor that will most influence the return on these investments may not be the speed with which an organization can incorporate new models, but the ability to make all of them produce value on the same understanding of operational reality. In other words, the competitive differentiator is no longer only Artificial Intelligence and comes to depend on the architecture that will make it possible to use this Artificial Intelligence consistently over time.

This change of perspective is particularly important because it reduces a risk that is beginning to worry many organizations. Models evolve quickly, new vendors enter the market, protocols are replaced, and specialized agents become obsolete at ever shorter intervals. Continuously investing in these technologies is part of the natural evolution of corporate architecture. However, if each new initiative needs to rebuild the entire understanding of the operation, a significant part of the knowledge produced will remain tied to the technology used at a given moment, hindering the organization’s future evolution.

Operational context completely changes this logic. By turning the representation of the operation into a permanent asset of the company, it allows new technologies to be incorporated without needing to rebuild, at each innovation cycle, all the knowledge about applications, dependencies, integrations, services, and business processes. Agents change, models evolve, and platforms are replaced, but the understanding of the operation continues to belong to the organization.

That is why the biggest investment in Artificial Intelligence may not have been made yet. Not because companies need to acquire more agents or adopt new models, but because they will need to invest in the architecture that will make it possible to preserve and continuously expand the value of everything they have already built. This architecture does not replace existing investments. On the contrary, it increases their ability to generate results, reduces dependence on specific technologies, and creates the conditions for the organization to evolve freely as new solutions emerge.

If the return on AI investments depends less on the chosen model and more on the architecture built to sustain it, an inevitable question arises: what is the greatest risk an organization runs when structuring its Artificial Intelligence strategy?

At first glance, the answer seems to lie in the choice of vendor, model, or technology. But, as we will see in the next article, the most significant risk may be elsewhere.

From vision to practice

The return on artificial intelligence investments does not depend only on the ability to develop new agents or adopt more advanced models. It also depends on how much knowledge, integration, and infrastructure the organization can carry over from one initiative to the next.

ARGUS contributes to this approach by building a layer of shared operational context. Through its AIOps capabilities, it connects information from monitoring, observability, ITSM, infrastructure, services, and other sources, relating data that normally remains spread across different tools.

This context represents knowledge about the operation itself that can be used by teams, automations, and different AI applications. Instead of each new agent having to rebuild in isolation the relationships between systems and processes, the organization can make a common foundation available to support its decisions.

With AgentOps and AI Gateway, the ARGUS proposal expands to also include the orchestration, observability, and governance of agents, allowing you to track their activity as new use cases are added to the operation.

This approach relates to the need to integrate AI into corporate architecture and processes, instead of treating it as a succession of independent initiatives.

The opportunity lies in turning one-off investments into reusable capabilities, preserving the knowledge developed and creating better conditions to expand the value of the next AI initiatives.

Get to know ARGUS and see how it works in practice.

About the author

Paulo Florêncio de Menezes is an entrepreneur, partner, and co-founder of companies such as Target Solutions, BiPTT, and ARGUS.

He holds a degree in Data Processing from the Federal University of Amazonas (UFAM), an MBA in Business Management from Fundação Dom Cabral (FDC), and a Professional Master’s in Economics from Fundação Getulio Vargas (FGV).

In his Master’s research, he developed the IBEOP-IA (Brazilian Index of Potential Occupational Exposure to Artificial Intelligence), investigating the potential exposure of Brazilian occupations to AI capabilities by combining occupational data, artificial intelligence patents, and natural language processing techniques.