When AI produces too much information
Artificial intelligence seems to have entered a more demanding phase in companies. After the pilots, the first automations and the incorporation of tools into daily work, the question of return is gaining weight, since the technology has already proven it can produce good content at scale. Now it needs to show its contribution to better decisions, less rework, faster flows and tangible economic results.
AI has drastically reduced the cost of generating outputs. Reports, summaries, analyses, presentations, scenarios and recommendations can appear in a few seconds, which represents real progress for many activities. By eliminating repetitive effort, structuring scattered information and broadening repertoires, the technology helps professionals and organizations move forward faster.
Output, however, is not a result. An AI-generated answer must be turned into useful information, which in turn has to be validated, interpreted and connected to a decision. That decision only creates value when it improves execution, removes a concrete barrier or enhances the system's performance. Between producing content and generating results there is a managerial path that technology does not eliminate.
AI has changed the economics of information. For decades, companies invested to obtain more data, more analysis and greater processing capacity. Today, in many situations, the constraint has shifted to absorbing, validating and using what has been produced. The company gains speed in generating analyses, while it may lose speed and precision in deciding.
The management tradition developed by Toyota has always treated “overproduction” as one of the most dangerous wastes, because producing ahead of real need consumes resources, creates inventory, takes up space, hides problems and generates new wastes. The strength of this idea lies in the managerial mechanism it reveals: work done without clear demand tends to look like productivity until hidden costs spread throughout the system.
In the AI era, overproduction has taken digital form. The organization produces information before knowing who will use it, which decision it will support and what problem it will clarify. Reports, dashboards, simulations and presentations multiply with an appearance of efficiency, and part of this material sits stored in clouds, circulates in the form of messages, feeds meetings and occupies managers' attention. This volume of analysis and processing without real need becomes “cognitive inventory,” a phenomenon known as “infobesity,” or information obesity.
This inventory has a cost, which shows up both in the calendars of overloaded individuals and in the technology bill. On one side, there are too many meetings and the fatigue of those who must compare, review and interpret redundant content. On the other, there are the tokens, processing and infrastructure consumed with each interaction, plus document queries, agents in nonstop operation and endless rounds of corrections. Digital overproduction is becoming a highly significant variable cost.
The financial impact of this technology race is already reflected in Brazilian organizations' balance sheets. Market data indicate that companies' average spending on artificial intelligence in the country jumped 191% in a period of just two years, showing that new tools are often adopted without a rigorous analysis of return on investment (ROI).
Many companies still evaluate AI by the number of tasks automated or by the time saved, although this calculation can be misleading. An agent that generates excessive analyses, queries databases indiscriminately, takes long routes or delivers answers that require heavy review shifts the cost to later stages. The output comes out fast, while the total work remains heavy.
Tokens make digital overproduction measurable. They do not explain the whole cost, but they reveal an important dynamic: spending grows with the organization's behavior. Poorly framed problems, confusing processes, unlimited agents and redundant requests consume more tokens, so the AI bill, in many cases, reflects the quality of the management system.
Volume is only part of the problem, because the quality of the information that circulates has come to demand the same attention. AI produces well-structured texts, fluent answers and analyses with an appearance of maturity, which makes it easier for flimsy information to circulate. A piece of data without a clear source, an inference presented as fact or a recommendation without the proper degree of confidence can enter the decision-making flow quite naturally.
The lean way offers a useful discipline for this new environment: quality at the source. In any process, the idea is to prevent defects from moving forward. When something deviates from the standard, the flow needs to signal the problem, allow correction and protect the following steps. The principle is simple and demanding: quality must be built into the process itself, not merely checked at the end.
Applied to information, this reasoning becomes especially timely. An AI-generated analysis should carry minimum signs of reliability before it circulates. Its origin needs to be clear; its purpose, defined. Someone must be accountable for the interpretation and the degree of confidence, and the limits of its use need to be understood. Without this, the material can serve as an initial hypothesis or exploratory support, but it should not yet circulate as evidence to support decisions.
Overproduction and quality at the source form two complementary filters. The first examines the need for and timing of information, while the second analyzes its reliability. One protects the organization from excess; the other, from fragility. Together, they can help place AI back within a logic of value, trust and return.
Quality at the source also has an economic impact, because bad information consumes budget. An inaccurate answer generates new rounds of prompting, checking, review, dispute and rework. An analysis without a source forces teams to retrace their steps, while a poorly supported recommendation consumes meeting time and may lead to weak decisions. The company pays to produce the output and then spends again to find out whether it was reliable.
This cost does not always show up in traditional productivity reports. It appears as more hours of validation, more layers of approval, greater caution from managers, analytical rework and loss of trust in the very mechanisms available. Without quality at the source, AI speeds up the output but increases cost and noise. Speed goes up, but the decision-making flow remains slow and murky.
Governance standards need to reach the information flows created by the technology. Rules on usage, security and compliance are necessary, but capturing return requires more than that. Before automating an activity, it is worth knowing which problem it solves. Before deploying an agent, it is necessary to define which decisions it supports, which sources it may use and which limits it must respect.
This puts leadership back at the center of the discussion, because the quality of AI use depends not only on the isolated sophistication of the tool, but also on the managerial clarity that guides its application. Well-framed problems, clear standards, defined responsibilities and consistent decision routines allow AI to expand capabilities, while confusion, overlapping processes, poor information discipline and little clarity about value creation tend to perpetuate the wastes that already exist.
Technology is forcing companies to look at something that has always been hard to measure: the quality of managerial work itself. When technology clarifies the problem, reduces uncertainty and helps the work move forward, there is a real gain. However, when it needs to be explained, checked, redone or defended before it can be used, it merely transfers effort to another point in the process or decision-making stage.
That is why the bill for this new wave will be much larger than the invoice for tokens, licenses and infrastructure. It will include the lack of clarity in the problems framed, the low reliability of the information that circulates and leadership's inability to separate what helps from what merely takes up space and consumes energy. Technology can greatly expand organizations' capabilities, but it can also make their weaknesses more visible. In the end, the return on investment will depend less on the power of the tools and more on the managerial intelligence applied to their use.
