The AI paradox: why efficiency alone doesn't pay the bill?

McKinsey estimates that generative AI could add up to US$4.4 trillion per year to the global economy. And here lies the major challenge that leadership faces today: while AI enhances operational efficiency and automates processes, these gains are not translating into proportional growth or competitive differentiation.

AI models, increasingly accessible, are leveling the playing field, as everyone has the same tools at their disposal. The differentiator, therefore, has shifted from the algorithm itself to the organizational capacity to connect AI with data, processes, and decision-making.

The disconnect between reality and expectation

In partnership with Sensedia, we launched a study conducted by MIT SMR Brazil, “Efficiency Doesn't Pay the Bills: The Paradox of AI in Organizations”. This study presents alarming figures about the AI landscape in large corporations:

• The gap between leadership's perception and technical reality: while 66.2% of executives state that AI is a defined priority or a core part of the business, only 7.4% of technology professionals report that it is truly integrated into multiple systems and processes;

• The experimentation trap: more than half of organizations (51.6%) are still operating in proof-of-concept (PoC) or isolated initiatives, while only 12.9% have effectively scaled and productionized use cases;

• The data blackout: 54.4% of technology professionals do not have structured access to corporate information to feed AI applications, and 31.6% of these rely on manual context input via prompts;

• Lack of control: 71% of professionals operate with no or basic control over AI responses, performance, and costs.

Knowing this, it's not surprising that 95% of AI projects fail, not because of the technology itself, but due to a lack of strategy, quality data, and integration within companies, a figure taken from the MIT report “State of AI in business”.

Integral design as a strategic response

To break the cycle of isolated automations and pilots that yield little practical results, Romulo Cioffi, Chief AI and Innovation Officer at Squadra, advocates for shifting the competitive advantage from execution to decision architecture.

With the commoditization of LLMs (Large Language Models), the focus shifts to the operational flow and the company's ability to integrate and orchestrate data and processes to feed AI with business context.

To mitigate the risk of context fragmentation, the integral design methodology proposes structuring any business problem into four fundamental models that lead to the solution's designation:

• Environment: mapping of internal and external factors that interact with the business ecosystem;

• Services: structuring of functional blocks, defining inputs and outputs;

• Processes: precise description of the execution dynamics and impact on the organization's stakeholders;

• Architecture: physical materialization through a scalable and real technological infrastructure.

“Integral design is the necessary foundation for creating organizations capable of operating distributed intelligence with consistency, security, and real scale.” - Romulo Cioffi

The next frontier of AI

The market is rapidly moving away from passive copilots. The new era is about orchestrated multi-agent ecosystems, which are delegated authority to act autonomously within an integrated system.

Here, Gartner issues a warning: companies that merely insert AI into legacy systems, without redesigning their architectures, without integrating and orchestrating data and processes to feed AI with the right and complete context, could face a margin compression of up to 80% by 2030. In light of this warning, beyond integral design, new integration protocols are gaining prominence:

• MCP (Model Context Protocol): a standard proposed by Anthropic that connects the agent to tools, APIs, and corporate data in a contextualized manner;

  • However, 68% of companies have not yet adopted it in production;

• A2A (Agent-to-Agent): a protocol focused on collaboration and complex delegation between different specialist agents.

What are those who are already reaping the rewards of scale through AI doing?

The study, “Efficiency Doesn't Pay the Bills: The Paradox of AI in Organizations”, also presents real-world cases from the Brazilian market of companies that have successfully applied artificial intelligence robustly and generated real value, covering the costs of AI within the organization.

Bradesco

Developed the Bridge platform and adopted virtual squads based on multiple agents. The result? A 95% reduction in model development time and an estimated value generation of R$250 million per year.

Mapfre

Integrated AI directly into the customer journey (such as in vehicle assistance and claims handling via WhatsApp), generating damage analysis and repair estimates seconds after photo submission.

Klabin

Transformed the Luna platform from a basic chatbot into a centralized orchestrator agent, connecting HR, IT, and even industrial procedures on the factory floor.

The data clearly shows that operating intelligence in isolation is the fastest way to scale costs, not the business.

The future belongs to companies that can integrate AI and orchestrate their processes with control and governance, while ensuring that artificial intelligence receives the complete data and contexts necessary for its success.

The full study “Efficiency Doesn't Pay the Bills: The Paradox of AI in Organizations” deeply analyzes the technical challenges, impacts, and practical steps taken by market leaders to effectively access the scaled benefits that AI can provide. Did it pique your curiosity and do you want to analyze the study's data in full? Access and download it now.

Also available at: InforChannel | TI INSIDE | SEGS | Jornal Tribuna