Sber explains why businesses need an AI Overlay layer instead of restructuring departments

Sber is urging companies not to tear up their org structure for the sake of AI, but to add a separate technology layer on top of the existing business: an AI Overlay. The logic is simple. Real returns come not from the number of pilots but from a shared architecture that connects models, data, processes, and metrics into one system.

Why pilots stall

The AI market moved past the experimental stage some time ago. Roots Analysis estimates that the global market for AI agents could grow from $9.8 billion in 2025 to almost $221 billion by 2035, a compound annual growth rate of 36.55%. Meanwhile, 88% of companies already use generative AI in customer service, marketing, logistics, HR, finance, manufacturing, and analytics. Yet only 6% of organizations report steady growth in operating profit. For business, that is an uncomfortable gap between the hype around the technology and measurable financial impact.

In Sber's view, the cause is fragmented adoption. Companies typically launch dozens of local initiatives: a support chatbot, a marketing assistant, document generation in the back office, knowledge-base search for employees. Each team tackles its own problem in isolation, with its own data, integrations, and metrics. As a result, AI speeds up individual operations but doesn't rebuild the end-to-end process. The impact gets diluted, costs climb, and leadership finds it increasingly hard to tell where new profit is actually emerging and where there's just more technological noise.

The logic of AI Overlay

Instead of a painful restructuring of departments, Sber proposes a different approach: lay a horizontal AI Overlay over the current structure. It isn't a new ERP, a CRM replacement, or an attempt to rebuild the entire business around a single model. It's a layer that connects existing systems, data-handling rules, and a set of AI services. This layer lets agents and models see a shared context, follow common standards, and take part not in a single episode but in the full decision-making chain.

In practice, this means a company can scale AI without organizational upheaval. Functional teams keep their areas of responsibility but gain a shared technology framework that carries data, tasks, automation, and quality control. If a use case proves effective in sales or support, it becomes easier to carry over to adjacent processes. AI then stops being a collection of scattered pilots and becomes an operational layer of the business that can be developed, measured, and centrally controlled.

What the layer is made of

For this approach to work, buying access to a strong model or deploying a couple of bots isn't enough. It takes an architecture in which AI is embedded in the company's operations as a managed system. Sber identifies several essential building blocks: unified access to data, an orchestration layer for models and agents, integrations with enterprise platforms, security mechanisms, and clear outcome metrics. Only when these work together does AI start to affect not just individual employee actions but the speed, quality, and economics of the entire process.

• Unified access to data and knowledge
• Orchestration of models, agents, and workflows
• Integrations with CRM, ERP, and internal services
• Control over security, roles, and response quality
• Performance metrics and links to P&L

If even one block is missing, the system degrades quickly. Without high-quality data, agents make mistakes and hallucinate; without orchestration, they duplicate actions; without integrations, they hit a wall of manual operations; and without metrics, the project turns into a polished demo showcase. That's why an AI Overlay rollout should start not with an abstract transformation but with specific process chains where you can measure time savings, conversion gains, fewer errors, or faster service. That's where a horizontal layer best proves its value.

What it means

Enterprise AI is shifting from the craze for standalone copilots toward a platform approach. For executives, the question is no longer which bot to buy first, but how to assemble a unified AI framework on top of the existing business. Those who build such a layer ahead of others are more likely to turn dozens of pilots into consistent operational results without an expensive restructuring of the whole company.