Only 7% of organizations have fully scaled AI even though 88% use it in at least one function, according to McKinsey's 2025 global survey. Consultant Dessy Pavlova argues the gap comes from layering AI onto fragmented operations instead of redesigning workflows first. For business leaders, the message is direct: without process redesign, AI spending accelerates disorder rather than returns.
Fragmented handoffs behind the 88% to 7% gap
McKinsey links workflow redesign to stronger financial impact from generative AI, yet most companies still automate isolated tasks. Pavlova describes a familiar chain: a customer arrives via website, data moves to a spreadsheet, then to an operating platform, then back to the website, while finance gets a separate fragment for later reconciliation. Each handoff adds delay, duplication and error risk, even when every tool in the chain performs well on its own.
Many firms try to patch this stack with connectors such as Zapier or Make, forcing separate applications to cooperate. Pavlova proposes the opposite approach: build the system around the business itself, where one human decision triggers cascading updates. Her example is a training company with dozens of classes, where a student transfer or teacher schedule change must flow through website, operating system, scheduling and financial records. In that design AI carries the updates forward instead of staff re-entering data at each step.
The model keeps a person as architect and AI as operational engine, with explicit stopgates for review. A human sees what changed, intervenes when something looks wrong and confirms the endpoint, while AI moves information and flags anomalies. Research on 5,179 customer-support agents supports this division: access to generative AI raised productivity by 14% on average, with larger gains for less experienced staff. The tool helped people do the work, but did not remove the need to understand it.
What process-first AI means for operating costs
For companies adopting AI, the practical shift is to map the full customer journey before buying more software or scaling acquisition. Pavlova lists the checkpoints: entry point, data flow, responsible actors, customer-facing output, finance records and points where manual error can enter. Firms that fix this backbone first can cut rework across sales, scheduling and accounting, while small teams gain the most because one decision propagates without extra hires.
The limits deserve equal attention, especially as agent deployments grow faster than control mechanisms. IBM's 2026 research found only 11% of surveyed technology leaders felt completely prepared for large-scale AI-agent deployment. Buyers should therefore ask vendors where human approval is required, how anomalies are surfaced, what audit trail exists and how finance reconciliation is verified. The source does not promise that redesign alone guarantees profit, only that isolated automation leaves substantial value unused.
The marker to watch is whether workflow redesign starts appearing in AI project disclosures and vendor case studies alongside adoption rates. If the share of fully scaled deployments rises above the current 7% while stopgates and end-to-end ownership become standard, process-first design will have moved from advice to practice. That shift would show up in fewer reconciliation errors and faster order-to-cash cycles.
