From AI Pilots to Workflow Design
Connect AI pilots to business outcomes by designing workflows, decision rights and human review together. Practical questions for leadership teams.
Three AI tools support a customer request: one summarizes it, another drafts a response and a third retrieves information. The customer still waits. Approval queues, missing data and handoffs between teams have not changed. Evaluating an AI pilot sometimes means stepping away from the model and examining the whole job.
McKinsey’s August 2026 management playbook examines 20 companies that created economic value through AI-enabled transformation. Two-thirds of this selected group concentrated on three or fewer business domains. These are successful examples, not a representative estimate of everyone’s chances of success. They nevertheless provide a useful starting point for discussing connected workflows and organizational capability. Source, pp. 2–5.
Put one workflow on the table
Map the process from its trigger to a completed business outcome. At every stage, identify the information needed, the decision being made, its owner and the output passed forward. Add tool names last.
Consider proposal approval. Understanding the request, setting a price, assessing risk and responding to the customer are connected activities. Accelerating one step can also mean moving incorrect information faster. The unit of design should, where possible, be the complete customer outcome.
Clarify three handoffs
From people to AI: What context can the system access? May it proceed with incomplete information? How will the organization’s approved-tool and data rules apply?
From AI to people: Which outputs require verification? Does the reviewer have the time and authority to challenge alternatives, or merely a button to approve them?
Between teams: Who takes over when an exception occurs? Where is the decision recorded? Who remains responsible for the response the customer receives?
These are design questions we recommend asking. A software feature list cannot answer them for your organization.
Define a stopping rule before expanding
A pilot needs criteria for stopping as well as succeeding. If misrouting exceeds an agreed quality threshold, for example, requests might return to human review. The team should set that threshold using actual business costs and baseline evidence, rather than borrowing a universal number.
The purpose is a workflow with connected responsibilities and observable improvement. Our Applied AI for Managers program brings practical tool use together with these managerial questions. Start with a workflow small enough to examine in detail and important enough for its improvement to matter.
