Imagine a sales rep drafting a quote in five minutes, only to have it sit in the approval queue for two days. That is the sort of gap worth looking for when someone says AI has made the team faster. McKinsey’s October 5 article on the AI operating model says 80 percent of respondents to its 2026 AI survey report improved individual productivity, while 37 percent report a positive impact on earnings from AI. Those are different measures, not a conversion rate. Still, the distance between a quicker task and a business result deserves attention.

If you are asking what this means for your business this week, look at one task your team says AI has made faster. Then follow the work until the customer, colleague, or account actually gets the result. The saved minutes may be sitting in a queue.

TL;DR

Time saved at one desk is potential capacity, not a realized return. Measure the whole path from request to finished outcome. Name the person who can change the handoffs, then decide where the reclaimed capacity should go. If nobody makes that decision, a productivity win can disappear into the workday without changing the business.

The productivity number has two different denominators

McKinsey’s 2026 State of AI survey says 37 percent of respondents attribute at least some impact on operating profit to AI use. That is not the same as saying the other 63 percent have failed. Some may not be measuring impact yet; some may have gains outside that metric. Nor does an employee’s report of feeling more productive prove that the business saved an hour on its books. The survey numbers describe different outcomes reported by respondents, not a controlled before-and-after test of the same teams.

The quote example is illustrative, not a finding from the survey. Its point is that the delay may be nowhere near the person using AI.

The temptation is to count the time saved by each user, multiply by a salary rate, and call the result ROI. That calculation assumes saved time is recoverable and directed toward something else the business needs. Neither follows automatically. Fifteen minutes freed in fragments across a day may make work less frustrating without allowing a company to serve another customer, reduce overtime, or close its books sooner. Relief matters. It just needs a different label from measured financial return.

AI can save a person’s time without saving the organization’s time.

Trace one piece of work across the boundary

Consider a service team that uses AI to draft responses to incoming requests. The old process takes a person ten minutes to assemble a first draft; the new one takes three. That seven-minute difference is a useful observation, not yet the answer. Does the response still need the same review? How long does it wait for an account detail from another department? How many replies come back because the first answer was incomplete? What happens to the staff time no longer spent drafting?

Take a small set of actual requests and record when each arrived, when a response went out, and whether the issue was resolved without a second round. Keep the messy cases. Compare with similar requests handled through the old path, not the best-looking AI example.

Then ask the team’s manager what outcome is worth buying. If the queue is growing, use the recovered time to handle more requests without extending the day. If turnaround time is the problem, change the review schedule so the faster drafts do not pile up at one approval point. If quality is uneven, spend the extra minutes checking the answer against the customer’s actual record. These are different uses of capacity. Each needs a different measure.

The gap becomes a habit

McKinsey’s October 5 article says only 11 percent of organizations in its AI readiness research have reached what it calls the “reinvention” horizon: rethinking how work gets done and how value is created. That figure is from a separate readiness assessment, so it should not be merged mathematically with the 80 and 37 percent survey results. It does, however, make the operating choice visible. A company can distribute AI access broadly while leaving its approval routes, staffing assumptions, and service promises intact.

Waiting for a better model will not fix an approval queue. Nor will another round of prompts tell you whether the customer got an answer sooner. Someone has to time the work on both sides of the handoff.

Give one process owner a narrow assignment: choose a recurring request, establish the old end-to-end time and error rate, let the team use AI for the part it can inspect, and track the finished result for several weeks. Agree in advance what would count as better and what would make you stop. Ask the people receiving the handoff whether your “faster” work arrived any earlier.

A team may discover that AI helps employees breathe but does not change throughput. That is honest information, and perhaps enough reason to keep the tool. If the goal is a business return, decide where the breathing room goes and measure what reaches the customer. You might find the next bottleneck is a manager’s inbox.

Research and structure: Mai. Direction and voice: John Lipe. Sources: McKinsey, “AI is changing work. Now it has to change the organization,” October 5, 2026; McKinsey, “The State of AI: Global Survey 2026,” August 25, 2026.