Why isn't AI making your team faster?

Last month I watched a three-person team celebrate that every reply now starts in ChatGPT.

Research took minutes. First drafts took minutes. Summaries took minutes. Then the quote still sat for four days, because sales still waited on ops, ops still waited on the founder, and nobody owned the whole path from enquiry to send.

They had more AI than last year. The customer still waited the same.

That is the whole essay. AI tools make individual steps faster. A team gets faster only when someone redesigns the handoffs those steps sit between.

Why do most teams use AI and still feel slow?

Because adoption is almost free, and redesign is expensive.

McKinsey's State of AI: Global Survey 2026, fielded in May and June 2026 with 1,719 respondents, finds nearly nine in ten organisations now use AI in at least one business function. Eighty percent of respondents say AI improved their own productivity. Fifty percent say it helps them make better decisions.

The enterprise numbers barely moved:

A companion McKinsey piece, Why AI's easiest wins are misleading CEOs (7 October 2026), puts the gap plainly: adoption went from 88% to 89% year on year, and the high-performer share stayed at 6% both years. Thirty-seven percent report some positive effect on earnings. The gap between broad use and real payoff is not closing.

Why doesn't a faster step make the process faster?

Because the delay usually lives between the steps.

The same McKinsey commentary walks through a bank using AI to read documents for small-business loans. Document review might save two days. If the application then waits on separate handoffs among sales, credit, compliance and operations, the customer sees almost nothing. A faster first step does not fix the delays and errors that arrive later.

That matches what I see in small teams. The AI lives in the tab. The process still lives in Slack threads, "can you take a look?", and a founder who is the only person allowed to say yes.

A faster first step is not a faster process. The queue between people is usually longer than the work inside any one tab.

The studies executives love to quote are real, and they are also narrow. McKinsey cites a study of about 5,200 customer-support agents where AI assistance raised issues resolved per hour by 15%, and randomised trials with about 4,900 software developers where an AI coding assistant completed about 26% more tasks. Contact centres and software teams already had queues, outcomes and a way to check the work. AI made an existing system faster. Most small-business workflows do not have that system yet.

What are AI high performers doing differently?

They rewrite the workflow. They do not paste a chatbot into the old one.

In the 2026 State of AI survey, nearly three-quarters of high performers say they have fundamentally redesigned workflows because of AI, up from 55% the year before. Among everyone else, only about one quarter say the same. High performers are also more likely to chase growth or innovation alongside efficiency, to put a senior owner on the work, and to measure whether the change did anything.

McKinsey's October commentary adds a sharper early-adopter cut: among organisations still in the earliest stage of AI adoption, those that redesigned workflows were 5.3 times as likely to report enterprise-level value as those that did not — 32% versus 6%.

What should a small team do instead of buying another AI tool?

Pick one path the customer feels. Rewrite that path. Then put AI inside it.

A setup that fits a team of three to thirty:

A few rules that keep this honest:

• Measure the path, not the tab. Time from enquiry to send. Error rate on the final message. Not "hours saved writing."
• Write the rules where the AI can read them. Tone, price floors, what never goes out without a person. If the rules only live in someone's head, every draft will need them again.
• Keep one gate that is always human. Big client messages, payments, anything with legal or reputation risk. The agent needs a yes essay is the longer version of this.
• Do not scale agents until one process works. McKinsey finds smaller organisations still stuck at 22% scaling AI agents, flat year on year, while large ones moved from 27% to 40%. The missing piece is rarely the model. It is a process clear enough to hand to one.

So should you keep using ChatGPT?

Yes. Just stop pretending the tab is the system.

Eighty percent of McKinsey's respondents already feel more productive with AI. That gain is real. It is also the easy part. The hard part is the four days between a finished draft and a customer who has heard nothing.

In Who trains the junior when AI does the junior work? I wrote that research, summaries and first drafts have mostly moved to agents. The verbs that have not moved are checking, deciding and owning the path. Those are still human jobs, and they are the ones that decide whether the customer gets faster.

Write down one process. Give it one owner. Put AI on the dull steps. Then measure whether the customer waited less.

That is the system. The chatbot was only the first step.