Your people got faster. Your company didn’t.

Eighty percent of leaders say AI made them faster. The share seeing that reach company profit hasn’t moved in a year. That gap is the one a dealer principal feels but can’t name — and the AI ROI question behind it is an organizing problem, not a buying problem.

HEADLINE STORY

McKinsey published its annual State of AI survey on August 25, drawing 1,719 business leaders across 97 countries. Two numbers belong on your desk.

Eighty percent said AI has improved their own productivity — close to universal agreement that the tools work. Thirty-seven percent said AI has contributed to their company’s EBIT, essentially unchanged from a year ago. More companies are scaling it and more money is going into it, and that number has not moved.

That gap is the thing a dealer principal feels but can’t quite name. Your inside sales team says the tool saves them time, your marketing person is producing twice the output, and then you look at the P&L and nothing is different.

Individual speed doesn’t reach the P&L on its own.

If your quote desk saves twenty minutes on a bid and nothing else changes, you bought twenty minutes of unallocated time. It goes into more email and a less rushed afternoon. Neither shows up in gross margin.

The survey is specific about what separates the few who do see it. Among the 6 percent McKinsey calls AI high performers — companies reporting at least a 5 percent EBIT impact and significant value from AI — nearly three-quarters redesigned the work instead of inserting AI into the work they already had. Among everyone else it’s one in four. Redesign is a big word for something small. It means deciding in advance what the saved time is for. If the quote desk gets faster, do more bids go out, or the same bids with fewer people, or the same bids same-day instead of next-day? Pick one, then measure it.

WHAT ELSE MATTERED

The gap with your big competitors is opening in one place

The same survey splits by company size. Among companies above $1 billion in revenue, the share running AI that acts on its own inside a business function went from 27 percent to 40 percent in a year. Among smaller companies it sat flat at 22 percent. That’s you on one side of a line and BFS, US LBM, and QXO on the other.

But look at where the line falls. On basic AI use, the kind your people already do, the gap is much narrower. What blew open in twelve months is specifically software that runs on its own — the most expensive and least proven thing on the menu. Your large competitors are spending IT departments you don’t have on the part with the thinnest evidence. Close the basic gap first.

The job cuts didn’t happen

Last year’s survey found 32 percent expected AI to cut their headcount within twelve months. This year, 14 percent report it happened. Yet 39 percent now expect cuts in the coming year — a higher number than the prediction that already missed. If you need the person, hire the person.

THE PATTERN BEHIND THE WEEK

Eight in ten people say the tools make them faster. The share reporting any company-level financial impact hasn’t budged in a year. And the companies with the most money to spend are pouring it into the least proven part of the technology. None of that argues for waiting. It argues about where the difficulty lives.

This is an organizing problem, not a buying problem — and no purchase you make this year solves it.

That’s the good news, because organizing is what a 90-person dealer does faster than a company with 300 locations.

WHAT TO DO NOW

1. Pick one function and measure there, not company-wide. McKinsey found real financial impact at the function level even where company-wide profit was flat. For a dealer that means dispatch and delivery, inside sales support, or credit and collections. Pick one, set a number, check it in ninety days. Measuring company-wide is how this gets declared a wash.
2. Before you accept an AI feature from your ERP vendor this fall, ask what work it removes. Faster is what everybody already has, and faster is what isn’t showing up in profit. If the vendor can’t name work that goes away, you’re buying speed you’ll have to figure out how to spend.
3. Ignore the agent pitch until the basics are done. If your team still can’t reliably get a clean draft or a competitor’s number rebuilt from source, software that acts on its own isn’t your next purchase. The companies buying it have staff you don’t.

WHAT I’M WATCHING

• OpenAI’s new top-end model landed September 3. The announced ChatGPT rollout covers Plus, Pro, Business, and Enterprise — not Free or Go. If part of your team works off free accounts, you’re beginning to open a capability gap inside the same building.
• Build-versus-buy is live again at the enterprise level. Thirty-two percent of McKinsey’s respondents skipped a software purchase because AI let them build it themselves. It hasn’t reached the mid-market. Watch whether it does.

ONE QUESTION WORTH ASKING

If everyone on my team says AI is saving them time, where did the time go?

My answer, from running this myself: it went back into the same jobs, done a little more comfortably. That isn’t a failure, but it never became a result either. The result shows up when somebody decides what the time is for before it gets saved.