I left McKinsey six weeks ago to set up Stenfert Kroese Consulting independently. The first month and a half has been spent in three rooms: a private equity firm thinking through AI in deal sourcing, a multinational reshaping how it responds to high-volume tenders, and an industrial group rethinking how it sells through channel partners.
Different sectors, different commercial questions. One pattern keeps surfacing, and it is not a technology pattern. It is a human one.
Why the wrong question gets asked
New technology always brings discomfort. Discomfort produces defensive thinking. The natural reflex is to ask "how does this make my current work easier?" rather than "should my current work exist?"
The first question feels safe. It preserves the shape of what we know. The second question is unsettling, because it questions the value of the work itself, and by extension, the team doing it.
This is not a failure of intelligence. It is a feature of how humans encounter change. But it produces predictable outcomes: a list of small productivity gains, a roadmap that mirrors the org chart, and a slow erosion of the original AI ambition into a series of accelerations that change nothing structurally.
The question I keep hearing
Every executive I have spoken to about AI starts with a version of the same question:
"How do we use AI to make our analysts, sales reps, engineers, or proposal team faster?"
It is a reasonable question. It is also the wrong one.
The reason it is wrong is that "faster" is bounded by the size of the existing task. If a proposal takes a team five days and AI halves it, that is two and a half days saved. Useful. But the proposal still exists. The team still exists. The cycle of writing, reviewing, and submitting still exists. You compress the same shape.
The companies pulling away from the pack are asking two different questions.
The better questions
"What entire workflow should no longer exist?"
"If we were designing this business today, with no legacy, would this workflow even be here?"
Both force a different conversation. Not a productivity conversation. A structural one.
At the multinational running high-volume tenders, the original question was not "how do we write tenders faster?" It was "should writing tenders be a discrete project at all, or should responses be generated continuously from a live system that absorbs every RFP we have ever submitted?" The first framing produces a 30 percent speed-up. The second produces a function that responds in hours, with the human role moved entirely to qualification and pricing judgment.
At the PE firm, the question was not "how do our analysts screen deals faster?" It was "should screening be a discrete project, or a continuous signal stream that surfaces companies before they appear on a banker's pitch list?"
Same pattern: the workflow itself comes into question, not the speed of doing it. And once you start asking the second question, the greenfield one, you notice that several of your most expensive teams exist because of a legacy you would never recreate today.
Three questions to run before any AI initiative
When I am brought into one of these conversations now, I run the team through three questions before talking about technology.
1. What is the unit of work today, and why does it exist? Be specific. Not "sales." A proposal. A pitch. A pricing exception. A hire. A deal screen. An incident ticket. Then ask why it exists. Often the answer is historical: an org chart from a different decade, a system that used to require human translation, a customer expectation set by a slower industry.
2. If we were starting this business today with no legacy, would this unit still be here? This is the greenfield question. It strips away the comfort of "this is how we have always done it" and forces a clean look. The honest answer is often no, but the work continues because no one has had the standing to question it.
3. If it disappears, where does the constraint move to? AI does not remove constraints. It moves them. If your proposal team is no longer the bottleneck, your qualification team probably is. If your screening analysts are no longer the bottleneck, your investment committee is. Know where the constraint is moving, so you can plan for it before the AI delivers what you asked for.
The Pattern
"Faster" is bounded by the size of the existing task. "Should this exist?" is bounded only by the imagination of the leadership team. The companies producing real AI value are doing the second question well, not the first one quickly. AI is not a tool you add to your business. It is a reason to re-examine the business you would now design.
Why most companies will not do this
The honest answer is that the question crosses org-chart boundaries, and that is uncomfortable.
A head of proposals can authorise a faster proposal team. She cannot authorise "proposals no longer exist as a discrete step." That decision spans sales, operations, and commercial leadership. It needs a sponsor at the executive table.
This is why most AI roadmaps look like a list of functional accelerations rather than a portfolio of process eliminations. The roadmap mirrors the org chart. That is a fixable problem, but the fix is not technical.
What this means
If you are about to commission an AI initiative, the most useful thing to do before spending a euro on technology is sit with the three questions above for a week. Ask them about your most expensive process. Ask them about the process that frustrates customers most. Ask them about your largest functional team.
The answers will tell you whether you are about to compress an existing shape or change it.
The first is helpful. The second is where the value lives.