If AI can produce the deliverable, what are clients paying you for?

Clients are paying for diagnosis, and they always were. AI answers the question it's handed, confidently and well, including when that question is the wrong one. Research on AI-assisted professionals shows the failure lands exactly there. The deliverable was the visible part of the work rather than the valuable part.

First, the strongest version of the panic

Let me steelman this properly, because the reassuring version of this article is everywhere and most of it is written by people whose income depends on the answer.

A strategy consultant can now watch a model produce, in an hour, work that used to take weeks. Market assessments. Competitive benchmarking. A five-year model with sensitivity analysis attached. Not flawless, but in the region of 80-something percent, and close to presentable once it's run through a house template. That isn't a demo. That's Tuesday.

And the sharpest objection to any comfort I might offer is this: senior people at elite firms have said for years that their most valuable skill was pattern matching. Recognising that this situation resembles eleven previous situations and knowing how those went. Pattern matching across an enormous corpus is precisely what these systems are built for, and they do it faster and without arithmetic errors.

So if your expertise was pattern matching as a service, that really is over, or at least it's headed for commodity pricing. Anyone telling you otherwise is selling reassurance.

Sit with that before reading on, because everything below is worthless if it's built on denial.

What did the deliverable ever prove?

A deliverable is the visible residue of thinking. It's what got produced, not what got decided.

In professional services that residue has been doing a lot of unearned work for a long time, because clients can't directly inspect judgment. They can inspect a document. So the document became the proxy for the thinking, and for plenty of firms the document quietly became the product. Anyone who has sat through a forty-slide deck delivering what two honest pages would have delivered knows this already. The volume was evidence of effort rather than communication, and it got priced accordingly.

What's happening now is that the cost of producing evidence-of-effort has collapsed toward zero. Terrible news if your value was the artifact. Clarifying for everyone else, and uncomfortable in a specific way, because it removes the place where thin thinking used to hide.

Every collapsed moat has done this. Geography protected firms until the internet took it away. Access to information protected them until search took that. Each loss removes a source of protection that had nothing to do with being good, and leaves being good as the remaining option.

Where does AI actually fail, and is that reliable?

This one isn't a matter of opinion, which is fortunate, because opinions on it are cheap right now.

In 2023 a team led by Fabrizio Dell'Acqua ran a field experiment with 758 consultants at Boston Consulting Group, roughly 7% of the firm's individual contributors. The study gave some of them GPT-4 and some nothing, across two categories of task. On tasks inside what the researchers called the jagged technological frontier, the AI-assisted consultants were substantially better: faster, higher quality, more output.

On a task that sat outside that frontier, the consultants using AI were about 19 percentage points less likely to reach a correct solution than the consultants working without it.

Read that twice. The tool made capable professionals worse than they would have been alone, which is a stronger result than simply failing on a hard problem. It produced confident, well-structured, plausible work for a situation it had misread, and the humans went along with it.

The reason isn't mysterious. The frontier is jagged, which means you can't tell from the outside whether a given problem sits inside it or outside it. Two questions that look equally difficult can fall on opposite sides. Knowing which side you're standing on is not something the model can do for you, because that determination happens before the model starts.

What does diagnosis actually look like?

The clearest example I have isn't from business at all.

Years ago, when I was a youth pastor, a family brought me their son. He was in about grade five and he was having violent outbursts several times a week. People were getting hurt. Things were getting broken. There were angry words. It was serious, and they wanted help with a boy who they understood to have a behaviour problem.

That was the brief. Fix the behaviour.

One of the first things I asked him was when it was happening. Almost always after school. And the detail that stopped me was this: we knew this kid, and he was a shy introvert. That's not the profile you expect. A boy who is quiet and withdrawn everywhere else, exploding at home, on a schedule.

So I said, think about how you are after school. Kids are dehydrated by then. They haven't been drinking enough and their blood sugar is low, and both of those things affect your capacity to regulate your emotions. On top of that, if there are chores waiting the moment he walks in, he's exhausted. And for an introvert, six hours surrounded by people is depleting in a way that doesn't show until it comes out sideways.

The plan we put in place was almost embarrassingly small. When he got home there would be a glass of water or juice waiting, and a healthy snack. Nobody would talk to him. The rule was that he had to drink and eat, and then he got half an hour to do whatever he wanted. Television, video games, Lego. Entirely up to him. He was in control of that part.

Several months later I ran into his mum and asked how things were at home. She said we'd gone from several outbursts a week to a handful a month.

That boy didn't need more discipline and more consequences. He needed fuel and recovery, and once he had the tools he could manage his own behaviour. The violence was a symptom of something physiological and situational, and everybody involved had been treating it as the problem itself.

Now hand that original brief to a capable model. My son is having violent outbursts several times a week, give me a discipline plan. You will get an excellent discipline plan. Structured, evidence-informed, well-organised, and aimed squarely at the wrong thing. It will not ask when. It will not notice that a shy child exploding on a schedule is a contradiction worth chasing. It answers the question it was given, beautifully.

There's a comfortable lie sitting underneath that brief, too. In a lot of families, and particularly in conservative faith communities, the reflex is that tough love is the answer: more discipline, more consequences, more responsibility. Sometimes that's right. In this case it would have made a hungry, exhausted, overstimulated child worse, and everyone would have concluded he was defiant.

Why does this make judgment more valuable rather than less?

Because clients have never been able to buy diagnosis directly, and now they have to.

Think about what a founder actually brings you. Almost never the real problem. They bring the presenting problem, the symptom that finally got loud enough. Leads are down. The team keeps turning over. The pricing feels wrong. A competent advisor's first job is to work out whether the presenting problem is the actual problem, and it usually isn't. A complaint about lead flow is frequently a positioning failure a layer up. A request for a sales script is frequently a capacity problem in disguise.

Hand a model the presenting problem and it will do outstanding work on it. That's the risk. It won't push back on the brief, because it has no relationship with the business, no memory of what the founder said six weeks ago in a different mood, and no stake in what happens in year three.

This is also why the proven-process business is in more danger than the judgment business, which I find satisfying, because it was always the weaker offer. If your value proposition is a repeatable methodology applied uniformly, you've described something a model can now generate on demand and tailor faster than a junior associate. The firms most exposed to AI are the ones that had already turned thinking into a formula. Chess has principles. It doesn't have a formula, and neither does a business.

What that means practically is that a practice organised around producing volume needs a different shape from one organised around judgment: fewer clients, a real diagnostic step, and a service model where the thinking is the paid product rather than the free preamble to a proposal. Plenty of firms have the second ambition and the first operating model, and the gap between them was survivable while artifacts were expensive.

Putting it to work

Take your last three engagements. For each one, write down the problem the client stated when they first made contact, and the problem you actually ended up solving.

If those two are identical all three times, sit with that. Either you're in genuinely well-diagnosed work, or you're taking briefs at face value, and the second one is now a commercial risk rather than a professional habit.

Where they differ, write down the moment the gap appeared. What did you notice? Was it something they said, something they avoided, or something that didn't match the numbers? In the story above it was a scheduling detail and a personality that contradicted the complaint. Yours will be equally small and equally specific.

That moment is your product. Most professionals have never described it to a client, because it happens invisibly and feels like just doing the job.

Then do the uncomfortable part. Take the stated problem from one of those engagements, hand it to a capable model, and read what comes back. It will be competent, well-organised, and aimed at the wrong target. Keep the output. It's the clearest demonstration you'll ever have of what you actually sell, and it's more persuasive in a sales conversation than anything you could write about yourself.

One last habit worth building: when a situation doesn't match the pattern, slow down instead of speeding up. The contradiction is the signal, and it's the exact place the tools are least reliable.

Sources

• Dell'Acqua, F. et al., "Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality", Harvard Business School / BCG, 2023. 758 BCG consultants; large gains inside the frontier, and roughly 19 percentage points lower likelihood of a correct solution on a task outside it.
• Navigating the Jagged Technological Frontier, Organization Science (peer-reviewed publication of the study)

Frequently asked

Will AI replace consultants and accountants?

It will absorb the parts of the job that are process applied to a well-specified brief, which is a large share of current billable output. What it doesn't do is decide which brief is the right one. Practitioners whose value was execution are exposed; those whose value is diagnosis are not.

Isn't "judgment" just a comfortable thing for advisors to claim?

It would be without evidence. The BCG field experiment supplies some: on a task outside the model's frontier, AI-assisted consultants performed meaningfully worse than unassisted ones. That's a measurable failure in exactly the place judgment is supposed to matter.

How do I show a client I'm worth more than what they could generate themselves?

Make the diagnostic step explicit and paid. Show them the gap between the problem they described and the problem you found. Running their original brief through a model and comparing it with what you actually recommended is an unusually direct demonstration.

Should I be using AI in my own practice?

Almost certainly, for the work that sits inside the frontier. The risk isn't using it, it's failing to notice when a task has crossed the boundary, because the output looks equally confident on both sides. Treat the situations that don't fit the pattern as the places to slow down.

Ready to look at the architecture honestly?

Book the conversation. We'll tell you what we see, and whether the work we do fits where you are.

Book the conversation →