Your client arrives with an answer

Why this matters to you now

Three researchers tracked 5,179 customer support agents whose employer rolled out a generative conversational assistant in staggered waves. Erik Brynjolfsson, Danielle Li and Lindsey Raymond published the result in Generative AI at Work: issues resolved per hour rose 14% on average. The average is the least interesting part. The least experienced and lowest-skilled agents gained 34%. The most experienced and highest-skilled gained almost nothing. The tool was not making the best people better. It was spreading their practices to everyone else.

Read that sentence thinking about your client instead of yourself. Inside that organization, the two-year analyst now produces what the six-year analyst used to produce. The first draft exists before you arrive: the data is gathered, the spreadsheet is built, the document is formatted. And when you bill days of analysis, that is precisely the layer you are selling.

Edition #31 on the delay your client buys looked at this technology from your side of the table: rare knowledge levelled, the large firm stripped of part of its advantage, turnaround becoming your argument. All of that still holds. What nobody looks at is the same tool in the hands of the team sitting across from you.

The effect is not symmetrical. On your side, it makes you faster at work you already knew how to do. On theirs, it creates a capability that did not exist, and that capability goes straight at the first half of your offer.

If your proposal is described in days, it describes the part your client has just learned to do.

Pricing: what degrades is worth more than what improves

The move: Find the two or three places in your offer where a generative tool makes the work worse, not better. Put your price there, and detach it from the volume you produce.

Why it works: A preregistered experiment split 758 Boston Consulting Group consultants, about 7% of its individual contributors, between a group with no generative model and two groups with one. Fabrizio Dell'Acqua and his co-authors published the results in Organization Science, under the title Navigating the Jagged Technological Frontier. On eighteen tasks inside the model's frontier, built around launching a footwear product, the consultants with the tool completed 12.2% more of them, 25.1% faster, with quality rated more than 40% higher. Then a single task deliberately chosen outside the frontier: a management call that had to be made from quantitative data and interviews that contradicted each other. There, the consultants with the tool were 19 percentage points less likely to produce a correct solution. The control group got it right 84.5% of the time; the two AI groups, 60% and 70%.

The word that matters in that title is "jagged." The two halves of the experiment looked similar in difficulty. From the inside, nothing told you which side of the frontier you were standing on. Your client cannot tell either. So your price is not justified by what their team produces faster. It is justified by what their team produces with confidence it has not earned.

The trap: Dropping your price by the share of the work the client did. What is left is not proportional to what disappeared. Validating a document a tool produced costs more than writing it: you have to hunt for what the tool could not see, and you have to put your name under the answer. It is the flaw edition #24 on the offer that bleeds you described, with a new trigger: this time the client is the one offering you the discount, and they think they are doing you a favour.

This week: Open your last proposal and read it deliverable by deliverable. Beside each one write a single word, improves or degrades, based on what a generative tool would do to that task. Then reprice using the "degrades" column only. If the total collapses, your offer is still described in days.

Sales and business development: ask what they have already produced

The move: Before you write a proposal, ask one question, and ask to see the output. "What has your team already produced on this with its own tools, and how did that go?"

Why it works: In Mastering the Complex Sale, Jeff Thull argues that a client almost always arrives with a self-diagnosis, and that the consultant who accepts it as given sells the wrong thing, sometimes very well. What changed is not that the self-diagnosis exists. What changed is its confidence. A diagnosis improvised in a meeting invites discussion. A diagnosis that is formatted, structured and footnoted looks like a conclusion. You are no longer negotiating against a hunch. You are negotiating against a document.

The trap: Correcting the document in front of them. You give away your value in the meeting, for free, and you make the team that produced it look bad, the same team that will shape the decision. This is the 2026 version of the problem in edition #28 on the diagnostic you give away, and it costs more: before, you gave away your diagnosis. Now you give away your judgment of theirs, which is exactly what they cannot get anywhere else. I take this to be the hardest rule in this edition, because it asks you to stay quiet at the exact moment you are most useful: read, take notes, and do not correct anything in the room.

This week: On your next discovery call, ask the question and write the answer down as given, without commenting on it. You are after two facts: how far they got, and who inside the organization did it. Do it three times before you judge what it changes.

Collaboration networks: the map nobody has alone

The move: Build a group of three to five peers who serve sectors other than yours, and compare one thing only: where, at your clients, the work done in house stops.

Why it works: Etienne and Beverly Wenger-Trayner describe communities of practice as groups that learn by sharing cases rather than rules. What circulates in a group like that is not information, it is judgment about real situations. And the frontier that BCG study describes is jagged, which makes it impossible to map from a single vantage point. You may see fifteen clients a year. Five consultants pooling observations see seventy-five, across sectors that do not move at the same speed.

The trap: The group that talks about tools instead of clients. Naming models, comparing software, trading tips: that is the easy conversation, and it expires in three months. What does not expire is the exact point where a client stopped, and the reason they gave. My bet is that a group built on tools does not survive its third month: the novelty runs out, and there is nothing left to compare.

This week: Write to three peers. One question, the one from the previous section, applied to their clients. Promise them the synthesis of what you get back: that is what makes all three answer.

Value creation: start where their document stops

The move: Rebuild your deliverable so it starts where theirs stops. Cut the context section, the methodology recap, and the summary of what they told you.

Why it works: In Flawless Consulting, Peter Block makes a case that unsettles anyone who sells reports: your product is not the document, it is the client's capacity to decide and to own the decision. Once their team has produced the first draft, they already own the description of their problem. What they are missing is someone who settles it and puts their name beside the recommendation. Your method is your product, said edition #6. The question is no longer whether that is true, it is which part of the method survives an equipped team: not the order of the steps, which can be copied in an evening, but the judgment that decides which exception counts.

The test: If your deliverable differs from theirs mainly in layout, you added nothing. Edition #30 on the value you bury showed how consultants hide their contribution under volume. Volume just lost its last alibi: their team now produces as much of it as you do, and faster. I doubt a context section has ever proved anyone's rigour. It proves you listened, which is not the same thing and does not command a fee.

This week: In your next report, cut the context section and open on the recommendation. Leave everything else alone. If someone asks for the context, you will add it as an appendix, which is exactly what it is worth.

AI: know what your client can do before you write the proposal

The move: Measure your client's frontier before you set your price, not after they tell you where they stand. Five uses for that.

1. The capability sweep. Before a first call, gather public material only: the client's website, its posted job openings, its annual report, its communications. Ask for a sweep of the signs that internal analytical capability already exists, the roles posted, the titles, the tools named. You will know which layer they are on before they tell you.

2. The test of your most expensive step. Take the step you bill the most for and have a model run it, on a case you know by heart. Read the output not to use it, but to measure what it misses. That is your own frontier, measured on your own method instead of assumed.

3. The first-draft reader. On a document the client handed you, ask for everything it asserts without evidence: numbers with no source, causes presented as facts, recommendations with no trade-off. That document belongs to your client. Submit it only with their agreement, and strip out anything that identifies people or third parties.

4. The proposal reviewer. Have your own proposal read with one question: which sections could an equipped internal team produce on its own? What comes back is the list of what you can no longer sell at last year's price.

5. The three wordings. Draft three ways of asking the question from the previous section. You are looking for the one that does not sound like an interrogation, because the person who produced the document is often in the room.

The warning: Do not confuse a client's capability with their results. In The Simple Macroeconomics of AI, Daron Acemoglu estimates that AI will raise total factor productivity by no more than 0.66% over ten years, and by less than 0.53% in his more conservative estimate. The gap between what an organization believes it has automated and what has actually changed in its results is wide. The client who tells you their team now does the work is often fooling themselves. Do not sell against their fear, and do not sell against yours: sell against their document.

This week: Use 2, on one step. Thirty minutes.