Cutler Says Let the Customer Pick the Score. Kill the Scorecard Menu.

John Cutler wrote the best post on the roster this year about the engagement as the product, and one recommendation in it gives away the thing a forward deployed firm cannot afford to give away. This is a Kill List entry: the scorecard menu, and what replaces it.

The short version

Cutler's November 5 post treats a consulting engagement as cooperative game design, from a team he describes as a mid-market Palantir, low-to-mid seven figures, outcomes in 3 to 6 months, "forward-deployed." The score should be capability milestones that retire risk, Gemini before Apollo, not a Gantt chart at a percent complete; the firm's incentive to extend the engagement rather than finish it should be named; both sides need people whose full-time job is the work. Then: prototype a few versions of the game, a milestone-and-RAG version, a North Star framework, a graded capability scorecard, and walk the client through them to see which they would play and which answers the boss. A score the customer picks from a menu is a score the firm can lose without anyone noticing, because the one chosen answers the boss, not the slip. In a forward deployed engagement the score was fixed at signature: the eval row set is the statement of work, done is the rows passing, the number is known at scoping (60 percent in Colin Jarvis's case), and the scorecard is four numbers, time to landing, reversal rate by day 60, generalization rate, and hours per deployment, each against the firm's own last quarter. The pull to extend is fixed by a contract line, hours falling over the term and the renewal price written before the pilot, not by naming it. Kill the menu; prototype the page the boss reads, not the number.

What Cutler published

The context first, because it matters for who he is writing to. Cutler works on a small team called Clonal inside a larger consulting firm, which he describes as a mid-market Palantir: operational intelligence, digital twins, disparate data sources combined for tangible outcomes, affordable at low-to-mid seven figures, with the goal of driving outcomes in 3 to 6 months. He names the industry term himself: forward-deployed. The post is one he wrote for his team.

The frame is game design. A good game has clear, attainable goals, a winnable level of difficulty, meaningful choices, timely feedback and visible progress, a compelling narrative, mastery over time, and social interaction. Consulting is cooperative game design with two parties who have different roles, incentives, and ways of working, and the trap is a one-sided game: the firm wants loose scope and light process because it suits the firm, the client wants dates and visibility because it suits the client, and both rationalize their preference as professionalism until trust goes and every move reads as adversarial.

Three self-serving assumptions he names: "we know what's best for you," "the customer is always right," and "follow the money," the last covering the perverse incentives nobody says out loud. His list for the firm: more incentive to extend the engagement than to make itself unnecessary, to say yes to scope changes than to push back, to produce visible activity than progress the client can feel, to protect the renewal than to deliver bad news early.

Then the Score, which is the best section. Before Apollo could land, Gemini had to prove two weeks in space, work outside the capsule, and rendezvous in orbit; Apollo then built up mission by mission. You can feel the risk coming out. Against that, a Gantt chart with arbitrary phases and a percent complete that tells you nothing except that you are N percent closer to finding out whether the thing works.

And the recommendations. Don't inflict ways of working the customer can't support. If you create a good score, follow through and hold yourselves to it, no gimmes. Be honest about incentives. Find people whose full-time job is the work. Don't assume expertise the customer doesn't have. Think hard about the promise. Then the one this post argues with, in his words: "Go design the game, and test it with them. Don't just default to your standard engagement model. Prototype a few versions of the game: a milestone and RAG version, a North Star framework, a graded capability scorecard, whatever fits. Walk the client through them and see which ones they'd actually play, and which ones answer 'So where is this thing at?' for their boss."

Where I get off

I want to be precise about what I agree with, because it is most of the post. The Gemini score is right. Naming the incentives is right. The full-time counterpart is right, and it is exactly the line in Buying From an FDE Company: Write the Move-Out Into the Contract that makes a named counterpart on the customer's side a deliverable.

The menu is wrong, and it is wrong in the one place a forward deployed firm cannot afford it.

A score the customer picks from three prototypes is a score the firm can lose without anyone noticing. The version a client chooses is the one that answers "so where is this thing at?" for their boss, because that is the test Cutler sets for it. It is not the version that would have caught the slip. A milestone-and-RAG strip goes amber when someone decides it goes amber. A North Star framework is a sentence. A graded capability scorecard is the Gemini score, and it is the only one of the three that can turn red on its own, which is why it is the one a client under pressure from their boss will not pick.

In a forward deployed engagement the score is not up for a vote, because it existed before the game started. The FDE Writes the Eval: Almost Right Has a Number is the rule: the eval is the deployment artifact, one versioned set of the customer's real cases with agreed-correct answers, written by the engineer before the build with the customer's experts supplying the labels. It is the statement of work, because done is the rows passing. The number is known at scoping, which is how Colin Jarvis's team could say 60 percent before building anything rather than discover it at go-live. That is Cutler's Gemini score with the customer's own data in it, and nobody chose it from a menu.

The score that was fixed at signature

The FDE Scorecard: Count What Stayed, Not Who Was Sent is the rest of it. Four numbers, each a trend against the firm's own last quarter because no benchmark exists. Time to landing: days from signature to the first change the customer makes without you. Reversal rate: the share of go-live workflows the customer switched off, overrode, or routed around by day 60. Generalization rate: the share of handback packets that got a written decision in two weeks. Hours per deployment, for the same class of customer, falling every quarter. The first two are the engineer's. The last two are the function's and product's.

None of those is a format. They are the number, and the AI Product Management thesis is that the part of the job that survives cheap building is deciding what the number is before anyone builds. The FDE Transition has the rest of the series.

Cutler's hardest incentive, the pull to extend rather than finish, is where the menu does its damage, because a firm that lets the client pick the score has also let the client pick the score it will be held to at renewal. Naming the incentive in a team post does not fix it. A contract line does: FDE hours fall over the term, by contract; the customer's team passes an independence test before the engineers leave; the renewal price is written before the pilot. When the hours are contracted to fall, the firm's incentive and the client's point the same way, and the score cannot be the thing that says otherwise.

What to prototype instead

Cutler's instinct to test the engagement design with the client is right. Point it at the page.

One screen, the fixed score, in whatever shape the customer's boss would actually open: a RAG strip driven by the eval pass rate and the day-60 reversals, a capability ladder with the rows passing at each rung, a trend line of hours per deployment. Prototype three of those with the client. Let them pick. The display is theirs. The number underneath is the same on all three, it was set at signature, and it goes red without anyone's permission.

Kill the scorecard menu. Keep the Gemini score, keep the named counterpart, keep the incentives out loud, and keep the prototyping for the page.

This week, if you run a forward deployed engagement: open the score the customer is currently looking at and ask whether it can turn red on its own. If someone has to decide, you let them pick the menu.

Related answer: Should the customer pick the scorecard for a forward deployed engagement?

Sources: John Cutler, "TBM 449: Game Design, Consulting, and the Engagement as Product," The Beautiful Mess, November 5, 2026, read in full; every description of Clonal, the game-design components, the incentive list, the Gemini and Apollo example, and the quoted recommendation are as Cutler published them. Falk Gottlob, "The FDE Scorecard: Count What Stayed, Not Who Was Sent" (the four numbers), "Buying From an FDE Company: Write the Move-Out Into the Contract" (the contract lines), and "The FDE Writes the Eval: Almost Right Has a Number" (the eval as the statement of work; Colin Jarvis's 60 percent at scoping as reported there).