July 29, 2026
July 29, 2026
Revenue up, cost down, or risk down. Name which one before you start.
The cleanest scoping test we have heard put on a stage, and the question we now ask before any engagement begins.
The cleanest scoping test we have heard put on a stage, and the question we now ask before any engagement begins.
Pauline Brunet runs forward deployed engagements at a fast-growing software company. She had twenty minutes on stage at the AI Engineer conference, and most of it was about how to keep a project honest.
Her test is three questions long.
“Am I increasing revenue? Am I decreasing costs? Or am I mitigating risks? That’s it. Every company, as complex as they are, that’s what they care about. Which one are you doing?”
Could be all three. At least one. If you cannot name which, you have not scoped the work. You have described an ambition.
The baseline is the part everyone skips
The second thing she does at the start of an engagement is agree what success looks like, out loud, in numbers, before anything gets built.
“If I automate this process from start to finish and now it takes 20 minutes instead of 3 hours, which is current baseline, is that sufficient for you? Does that measure success?”
Two things happen inside that sentence. Somebody has gone and measured the three hours. And the client has said yes or no to twenty minutes while there is still time to change the plan.
Skip it and the project ends in the worst available place. Something got built, everyone is fairly sure it helped, and nobody can say by how much. That project does not get a sequel.
What this looks like on our side
We scope directionally, never open ended. A named problem, a defined finish, and a baseline measured before we start.
Here is a real one. Inside a client business, one capability of one layer, running for two months:
1,241 completed runs. 207 hours of manual work removed. The rest of the stack was still switched off.
That number exists because somebody wrote down what the work cost before we touched it. It is the only reason the figure can be checked, and being checkable is the entire point.
The other half of the test
She was just as direct about refusing work. An embedded engagement is the wrong instrument for a company that already has its own engineers. They should buy tools and build it themselves, and someone should tell them so.
Saying who you are not for is the fastest way to show you know what you do.
Pauline Brunet runs forward deployed engagements at a fast-growing software company. She had twenty minutes on stage at the AI Engineer conference, and most of it was about how to keep a project honest.
Her test is three questions long.
“Am I increasing revenue? Am I decreasing costs? Or am I mitigating risks? That’s it. Every company, as complex as they are, that’s what they care about. Which one are you doing?”
Could be all three. At least one. If you cannot name which, you have not scoped the work. You have described an ambition.
The baseline is the part everyone skips
The second thing she does at the start of an engagement is agree what success looks like, out loud, in numbers, before anything gets built.
“If I automate this process from start to finish and now it takes 20 minutes instead of 3 hours, which is current baseline, is that sufficient for you? Does that measure success?”
Two things happen inside that sentence. Somebody has gone and measured the three hours. And the client has said yes or no to twenty minutes while there is still time to change the plan.
Skip it and the project ends in the worst available place. Something got built, everyone is fairly sure it helped, and nobody can say by how much. That project does not get a sequel.
What this looks like on our side
We scope directionally, never open ended. A named problem, a defined finish, and a baseline measured before we start.
Here is a real one. Inside a client business, one capability of one layer, running for two months:
1,241 completed runs. 207 hours of manual work removed. The rest of the stack was still switched off.
That number exists because somebody wrote down what the work cost before we touched it. It is the only reason the figure can be checked, and being checkable is the entire point.
The other half of the test
She was just as direct about refusing work. An embedded engagement is the wrong instrument for a company that already has its own engineers. They should buy tools and build it themselves, and someone should tell them so.
Saying who you are not for is the fastest way to show you know what you do.
