04 / AI AND THE STACK
Not a tool recommendation and not a workshop. Documented workflows with the actual prompts, written for your buyer and your process, taught until people use them without being asked.
SOUNDS LIKE YOUR TEAM
“We bought the tools and nothing measurably changed”
“Reps still keep the real pipeline in a spreadsheet”
“Half the stack goes unused and I cannot tell you which half”
“I cannot tell the board what is working, what is not, or why”
WHAT IS ACTUALLY BROKEN
Most teams bought the tools before they defined the work. The tools then did the undefined work faster.
If stage three means whatever the rep updating the record decided it meant, then a model trained on that pipeline will forecast confidently off it. Confidently, wrongly, at speed, with a chart.
If your ICP is a guess, AI will now execute that guess a thousand times a week, politely and at scale, into a market that will remember you for it.
If nobody ever defined what good discovery sounds like, an AI call summary is a transcript with an opinion attached. You cannot coach against output nobody set a standard for.
None of that is a tooling failure. Every one of those companies bought a competent product and pointed it at an undefined process.
AI does not fix a short column. It loads it faster.
Artifacts, not recommendations. Every one is a thing your team holds and runs after I am gone.
WHERE IT GENUINELY EARNS ITS PLACE
The gains are real and they are almost all in one category: work your team already skips because it takes too long.
Prospect research. Forty minutes of account work compressed into three, so a rep actually does it instead of winging the first call. Meeting preparation. The account history assembled before the call rather than remembered during it. Win-loss reading. Two years of closed-lost notes read in one pass. Call analysis. Not transcription: the pattern across thirty calls. Signal monitoring. A weekly sweep of a target list that returns only what changed. First-draft writing, from a real brief, where the human supplies the point of view.
WHERE THE HUMAN STILL CARRIES MORE WEIGHT
Three things, and they are the three that decide whether you hit the number.
Discovery is judgement under uncertainty. A model produces a confident, plausible summary of a conversation it did not have to steer. Knowing which thread to pull, and when to stay quiet, is not a summarisation problem.
Coaching works because somebody was watching. Specifically, somebody whose opinion the rep cares about. Automate the observation and you have removed the part that changed the behaviour.
Deciding which accounts deserve your team’s time is the highest-leverage call you make all quarter, and it is the one most often handed to a scoring model. If a rep cannot say why an account scored 82, they will not work it. An opaque score is a number, not a decision.
HOW IT ACTUALLY GETS TAUGHT
It starts with one prompt. Not a platform and not a strategy. A single prospecting prompt, written for your company, your industry and your buyer, that pulls everything a rep needs to walk into a first call prepared and dangerous.
Then the part that makes it stick: the rep changes one field. The company name. Everything else is already right, because it was written around your business rather than downloaded. Research nobody performs is not a capability.
Not a generic library, because a generic library is a bookmark nobody opens. And no prior knowledge assumed. Half the teams I work with have done nothing and feel behind. They are not. Nobody is as far ahead as the conference agenda suggests.
WHAT YOU ARE LEFT HOLDING
The workflow documentation with the prompts, the recorded sessions, the written decision boundary, the stack audit with its recommendation, and the CRM configured to match the process.
NINETY DAYS OF REVISIONS
Models change, tools change, and a prompt that worked in March needs adjusting by June. For ninety days after handover that is part of what you already paid for.
THE FRAMEWORK
None yet. This is the newest part of the practice and I would rather name a framework once it has earned it than invent one to fill a slot.
THE ARTICLES THAT PROVE IT
Being written. This component currently has the least published work behind it, which is worth saying plainly rather than padding the page with adjacent material. Next: Most companies are AI-curious. That is the expensive one.
BEST FOR
Teams who have tried the tools and suspect the problem is upstream of them, and teams who have done nothing and would rather start correctly than start fast.
NOT FOR
Anyone looking purely for a tool recommendation. The recommendation is downstream of the process, and the process is the work. If you want a shortlist, there are analysts who sell that and they are cheaper than me.
Run the Six Reads. Six questions, three minutes, and it names which of six problems you actually have.
THE OTHER THREE PARTS
The next step
Thirty minutes, no deck and no discovery script. Tell me what the last two quarters actually looked like and I will tell you which of the six you are dealing with, and whether I am the right person to fix it.