Say what is true, including what you do not have. Then lay down a position you can defend: what you believe about where this is going, what would prove you wrong, and what follows for the buyer if you are right. Use the evidence that does exist for the questions it can answer, and agree a thesis with the buyer for the rest.
The question, asked about AI
Who else is running this in production, at our scale, for long enough to show it worked?
The question, as an AI buyer asks it. Page 20 in the reader
The honest answer is almost nobody. The technology buyers are asking about reached most companies in late 2022. The AI models have been replaced several times since.
Nobody has done this long enough to prove it. Here is what we believe is true about where this is going. Here is what would prove us wrong. Here is what follows for you if we are right. Tell me where you disagree.
The answer the book offers. Page 20 in the reader
That answer does not point at anyone. It lays down a position and offers to defend it. The old word for a position laid down to be defended is a thesis.
Why the reference is missing
Three things broke at once.
- The technology is young. Generative AI reached most companies after November 2022. A real reference takes time to grow.
- The ground keeps moving under the stories that do exist.
- Most of the stories are pilots.
So when the buyer asks who else is doing this, the seller has three bad options: a story that is too young, a story that is out of date, or a pilot wearing a production badge.
What the missing reference does not take away
What is usually missing is one kind of evidence: a customer like this buyer, who made this decision, on this horizon, and ran it long enough to report the result. The large bet is the part nobody can point at yet. A great deal of other evidence is still there.
Questions about the product, like whether it stays up for twenty thousand employees on a Monday morning, get an answer from evidence: a load test, a deployment at that scale, a trial designed to prove it. The thesis is for the part no evidence can reach yet.
There is a simple way to tell which of the two a decision needs. If the decision is small, easy to reverse and short in horizon, run the experiment and skip the thesis: a team buying twenty seats of an assistant for a quarter does not need a memo about the future of its industry. The thesis earns its place when the decision is large, hard to undo and long in horizon, such as the foundation the company's work will be rebuilt on, because there no experiment can reach the end of the bet.
A compass, not a map
A thesis is that compass. It tells a buyer which direction is right before anyone has walked the route, and it can be checked and corrected along the way. It is how decision makers have to decide now, and it is what this book teaches you to bring them.
What to do this week
- Date your stories. Take the three customer stories you use most. Next to each, write when the customer made the decision, and what AI model and price it was made on. Ask whether that customer would make the same choice today.
- Sort them. Mark each story as production or pilot. Count how many are in production with a measured result.
- Answer the question straight, once. The next time a buyer asks who else is doing this, say what is true, including what you do not have. Watch what happens in the room.
- Write one sentence. Finish this line: One thing I believe about this market that would still be true if my company disappeared tomorrow is... Keep it. Chapter 5 will test it.