What to promise, and what not to
The honest edge of the promise, drawn from things this org actually does wrong — because a partner who knows the limits is the one clients keep.
You can now show twelve minutes that make a business want this. The last thing to learn is where the edge of it is, and the reason is not modesty: the fastest way to lose a client in month three is to have been slightly too confident in month one.
What you can promise, flatly
- Nothing is lost. Every lead that arrives is captured with its source, once, without a second record. You watched it happen in beat three.
- Nothing waits unnoticed. Enquiries nobody has touched surface as a list of names every morning, biggest first.
- The reasoning is readable. Every score, every route, every message the AI wrote is a run you can open and a sentence somebody typed. There is no black box to apologise for.
- Their words, not ours. The stage descriptions, the scoring rubric, the refusals, the follow-up copy — all editable by their team, in English.
- Fast to start. There are
5workflows in this org that need nothing from anybody to run today.
What to say carefully
| They will hear | Say instead |
|---|---|
| “The AI will answer customers” | It answers from documents you approve, and refuses what you tell it to refuse — including rates and approvals |
| “It will call your leads” | It will, once you have a number, and somebody must be willing to answer the callbacks it generates |
| “It scores every lead” | Against a rubric your team wrote and can argue with. It is a shortcut to a conversation, not a verdict |
| “It's live next week” | Parts of it are live next week. WhatsApp is live when your business verification clears, and that queue is not ours |
“The AI will answer customers”
- Say instead
- It answers from documents you approve, and refuses what you tell it to refuse — including rates and approvals
“It will call your leads”
- Say instead
- It will, once you have a number, and somebody must be willing to answer the callbacks it generates
“It scores every lead”
- Say instead
- Against a rubric your team wrote and can argue with. It is a shortcut to a conversation, not a verdict
“It's live next week”
- Say instead
- Parts of it are live next week. WhatsApp is live when your business verification clears, and that queue is not ours
What not to promise, ever
- That the AI will not be wrong. It will. The promise is that when it is, you can see exactly which step did it and what it was given — and that it was bounded by a sentence you can change.
- That it replaces anybody. Everything in this org creates work for people: tasks, call-backs, chases. It removes the deciding-what-to-do-next, not the doing.
- Numbers you have not measured in their business. Conversion lift, response times, hours saved. You have this org's numbers, not theirs.
- That it works untouched. Thresholds, stage names, message copy and rubrics are all theirs to set, and all of them are wrong on day one by definition.
The four things this org itself gets wrong
These are real, and they are in the org you just demoed. Knowing them is what separates a partner from somebody who has watched a video, and saying one of them out loud unprompted buys more trust than the entire demo.
| What | Why it matters in a client build |
|---|---|
| Two workflows raise the same task on one enquiry | An advisor sees it in week one. Decide which owns it before go-live — The first five minutes: call-back and routing |
| An agent signs off with a placeholder it was never given | The prompt asked for something the data did not carry. Read what it writes before you switch it on — Two kinds of follow-up: the cadence and the written-for-one message |
| A deal moves stage even when the calendar write failed | The record then claims an invite that was refused — Booking the meeting, and chasing the no-show |
| A sensible-looking retrieval threshold deletes correct answers | The assistant starts saying it does not know things it knows — What a knowledge base is, and how search finds things |
Two workflows raise the same task on one enquiry
- Why it matters in a client build
- An advisor sees it in week one. Decide which owns it before go-live — The first five minutes: call-back and routing
An agent signs off with a placeholder it was never given
- Why it matters in a client build
- The prompt asked for something the data did not carry. Read what it writes before you switch it on — Two kinds of follow-up: the cadence and the written-for-one message
A deal moves stage even when the calendar write failed
- Why it matters in a client build
- The record then claims an invite that was refused — Booking the meeting, and chasing the no-show
A sensible-looking retrieval threshold deletes correct answers
- Why it matters in a client build
- The assistant starts saying it does not know things it knows — What a knowledge base is, and how search finds things
None of those is a reason not to sell it. Every one of them is a five-minute fix by somebody who knows where to look, and that person is what the client is buying. A demo that admits a flaw and shows the screen where it is fixed is more convincing than one that admits nothing.
The costs to be straight about
- Every agent run costs something. A scorer on every lead, an assistant on every message, a weekly writer per person. It is small and it is not nothing, and a client should hear the shape of it before their first invoice rather than after.
- Deterministic work is nearly free. Routing, tasks, notes, stage moves — these are the bulk of what the org does and none of it calls a model. Reach for a rule before an agent, and say why.
- Switching something on is a commitment. The morning sweep raises up to fifty tasks tomorrow; a cadence starts a twenty-one-day sequence. Turn things on with a client in the room, not the night before — Draft and active: which workflows are actually running.
The sentence to end a first meeting with
“Everything I just showed you is running, on your data, before you give us anything. The four things you'd have to hand over are things you already own, and the longest of them is a queue at WhatsApp's end, not ours.”
It is true, it is checkable, and it moves the conversation from whether to when. You can say it because you have counted: 7 workflows running, 7 waiting on four credentials, 5 waiting on a decision, over 96 enquiries and 110 people.
Try it
- Write the three sentences you would use to describe what this does to a client who has never automated anything. No product names.
- Pick the flaw from the table above that would bite your next client soonest, and write how you would raise it in a first meeting.
- Go back to any workflow in The deals board and the seven stages's org and check you can still say what it does, what starts it, and what it would cost to run. That is the whole course.
That is the path. You can read a business out of its own org, read any of its 19 automations, say what each would cost and what it needs, show it working in twelve minutes, and draw the line where the honest promise ends.
Related lessons
Base rates — what a piece of evidence is actually worth
A face-recognition system that is 99.9% accurate and almost entirely wrong, and a number that sent an innocent woman to prison. Both are the same arithmetic, and it is the arithmetic that decides what any piece of evidence is worth.
ReadConfirmation and survivorship — what you never looked for
Two questions about evidence you did not go looking for. One is a rule you have to discover, and one is a pattern in five famous people — and in both, the thing that would have told you the truth is the thing nobody checks.
ReadLoss aversion, sunk cost and regression — what it costs you
Four questions you answer about yourself rather than about a scenario, and your own answers are the finding. Then the pattern that makes praise look useless and criticism look like it works, whatever you actually do.
Read