Scoring a lead against a written rubric
Every enquiry gets a score out of 100 and a band. The scheme is not in the model — it is a document somebody wrote, and changing it changes the scoring.
This is the workflow to show a sceptical client, and it is the one to understand best, because it answers the objection that ends most AI conversations: how do we know it is right?
Four boxes, and two of them are attached rather than in the flow
2.3 · Mark every lead hot, warm or cold — and say why
A new enquiry starts it. The agent reads the rubric and writes the result — and both of those are tools hanging off the agent, not steps after it.
Scroll for all 4 steps →
A tool is not a step. The two boxes underneath the agent run if and when the agent decides to call them, not in a fixed order. That is the difference between an agent and a flowchart, and it is drawn on the canvas: tools hang below, steps sit in the line.
Where the scoring actually lives
The agent is not asked to judge the enquiry. It is instructed to go and read how this business judges enquiries, and to score against that and nothing else:
## Instructions
- First use the rubric tool to read the scoring rubric from the sales playbook.
Score against that rubric only — do not invent your own weighting.
- Then write the result back with the write_score tool, called EXACTLY like this:
entity_type "deal", operation "update", entity_id the Enquiry id above, and
data containing only a next_step of the form
"SCORE 72/100 — WARM — comfortable obligation ratio, documents not ready".
- Keep the reasoning to one line naming the two factors that decided the band.“Score against that rubric only — do not invent your own weighting.” The rubric is a document in the sales playbook knowledge base. A person wrote it, a person can rewrite it, and nobody needs a developer or a model to do it.
What one real run did
Here is a run from this org, node by node — the shape you will read in the Executions tab yourself:
| Step | What happened |
|---|---|
| the trigger | A new enquiry: “Home Loan — Anil Kohli”, ₹33.2 lakh, source Direct, sitting at New Enquiry. |
the agent calls rubric | Its own search query — “score home loan enquiry rubric”. The rubric document comes back top, at 0.9337 similarity. |
| the agent answers | SCORE 45/100 — WARM — unknown credit signal, 65 days old enquiry, moderate urgency |
the agent calls write_score | entity type deal, the enquiry's own id, and a next_step of exactly that sentence. |
the trigger
- What happened
- A new enquiry: “Home Loan — Anil Kohli”, ₹33.2 lakh, source Direct, sitting at New Enquiry.
the agent calls rubric
- What happened
- Its own search query — “score home loan enquiry rubric”. The rubric document comes back top, at
0.9337similarity.
the agent answers
- What happened
SCORE 45/100 — WARM — unknown credit signal, 65 days old enquiry, moderate urgency
the agent calls write_score
- What happened
- entity type deal, the enquiry's own id, and a
next_stepof exactly that sentence.
Read the middle row again: the agent chose its own search words. Nobody wrote “score home loan enquiry rubric” anywhere. That is the part that is genuinely an agent — and the part after it, the write, is completely constrained.
Why the band comes with a reason
A score on its own is not usable and not defensible. “45/100” tells an advisor nothing and gives a client nothing to disagree with. “45/100 — WARM — unknown credit signal, 65 days old” names the two factors that decided it, so the advisor knows what to ask on the call and the manager knows whether the rubric is wrong.
The prompt asks for exactly that shape, in one line. It is a formatting instruction, and it is doing a governance job.
The answer to “how do we know it is right?”
| The worry | What you show them |
|---|---|
| “The AI decides who we chase.” | No — the rubric decides. Open it. It is a document with five factors and a band table in it. |
| “We cannot change how it thinks.” | Edit the document and run one enquiry again. Nothing is deployed, nothing is retrained. |
| “We cannot tell why it said that.” | Every run keeps what the model was sent and what it answered, side by side. |
| “It might score on something it should not.” | The Constraints section forbids it, in English, in a box they can edit. |
“The AI decides who we chase.”
- What you show them
- No — the rubric decides. Open it. It is a document with five factors and a band table in it.
“We cannot change how it thinks.”
- What you show them
- Edit the document and run one enquiry again. Nothing is deployed, nothing is retrained.
“We cannot tell why it said that.”
- What you show them
- Every run keeps what the model was sent and what it answered, side by side.
“It might score on something it should not.”
- What you show them
- The Constraints section forbids it, in English, in a box they can edit.
Every one of those answers is a screen, not a claim. That is the whole reason this workflow is worth understanding better than any other in the org: it is the one where “show me” is a better response than “trust me”.
Change one thing — and this is the demo
- Find the rubric document in the sales playbook knowledge base. Read the five factors.
- Change one weighting — make document readiness worth more, or urgency worth less. Save.
- Create an enquiry on the board and let the workflow run.
- Read the
next_stepit wrote, and the run behind it. The score moved because you moved it, and the audit trail says so. - Put the rubric back if you want the rest of the course to describe what you are seeing.
This is the moment to rehearse for a client meeting, not to improvise. Changing a document and watching the scoring change, live, in about ninety seconds, is the most persuasive thing in this org — and it only lands if you already know where the document is.
Next: The first five minutes: call-back and routing — the same idea, with the conversation happening out loud.
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