When a company decides to review calls automatically, the discussion usually starts with technology: which model, what recognition quality, how it connects to the CRM. Those are all solvable. The one that turns out not to be is different: what criteria the conversation gets judged by.
A sales book almost always exists. The problem is that it was written for a human — and a human fills in context the text does not contain.
Description versus criterion
Most sales books consist of descriptions: what to do and in what spirit. A description helps a person understand intent, but it does not let anyone answer yes or no about a specific conversation.
| Description | Criterion |
|---|---|
| Build rapport at the start of the call | The rep gave their name, company and reason for calling before the first question about the task |
| Uncover the need | A question about process volume was asked, and a question about who makes the decision |
| Handle the price objection | After the price objection the rep returned to the task, rather than naming a discount |
| Agree the next step | A date or a specific event was named that both sides will return to |
A simple test for a wording: hand it to two managers along with the same conversation. If their verdicts differ, the criterion is not ready — and the issue is the wording, not the system.
Qualification definitions
The MQL and SQL boundaries live in oral tradition at most companies. While three people who eat lunch together do the qualifying, that works. Automation forces the implicit agreement to be written down — and that is usually when it turns out there was no agreement.
Which facts must be established for a lead to count as qualified
What to do when some of them came from an e-mail or a form rather than the conversation
Which condition is blocking and which is merely desirable — otherwise everything lands in one bucket
Who settles the case where the criterion is formally met but substantively is not
That last point matters more than it looks. Contested cases will always exist, and the goal is not to eliminate them but to give them a route. A system forced to produce a definitive answer in an ambiguous situation starts inventing one — and trust in it ends at the third such case.
DemoCalibration: why it is unavoidable
Even well-worded criteria diverge from the manager's judgement on the first pass. That is normal, and there is only one remedy: take a set of historical conversations, score them by hand, and compare with what the system produced.
- 1
Assemble a set with varied outcomes
Not just closed deals: include losses and stalled ones, or the criteria tune to success and stop discriminating.
- 2
Score by hand before seeing the system's output
Order matters. Shown someone else's verdict first, a person almost always agrees with it.
- 3
Work through disagreements, not matches
Every disagreement is either a system error or an imprecise wording. The second is more common.
- 4
Fix the wording, not the result
The temptation to nudge the score of one conversation is strong. Do it and the criterion stops working for all the others.
- 5
Record the version
A score only means something together with the methodology revision it was given under. Otherwise periods cannot be compared.
What is better left unformalised
Not everything in a conversation yields to a criterion, and trying to describe all of it backfires: a forty-item checklist gets followed by nobody, and the system starts penalising people for speaking naturally.
Tone and empathy: judged by people; formalising them produces a script
The order of conversation blocks, where it does not change the outcome
Talk ratio and utterance length — measurable, but rarely meaningful
Anything you cannot point at an excerpt to justify
A good criterion is one where the system and the manager reach the same conclusion — and a rep who disagrees can point at an excerpt and contest it.
Methodology notes, by e-mail
The same material we publish here: how to model the economics of an initiative, where rollouts break, and what to verify before work starts. Once a month at most, no market news and no sales e-mail.
