Set up lead conversion scoring

You want to know which of the people in your CRM are worth working first. This is the guide for that goal, from the card you pick to the number you read.

What it represents

Lead conversion is the first of the four cards on Create a Score. It answers one question: of the people who are not customers yet, which ones go on to produce a meeting or an opportunity?

Pick it when the record you want ranked is a person · a lead, a contact, an inbound signup, a list you just imported. Pick Won business instead when the record you want ranked is a deal that already exists, because a deal has a stage history a person does not.

The two are close enough to confuse, and the difference is the object, not the ambition. Lead conversion ranks people who may become pipeline. Won business ranks pipeline that may become revenue.

Everything behind More starting points stays available and stays selectable. Those entries are variants of the same four questions on different primary objects, and none of them carries a cohort, so choosing one means you take on defining the population yourself.

How it’s calculated

The card ships as a complete target definition, not a bag of pre-filled fields. It carries the object it measures, what counts as success, the date that stamps success, and · the part that is new · the population it trains on.

The population line renders on the card itself, under the description: Trains on: people who are not already customers. On Salesforce that is the converted flag on the Lead object; on HubSpot it is the lifecycle stage. Stating it is the point. A scoring model that quietly included your existing customers would be learning what a customer looks like, not what a converting lead looks like, and nothing on the screen would have told you.

Success is a record on the meeting or event object linked back to the person, dated by when that record was created. That is a deliberately early definition of success: it measures whether the person engaged far enough to get on a calendar, which is the thing a top-of-pipeline team can act on this week.

Set it up

1. Pick the Lead conversion card

   It is the first of the four cards at the top of Create a Score. The four lead because they are the four questions the product has a complete definition for; the rest sit behind More starting points.

   If the card is greyed out with a line where the population should be, the connection you picked has no definition for this starting point yet. That is a real answer, not a loading state, and the card says which connection type it means.

2. Read the Trains on line before you touch anything else

   Trains on: people who are not already customers appears on the card and again above Success definition once the form opens. That sentence is the population, in the shipped filter grammar, and it reaches the query that builds your dataset.

   Read it against your own CRM. If your team marks customers somewhere other than the standard field, this line is what will be wrong, and the preflight in step 7 is where you will find out.

3. Confirm Connection and Primary entity

   Connection picks which CRM this score reads. Primary entity is already set to People by the card, and on Salesforce a note appears reading Lead + Contact merge auto-enabled, because a person who was a lead and is now a contact is one person and should be scored once.

   Change the primary entity only if you meant a different question. Doing so is what turns a preset into something else, and the product will say so.

4. Leave the success definition alone unless your data disagrees

   What defines success? holds the object that records the outcome, and Success criteria holds the condition on it. For this starting point that is the meeting or event object, with a condition that is satisfied by the record existing at all.

   Edit it when your team does not book meetings through the CRM, or when the first outcome you care about is an opportunity rather than a meeting. Editing is allowed and supported. What changes is that the score now carries a note saying it diverges from the starting point it names, which is how you find it again in six weeks.

5. Confirm the Success date field

   Success date field is the date that stamps the outcome. It is what every feature is computed from the far side of, so it is the single control that decides whether the model is a forecast or a description of the past.

   The default is the created date on the meeting or event record. Change it only if that object dates the thing you care about somewhere else.

6. Set the Non-success cutoff

   Two options. Auto-detect (recommended) uses the 90th percentile of your observed conversion times, which is the honest default when you keep no disqualified date. Custom date field lets you name one when you do.

   Take auto-detect unless your team genuinely maintains a disqualified or archived date. A person who has not converted needs a cutoff too, or their whole history counts and the converted records are the only ones held to a boundary.

7. Check what a high score means here

   What does a high score mean here? sits directly under the success criteria. For this goal the answer is More likely to convert, which is what the card sets, and it gives you hot · warm · cold bands where a high score reads as opportunity.

   The control is there because you can re-point the success criteria at something you would rather avoid, and the readout has to follow you when you do. The note under it is exact: "Only the readout changes · the model learns the same thing either way."

8. Press Create score and read the preflight

   The button reads Checking your data... while a preflight runs against your connection. On this card it asks three data questions: does the population resolve to real records, is a success date set, and · on renewal starting points only · does the org carry renewal typing. A card that separates record-keeping losses, like At-risk customers, gets a fourth check for that.

   A pass renders as a count: 4,120 records in this cohort. It is never a question about whether you meant to build this score.

If the preflight refuses

Two refusals can reach this starting point, and each names an exit rather than a door.

No records match this cohort means the population filter matched nothing. On this goal that almost always means your CRM marks customers on a field other than the standard one, so every person was excluded. The block names the fields it looked at, and offers Include every record instead, which clears the population filter and re-runs. Take it knowingly: you are choosing to train on everyone, customers included, and that is a real trade rather than a fix.

No success date field means nothing is set in step 5. There is no workaround for this one and there should not be: without a date, features cannot be computed from before the outcome, and everything the model learned would be written in the past tense. The exit is Continue in Advanced, which carries forward the part that already resolved so a refusal never costs you work you already did.

Only 22 records in this cohort is a warning, not a refusal, and it never blocks. Fewer than 30 records is too thin to read a rate from, and the result carries that count with it so nobody downstream reads a number the population cannot support.

What it means for you

A lead conversion score is a ranking, not a percentage. A person at 72 goes above a person at 41, and the bands · hot at 80 and up, warm from 50, cold below · are labels on that ranking rather than targets to hit. Work the top of the list down to whatever capacity you have this week.

Read the base rate the readout gives you as the sanity check on the population. On a healthy top-of-pipeline dataset it is small, because most people do not book a meeting, and a thin hot tail is the model doing its job rather than a fault to fix. A base rate that arrives suspiciously high is usually the population line telling you something: customers got in, or the success condition is satisfied by a record that exists for everyone.

When Advanced is the right door. Take Start from scratch (custom) when the question is genuinely not one of the four · a fit score with no time dimension, a target on an object the presets do not touch, a population that needs several conditions to describe. Advanced gives you the full criteria builder and asks you the same population question the presets answer, so custom means you define the cohort rather than there being no cohort. What Advanced does not give you is a definition somebody has checked, which is the whole value of starting from a card.

Editing a preset is not the same as taking the Advanced door. An edited preset keeps its lineage: the score records which starting point it came from and which of the six definition fields you changed, and the create flow says so in one line · “You changed what counts as success from the Lead conversion starting point.” · with the saved score carrying the equivalent note. The note is not a warning and it does not block. Keep my version retires it for good once you have confirmed the change was deliberate.

Check your understanding

You pick Lead conversion, the card reads Trains on: people who are not already customers, and the preflight comes back 9,840 records in this cohort. The first run reports a 6% base rate and a distribution with 4% hot.

Read that back against the controls. The 9,840 is the population line resolving against your converted flag, which is why customers are absent from the number. The 6% is what it costs to book a meeting with a stranger, and it is the reason 4% hot is a healthy shape rather than a thin one. Had the preflight instead read the whole person book, the base rate would have carried every customer you have ever closed as a positive, and the model would have spent its capacity learning to recognize them.