Skip to content

Predictive Lead Scoring: The 40/40 Rule for Small Teams

Predictive lead scoring needs 40 qualified and 40 disqualified leads. Count yours in ten minutes and use a simple points score until you get there.

Alex Pechenizkiy 11 min read
Predictive Lead Scoring: The 40/40 Rule for Small Teams

You heard that predictive lead scoring came with Dynamics 365 Sales, so you opened the setup screen expecting it to rank your leads. The Get started button would not respond. Microsoft documents this: “If you don’t have the minimum closed leads in the chosen time period, the Get started button is disabled” (Configure predictive lead scoring).

The myth is that AI lead scoring is a switch you flip. It is not. It learns from leads your team has already finished with, and a young or small CRM can easily have too few of them.

On Sales Enterprise, this is a data problem, not a license problem. Microsoft says that with Sales Enterprise “you can install and configure three premium features (sales accelerator, conversation intelligence, and predictive scoring) with a limited monthly capacity” (Enable and configure Sales Insights features). So you already have the feature. What you do not have yet is enough history to feed it.

Before you call anyone about licenses, count your closed leads.

How predictive lead scoring decides whether you qualify

Microsoft’s feature “uses a predictive machine learning model to calculate a score for open leads based on historical data” (Configure predictive lead scoring). Each open lead gets “a score between 0 and 100”, plus a grade from A to D, where A is most likely to convert (Prioritize leads through scores).

The rule is 40 qualified AND 40 disqualified leads. That is at least 80 closed leads, not 40. In plain terms, these are leads your team has finished with, one way or the other.

Open leads do not count toward the 80. They are what gets scored. Closed leads are what teaches the model.

The time window is the confusing part, because three Microsoft pages phrase it three different ways:

You do not need to reconcile these. The safest reading is: closed, both kinds, recent.

A short license aside. Sales Enterprise includes predictive lead scoring for 1,500 scored records a month per environment, while Sales Premium has “No limit” (Digital selling capabilities in Sales Enterprise, Digital selling FAQs). Microsoft’s admin page also says “Licensing options for Sales Insights features have changed” and tells you to contact “your Microsoft account team or partner for guidance,” so ask yours (Enable and configure Sales Insights features).

The number to watch is your closed leads of each outcome, counted separately.

Count where you stand in ten minutes

Microsoft’s FAQ includes a count check you can do yourself. Here it is as plain steps:

  1. 1

    Open All leads and filter by date

    In Dynamics 365 Sales, open the All leads view and filter Created On to the window you want to test. Two years is the default the model uses.

  2. 2

    Set Status to Qualified

    Add a filter so you only see leads your team has qualified.

  3. 3

    Exclude abandoned process leads

    If you use a business process flow (the stage bar across the top of the lead), add a filter for its Status Reason not equal to Aborted.

  4. 4

    Write down the number

    This is your qualified count.

  5. 5

    Switch to Disqualified and count again

    Change the Status filter to Disqualified and write down the second number.

Step 3 matters because Microsoft drops leads that abandoned the business process flow. They “will not be considered for training, scoring, and for establishing the minimum requirement for model creation” (Configure predictive lead scoring).

If either number is under 40, Microsoft’s advice is to “Increase the time frame to one to two years”. Once both numbers are above 40, Microsoft asks you to keep the two outcomes balanced “for better model performance” (Lead management FAQs).

One timing catch. Microsoft says “The system takes about four hours to sync the data with the data lake. If you’ve closed leads recently, the model won’t consider them right away” (Configure predictive lead scoring). When you publish, the same page says “To retrain the model every 15 days, select Retrain automatically,” and Microsoft’s FAQ says to “Enable auto-retraining so that the model can update itself with recent outcomes without manual intervention” (Lead management FAQs).

If you only want to see what scored leads look like, Microsoft says “To explore predictive scoring without using your company data, publish sales accelerator with sample data” (Lead and opportunity scoring). Sample data is “available only in your sandbox or trial environments” (Sales accelerator). It never scores your real leads.

Write the two numbers down. They decide everything that follows.

Agree on what a good lead is before you score anything

If sales marks a lead qualified for one reason and marketing for another, your 80 closed leads are 80 coin flips. The model learns your confusion. Microsoft lists the first factor in model accuracy as “The quality and amount of the data available for training the model” (View the accuracy and performance of predictive scoring models).

The fix is small. Write one sentence that defines “qualified”. Then agree on a short, fixed list of reasons to disqualify, such as wrong size, no budget or not reachable.

That one page is your shared lead definition. This is sales and marketing alignment in practice: sales and marketing agreeing on what a good lead is, on one page, not in a workshop.

Put the disqualify reasons into the system, not just on paper. If your sellers pick from a fixed list when they close a lead, you can later see why leads fail. If they type free text, you get fifty spellings of “no budget”.

Watch out for the Rating field. Dynamics 365 has a built-in Rating with three choices: Hot, Warm and Cold. Its default form value is Warm (Lead table reference). Every lead created on the form starts as Warm, so if nobody changes it, the field tells you nothing.

Alex Pechenizkiy has seen unclear lead definitions and messy lead records again and again, across hundreds of engagements over 20+ years in IT. That is a process lesson, and it applies here before any score exists.

One page, signed off by whoever sells and whoever markets, before anyone builds a score.

Below the line: a points score that works now

A points score you can explain beats an AI score you cannot have yet. Build it from two kinds of points. Fit points cover who the lead is. Engagement points cover what the lead did, and negative points flag bad signs.

Option A: Customer Insights - Journeys scoring models. Journeys was formerly Dynamics 365 Marketing. If you already run it, use its built-in lead scoring. You turn on the Lead Management feature switch, then build a model that is points-based and scores on “demographic attributes and interactions” (Create lead scoring models). It does not train on your history.

Points “can be negative or positive”, you can add “up to 40 groups per scoring model”, and “Scoring models refresh every two hours.” Grades such as Hot or Cold are optional. You can publish a model without any.

When a lead crosses your line, Journeys can mark it Marketing Qualified or Sales-Ready, then hand it to “Dynamics 365 Sales assignment” or trigger a sales sequence (Qualify the best leads). Two cautions from Microsoft. If a qualified lead’s score later drops, “the lead remains as Marketing Qualified.” And leave parent contact scoring off (Known issues).

Be clear on licensing. Journeys needs a separate Customer Insights license. The reduced “attach” license applies if you have Dynamics 365 Sales or another qualifying app “with 10 or more users” (Purchase Dynamics 365 Customer Insights).

Option B: a formula column on the lead. In our experience, this works well when there is no Journeys. A formula column is a field that calculates its own value. Microsoft says you build it “the same way you would build a formula in Microsoft Excel” (Work with formula columns).

The column adds up points from fields already on the lead, such as industry, company size and source. It runs inside Sales with no extra app. The limit is real: it scores fields only, and no Microsoft source says it can count email opens.

Here is what a first version looks like. The point values are illustrative; calibrate against your own data.

  • Target industry: +10
  • Company size inside the range you sell to: +10
  • Came from a referral or your own website form: +5
  • Job title is someone who signs: +5
  • No phone number and no company name: minus 10

Then agree with your sellers what the total means. For example, above a set score gets a call the same day, the middle gets an email, the bottom waits. The cut-off is yours to set, and you move it when sellers tell you the top leads are not real.

Whichever option you pick, review the points monthly with whoever sells. Change one rule at a time, so you can tell what moved the results.

Pick A if you already run Journeys and B if you do not. Keep your first version to five or six rules.

Three ways to score at your size

What it needs to start
Predictive lead scoring
At least 40 qualified and 40 disqualified closed leads in your training window
Journeys scoring model
Customer Insights license; points-based, does not train on history
Formula column score (our suggestion)
In our experience: a few agreed fields filled in on each lead
AI or rules
Predictive lead scoring
AI: a machine learning model trained on your history
Journeys scoring model
Rules: points you set, positive or negative
Formula column score (our suggestion)
Rules: points you set (in our experience)
What it scores on
Predictive lead scoring
Signals from leads and related records such as contact and account; a score between 0 and 100, grades A to D
Journeys scoring model
Demographic attributes and standard Dynamics 365 interactions
Formula column score (our suggestion)
In our experience: fields on the lead only, no email opens
License note
Predictive lead scoring
Sales Premium, Sales Insights or Sales Enterprise; Enterprise covers 1,500 scored records a month per environment
Journeys scoring model
Separate Customer Insights license; attach license needs 10 or more users of a qualifying app
Formula column score (our suggestion)
In our experience: no extra app beyond Sales
Lead scoring options for a small Dynamics 365 Sales team

The left column is the goal. The other two get you there.

Build the history on purpose

Every lead should end as Qualified or Disqualified, with a reason from your agreed list, on a schedule. A lead left open for a year teaches the model nothing. Pick a regular day each month to close out stale leads.

In our experience, the simplest routine is a monthly sweep. Sort open leads by Created On, oldest first. Each one either gets qualified, gets disqualified with a reason, or stays open with a named next step and a date. A scheduled Power Automate flow can email you the list of open leads each month, the same way it can alert you to failed workflows.

Clear out test and junk records too. Delete test leads rather than disqualifying them, because a disqualified test lead counts as a real outcome. Microsoft suggests that if “there are too many test or incorrect records during that period, choose a shorter time period, such as six months or one year, when the quality of your data is better” (Scoring model accuracy).

Watch for duplicates as well. When the same person arrives twice and one copy gets qualified while the other gets disqualified, your history contradicts itself. Merge first, then close.

Keep closing leads every month and keep both outcomes roughly balanced. The model trains only on leads created and closed inside the window you choose, up to two years, so old closes age out (Configure predictive lead scoring). Separately, for open leads, if a record “goes outside the 2-year window in the next scoring cycle, the score won’t be updated” (Lead management FAQs).

A note on the Sales Qualification Agent, because it is not a shortcut around this. In research-only mode, “The agent works on assigned leads based on your selection criteria, such as lead source, rating, or geography.” It gives insights about each lead, such as their background, recent opportunities and news about their company.

“The agent evaluates the lead’s fit with your target customer profile and generates an initial outreach email for leads that meet your criteria,” and a seller decides whether to send it (Sales Qualification Agent overview). It needs a Copilot Studio license (Set up the Sales Qualification Agent) and uses your tenant’s prepaid or pay-as-you-go capacity. Microsoft says “The agent doesn’t replace your judgment or decision-making process,” and no Microsoft source says it replaces predictive scoring.

Below 40/40, a rules-based score on a shared lead definition is the right tool, not a consolation prize. Every lead you close cleanly today is training data for predictive lead scoring tomorrow.

Run your count again every quarter. When both numbers reach 40, try building the model and see whether it publishes.

Get one real lead checked before you buy or build

Citizen Development Academy runs a 30-minute Lead-to-Revenue Review. We walk one real lead from form to first seller touch and check five points: shared lead definition, capture and dedupe, scoring readiness, routing and SLA, handoff feedback and reporting. Nonprofits can get help free.

Request a 30-minute Lead-to-Revenue Review with Citizen Development Academy

Related articles