I85PREP
⌘ K
Microsoft MB-220 study materials
Microsoft MB-220 study resources

Lead Scoring That Never Fires

Microsoft MB-220 · Put your knowledge to work in context.

All case studies

Lead Scoring That Never Fires

Tailspin configured a lead scoring model with two grades: A (score ≥ 50) and B (score < 50). Scoring rules: +20 for JobTitle containing 'Director' or above, +15 per whitepaper download (max 45), +10 for a website visit in the last 30 days, and −25 for a personal email domain. The model was published 3 weeks ago. The sales team reports that no lead has ever reached grade A, even though 40 leads have Director titles and 3 whitepaper downloads each. Investigation shows: the whitepaper download interaction is logged on the Contact, not the Lead; the Lead's JobTitle field is populated by a real-time marketing form but is often blank on leads created by the sales team manually; and the model's grade thresholds were set before the negative rule was added.

1. Explain why no lead reaches grade A despite qualifying activity, and rank the causes by impact.

Read the worked answer

Work the arithmetic for the best-case lead: Director (+20), three whitepaper downloads (would be +45), one recent visit (+10) = 75, comfortably above 50. So the threshold is not the problem for that lead - the data path is.

Ranked by impact:

  1. Interactions are on the Contact, not the Lead. Whitepaper downloads never accrue to the Lead record, so the +15-per-download rule contributes 0. Best case drops to 20 + 10 = 30 - below 50. This alone explains the failure for the 40 Director leads.
  2. Blank JobTitle on manually created leads. The +20 rule cannot fire when the field is empty. Every sales-created lead starts at a maximum of 25 (visit only) and cannot reach A. This is a large silent cohort.
  3. Thresholds set before the negative rule. If grade A was calibrated at 50 when the maximum positive score was 75, adding −25 for personal domains moves the effective bar to 75 for anyone with a Gmail address. That is a calibration error, not a data error, and it compounds causes 1 and 2.

The sales team sees 'no A leads' as one symptom; it is three independent defects with different owners.

2. Propose a remediation plan, including how to validate the scoring model before republishing.

Read the worked answer

Fix in dependency order, then validate:

  1. Re-point interaction rules to the Lead. Either score on Contact interactions and roll the score to the associated Lead, or capture whitepaper downloads against the Lead when a lead exists. Decide which entity is the scoring subject and be consistent - mixed scoring is how this broke.
  2. Backfill and enforce JobTitle. Backfill from the Contact where a lead has a linked contact, and make the field required on the sales-created lead form so the rule has data to read.
  3. Recalibrate thresholds. Compute the realistic maximum positive score (20 + 45 + 10 = 75) and the realistic minimum with the negative rule (−25). Set grade A at a score that a genuinely qualified lead can reach without a perfect record - for example 55 - and document the reasoning so the next rule change triggers a recalibration review.

Validation before republish:

  • Build a test cohort of 10 leads spanning the archetypes: Director with 3 downloads, Director with 0 downloads, non-Director with 3 downloads, personal-domain Director, blank-title lead.
  • Score each and compare to expected values computed by hand.
  • Confirm grade assignment matches intent for each archetype, then republish and re-check the same 10 leads in the live model.

Never republish a scoring change without a hand-computed expectation table; the model will happily produce numbers that are wrong.