Predictive Lead Scoring with Dynamics 365

Dynamics 365 Predictive Lead Scoring — Field Guide

What the model actually uses

• Firmographics & CRM facts: industry, revenue band, parent account, territory, source, campaign.

• Behaviors: email opens/replies, link clicks, web sessions (time/depth/pricing views), meeting accepts, form fills, webinar attendance, unsubscribe/bounce.

• Rep interactions: call outcomes, notes, stage changes, tasks completed/overdue.

• Outcomes: closed-won/closed-lost + reasons (the backbone for learning).

• Timing & sequence: latency from touch → response, order of actions, desktop vs mobile patterns.

Translation: it blends “who they are” + “what they did” + “how fast/in what order” + “how similar leads ended.”

---

Enable & set up (15–45 min once prerequisites are met)

1. Prereqs

   • D365 Sales (with Sales Insights/Predictive scoring enabled in your license).

   • Sufficient historical leads with CLOSED status (won & lost) and consistent close reasons.

2. Turn on scoring

   • Settings → Sales Insights settings → Predictive models → Lead scoring

   • Select target entity (Lead), confirm outcome fields, pick training data window (e.g., last 6–18 months).

3. Choose fields & filters

   • Include meaningful columns (don’t drown it; avoid PII not used for selling).

   • Exclude internal/test/dummy leads via filters.

4. Train & publish

   • Review Feature importance and Data health.

   • Publish; scores populate on leads (and refresh per schedule).

---

Data hygiene checklist (do this first)

• Outcome truth: Make Closed as Won/Lost mandatory; standardize Close Reason picklist.

• De-dupe & normalize: emails, domains, industries, geos.

• Interaction fidelity: ensure marketing events (opens/clicks/web) and sales activities sync to the same lead.

• Time stamps: capture created/last touch/first response; no backfills that distort order/timing.

• Bot noise: exclude spam/bot hits and internal testing traffic.

---

Turn scores into action (what reps actually see)

• Buckets by percentile, not fixed numbers:

  • A = top 10% (work now), B = next 20% (nurture fast), C = middle 40%, D = bottom 30% (automate only).

• Next-best actions: pair bucket + behavior:

  • A + pricing page in last 24h → immediate call task (SLA < 1h).

  • B + webinar attended → book demo sequence.

  • C + multi-open no reply → send value email, wait 3 days, then call.

  • D + unsubscribe → suppress; route to re-permission stream.

Power Automate sketch

• Trigger: Lead score updated OR enters bucket A.

• Conditions: score percentile + recent activity (pricing page, reply).

• Actions: create phone task → assign to owner; add to Sales sequence; set Follow-up by; post to Teams channel.

---

Views & dashboards for sellers and managers

• My A-Leads (last 7 days): columns = Score, Last Activity, Pricing Views, Owner, SLA timer.

• A/B conversion funnel: Created → First touch → Meeting booked → Qualified → Won.

• Lift chart: win rate by decile (prove the model separates signal from noise).

• Aging report: A-leads with no human touch in 24h.

---

Feedback loop (how it keeps getting better)

• Close every qualified opportunity with Won/Lost + Reason.

• Retrain quarterly (or after big GTM or pricing changes).

• Exclude outliers (mega 3-year deals) from training to avoid skew.

• Review Feature importance; if junk fields dominate, fix data or unselect them.

• Track data drift: are behaviors/industries shifting vs last quarter?

---

What to measure (prove ROI)

• Lift vs random: win rate in top decile ÷ overall win rate. Target ≥3×.

• Precision@k: % of wins inside the top k% (e.g., top 20%).

• Time-to-first-touch for A-leads (goal: <60 minutes).

• Meetings per 100 touches by bucket.

• Cycle length & win rate by bucket/segment.

• Rep adherence: % A-leads touched within SLA.

---

14-day rollout plan (low-drama)

Day 1–2: Data audit & close-reason cleanup.
Day 3: Enable predictive scoring; pick fields/filters; initial train.
Day 4: Create A/B/C/D buckets (percentiles); publish seller views.
Day 5: Automations: Power Automate + Sales sequences + SLA timers.
Day 6: Manager dashboard (lift, precision@k, SLA adherence).
Day 7–8: Pilot with 5 reps; collect edge cases; tweak filters.
Day 9: Retrain if needed; lock bucket thresholds; document playbook.
Day 10–11: Team enablement (30-min call + cheat sheet).
Day 12: Go-live; monitor scoring job health & automations.
Day 13–14: Review early metrics; adjust actions/SLA; backlog enhancements.

---

Common pitfalls (and fixes)

• “All my leads look hot.” Buckets by percentile; cap marketing score influence; filter out internal/test.

• High score, no conversions. Close reasons missing; behaviors mis-tracked; revisit features and retrain.

• Model drift after new product launch. Shorten training window; add new signals (e.g., technical Q&A attendance).

• Rep distrust. Show lift chart + 10 sample “why” cards; tie SLAs only to A-bucket at first.

• Gaming the system. Weight outcomes > activities; avoid giving points for vanity opens.

---

Sales talk tracks (use the score without sounding robotic)

• A-lead (recent pricing activity):
  “Noticed you compared packages after last week’s webinar—happy to align the plan to your timelines. Is X your priority this quarter?”

• B-lead (webinar attendee, no reply):
  “What stood out—or didn’t—from the session? Two clients in your space chose Y to cut Z by 30%. Worth a 15-minute fit check?”

• Re-engaging C-lead:
  “Sharing a 2-minute screen of how peer company solved X. If it’s not a 2025 priority, I’ll park it for now.”

---

Governance tips

• Document: training window, included fields, filters, retrain cadence.

• Limit who can change model settings; change control with notes.

• Keep a shadow view of leads excluded from scoring (for audit).

• Align privacy: only process signals you’ve disclosed in consent.