Skip to content

Change Management Plan – SaaSGuard Platform Rollout

Overview

SaaSGuard delivers P(churn in 90 days) scores, compliance risk signals, and AI-augmented executive summaries to Customer Success and Sales teams. This document covers the stakeholder map, training approach, phased rollout schedule, governance model, and success metrics for a production deployment targeting a $200M ARR portfolio.

1. Stakeholder Map

Stakeholder Role Impact Engagement Level
VP of Customer Success Executive sponsor; owns churn KPI and CSM headcount High – defines success metrics and signs off on rollout Champion
Customer Success Managers (CSMs) Primary end users; consume Customer 360 dashboards and AI summaries daily Critical – adoption determines ROI Daily users
Sales / Account Executives Consume GTM integration signals (renewal risk, upsell triggers) Medium – use Churn Heatmap for pipeline prioritisation Weekly users
IT / Security Reviews data residency, auth model, container hardening, CORS policy High – gate for production approval Approvers
Executive Team (CEO, CFO) Consume ROI dashboard and executive summaries in board-prep sessions Medium – use Uplift Simulator for investment decisions Monthly users
Data Engineering Owns dbt refresh pipeline, DVC model versioning, DuckDB infrastructure High – responsible for data SLAs Operators

2. Training Plan

CSM Onboarding Guide (30 min)

Target audience: All Customer Success Managers

  • Module 1 (10 min): Customer 360 dashboard walkthrough — how to read churn probability, risk tier, and SHAP feature drivers
  • Module 2 (10 min): AI executive summary generation — how to request a summary, interpret guardrail flags, and apply human judgement before sharing externally
  • Module 3 (10 min): Hands-on practice with 3 sample at-risk accounts; Q&A

Delivery: Recorded Loom video + live session facilitated by VP CS. Materials in docs/.

Executive Dashboard Walkthrough (15 min)

Target audience: VP CS, CFO, CEO

  • Superset Churn Heatmap — cohort-level risk view
  • Uplift Simulator — model ROI scenarios (e.g., "if we intervene on 20 high-risk accounts, what is the expected churn reduction?")
  • How to interpret confidence ranges and model limitations

Delivery: Slide deck + live session. See docs/demo.md for the standard demo script.

API Integration Guide (Engineering Partners)

Target audience: Sales Ops, CRM integration engineers

  • FastAPI endpoint reference: GET /customers/{id}, POST /predictions/churn, POST /summaries/customer
  • Authentication headers and CORS configuration
  • Webhook patterns for real-time churn alert pipelines

Delivery: docs/API.md + OpenAPI interactive docs at :8000/docs.

3. Phased Rollout Schedule

Week Milestone Scope Success Gate
Week 1 Pilot launch 2 senior CSMs on 10 at-risk accounts Both CSMs complete onboarding; ≥80% report dashboard "useful"
Week 2 Full CS team rollout All CSMs (~15 users) No P1 support tickets; AI summary guardrail pass rate ≥90%
Week 4 Executive dashboard access VP CS, CFO, CEO VP CS reviews Uplift Simulator in board prep
Week 6 Sales integration AEs receive GTM churn signals in CRM Renewal pipeline enriched for ≥80% of Q2 renewals
Week 8 API live for integrations Engineering partners connect CRM webhook First automated churn alert delivered to Slack
Week 12 First model retraining Data Eng runs dvc repro on new cohort data Calibration drift < 5pp; accuracy maintained

4. Governance & Data Freshness

Model Retraining Schedule

  • Frequency: Quarterly (aligned with new customer cohort data)
  • Trigger: Calibration drift > 5 percentage points OR accuracy drop > 3pp on holdout set
  • Process: dvc repro → review tests/model_accuracy/ metrics → dvc push → CI/CD deploys new image
  • Approver: VP CS signs off on model metrics before production promotion

Data SLA

Data Source Refresh Frequency Owner
raw.customers Nightly via dbt run Data Engineering
raw.usage_events Nightly Data Engineering
raw.support_tickets Nightly Data Engineering
raw.gtm_opportunities Nightly (Salesforce sync) Sales Ops
Churn predictions On-demand via API Prediction service

Human-in-the-Loop Policy

All AI-generated executive summaries include the watermark:

"[AI-GENERATED SUMMARY — HUMAN REVIEW REQUIRED BEFORE EXTERNAL DISTRIBUTION]"

CSMs are required to review and edit summaries before sharing with customers or in EBRs. AI outputs flagged by the guardrails system (hallucination detection, out-of-scope questions) must be discarded and regenerated with a more specific prompt.

Data Access Controls

  • CORS locked to approved origins (ALLOWED_ORIGINS env var — see docs/runbook.md)
  • DuckDB file mounted read-only to API container workers
  • Non-root container user (saasguard) — no write access to model artifacts at runtime
  • No PII in LLM prompts — customer IDs and aggregate metrics only

5. Success Metrics

Metric Baseline Target (90 days post-launch) Measurement
Churn rate Current cohort churn rate −5% relative reduction Customer data in DuckDB
CSM time saved per at-risk account 15 min manual research ≤3 min with Customer 360 CSM survey (n ≥ 10)
AI summary accuracy N/A (new capability) ≥80% rated "accurate" by CSMs Weekly Loom survey
Platform uptime N/A ≥99.5% (≤3.6h downtime/month) /health endpoint monitoring
Guardrail pass rate N/A ≥90% Prometheus guardrail_passed_total
CSM dashboard adoption 0% ≥80% weekly active CSMs Superset usage logs

ROI Framing

  • 1% churn reduction on $200M ARR = $2M saved annually
  • 10–15% churn reduction achievable through early CS intervention (Forrester)
  • CSM time savings: 3.3h/week × 15 CSMs × $85/hr fully-loaded = $220K/year productivity gain