Customer Health Score Agent

Demonstrating 7 Key Agentic Workflow Concepts in a Real-World Application

1. Planning & Decomposition
2. Multi-Agent Collaboration
3. Tool Selection Decisions
4. Research & Synthesis
5. Reflection & Self-Critique
6. Decision Trees / Routing
7. Iterative Refinement
0
TRIGGER

Weekly Scan or Real-time Event Detection

1
PLANNING & DECOMPOSITION

Orchestrator Agent

Breaks "Assess customer health" into sub-tasks:

  • Query product metrics
  • Pull support history
  • Check billing status
  • Identify stakeholder changes
  • Analyze communications
  • Synthesize into score
  • Take appropriate action
2
MULTI-AGENT COLLABORATION

5 specialized agents work in parallel:

Product Agent
Amplitude API
Daily Active Users, Adoption, Activity
Support Agent
Zendesk API
Tickets, Sentiment
Billing Agent
Stripe API
Payments, Annual Revenue, Renewal
Contact Agent
Salesforce API
Account Manager, Stakeholders
Comms Agent
Slack API
Response Rate, Engagement
3
TOOL SELECTION DECISIONS

Conditional tool selection based on detected signals:

Usage drop detected?
> Query product analytics
Tickets spiking?
> Query support system
Payment failed?
> Check billing history
Account Manager left?
> Query CRM for stakeholders
Low engagement?
> Check Slack/Email activity
4
RESEARCH & SYNTHESIS

Synthesis Agent

Combines findings from all sources into unified risk assessment:

Product: Daily Active Users -15%, Adoption 67%
Issues: 3 tickets, Sentiment -0.4
Billing: $125K Annual Revenue, 45 days to renewal
Contact: Account Manager departed
Support: 12 days no contact
Initial Risk Score: 34/100 (HIGH RISK)
5
REFLECTION & SELF-CRITIQUE

"Am I sure? What could I be wrong about?"

✓
Data Complete?

All sources returned data

✓
No Contradictions?

Signals are consistent

✓
Context Valid?

Not migration/holiday/etc.

✓
High Confidence?

Score is reliable

↺ If validation fails, loop back to Research & Synthesis
6
DECISION TREES / ROUTING
CRITICAL RISK
Score < 25 > URGENT: Executive escalation required
> Slack #cs-critical-accounts
> Alert Director of BD + Manager + President + Finance
HIGH RISK
Score 25-49 > Immediate escalation to Director of Business Development
> Slack #cs-critical-accounts
> Alert Director of BD + Manager + President
MEDIUM RISK
Score 50-79 > Queue for weekly review
> Notify Director of Business Development
> Monitor for changes
LOW RISK
Score 80+ > Slack #cs-expansion-opps
> Consider for advocacy program
> Re-check in 30 days

Secondary Routing by Issue Type:

Billing > Finance Team Product Friction > Product Team Support Issues > Account Management Lead
7
ITERATIVE REFINEMENT

Feedback Loop - Learning Over Time

  • Track which flagged accounts actually churned
  • Compare predictions vs. actual outcomes
  • Refine risk weightings based on data
  • Incorporate Account Managers feedback on assessments
  • Adjust thresholds based on false positives/negatives

Example Reports

Click on any example to view the full Customer Health Assessment Report

34

TechCorp Industries

Enterprise Customer

HIGH RISK

Account Manager departed, usage declining, multiple support tickets

62

Acme Corporation

Mid-Market Customer

MEDIUM RISK

Renewal approaching, moderate engagement decline

92

GlobalTech Solutions

Enterprise Customer

LOW RISK

Strong engagement, expansion opportunity identified

18

StartupXYZ Inc

Small/Medium Business

CRITICAL RISK

Payment failed, zero activity, super-connector left