System design, execution blueprint, and collaborative decision log for service intake automation, callback prevention, and zone-based dispatch routing.
Planning Draft
Local draft mode — decisions and actions save to your browser. Shared Notion collaboration pending.
Overview
Four dimensions of the build
Business Opportunity
72.5% of service callbacks are non-billable — issues that could be resolved remotely through guided troubleshooting. Five categories drive the majority: Service/Adjustment (33%), Defective Parts (32%), User Error (17%), Battery Failures (7%), Wind/Venting (11%). Every unnecessary dispatch consumes technician capacity reserved for billable work.
Target Outcomes
Non-billable rate: 72.5% → <40%
Intake: Variable/manual → 100% structured
Dispatch time: Manual review → <90 seconds
Utilization: Unoptimized → Max 7 jobs/tech/day
System Stack
Claude (Anthropic AI) as the triage brain — no OpenAI. Make.com as the no-code automation layer. Twilio for all telephony and SMS. Striven CRM as system of record. Birdeye for post-service review triggers.
TwilioMake.comClaude APIStrivenBirdeye
Implementation Phases
Phase 1 — Foundation (Days 1–7)
Phase 2 — Triage Logic & Scoring (Days 5–14)
Phase 3 — Monitor & Refine (Days 14–30)
Phase 4 — Full Automation (Days 30–60)
Key Metrics
Current state → target state
Non-billable callback rate
72.5%→<40%
Target: reduce by 45% within 90 days of full deployment via AI-guided troubleshooting
Intake structure
Variable / manual→100% structured
Every inbound contact produces a fully structured Striven job card before any human touches it
Charleston Location Code Matrix enforced — max 7 jobs/tech/day, Oct 1–Mar 31 peak season
System Architecture
Inbound communication flow — platform by platform
📞
Twilio
SMS & Voice
→
⚙️
Make.com
Automation
→
🤖
Claude API
Triage AI
→
📋
Striven CRM
Job Record
→
🔧
Tech Queue
Jeremy / Logan / Steve
→
⭐
Birdeye
Review request
Safety escalation
Three conditions bypass all AI and route directly to a human dispatcher within 5 minutes: gas smell, smoke inside the structure, or CO alarm activation. Claude returns safety_escalation: true and no further AI processing occurs.
Rollout Phases
60-day implementation pathway
Phase 1
Foundation — Communication & AI Backbone
Days 1–7 · No custom code required
Configure Twilio inbound SMS and voice webhooks. Set up Claude API in Make.com (claude-sonnet-4-6 for standard triage). Build the core Make.com scenario: Twilio → Claude → JSON parse → Striven job creation. Load the Charleston Location Code Matrix as a Make.com Data Store lookup table.
Phase 2
Triage Logic & Job Scoring Engine
Days 5–14
Deploy Claude Triage Prompt v1.0 with safety check, guided troubleshooting, and structured JSON output. Implement job scoring formula. Configure Striven dispatch queue sorted by score. Enforce 7-job daily cap during peak season. Add $100 trip charge for Secondary Zone jobs automatically.
Phase 3
Monitor & Refine
Days 14–30
Dispatcher reviews every Claude-generated job card before finalization. Track edits with reason codes. Any issue category or ZIP combination generating >20% edits triggers a prompt refinement. Update to Claude Triage Prompt v1.1 after 30-day calibration period.
Phase 4
Full Automation & Voice Integration
Days 30–60
Deploy Twilio AI Voice handler for inbound calls. Enable Birdeye automated review requests (2-hour delay post-job completion). Remove monitor mode gating for SMS triage where confidence_score >85. Run first monthly analysis against all target metrics.
Working Decisions
Eight open decisions needed before build can advance — edit inline, saves to browser