This is not a big-bang deployment. Every phase has defined milestones, go/no-go criteria, and a board checkpoint before further capital is committed. If Phase 1 doesn't meet the bar — we stop, and the board sees exactly why.
Phase 1 — POC
6 wks
Order Status only. ~140 hrs.
Phase 2 — Production
8 wks
Full intake + routing. ~230 hrs.
Phase 3 — Scale
4 mo
Outbound AI + multi-warehouse.
Ongoing
/mo
Monthly tuning & management.
6
Weeks
~140 hrs
Ark Fortune Engineering
Pillar 1
Scope: Order Status Only
10–15%
Call Volume on AI (Live)
Fixed Fee
No Overruns
-
Week 1–2RingCentral Call Log Audit Pull 2 weeks of real inbound call data. Categorize by type. Establish the actual baseline split between order status, intake, and complex calls. This is the foundation of the entire financial model.
-
Week 2Platform Setup — Retell AI Enterprise Configure Retell AI enterprise account. Sign Business Associate Agreement. Integrate with RingCentral via SIP bridge. Set up the routing logic so AI receives the right call types.
-
Week 3Core System API Bridge (Read-Only) Build a secure, read-only API layer between the AI and the Core System's order/delivery data. Implement caller authentication flow using hospice agency credentials and patient DOB. No PHI leaves the connection without encryption.
-
Week 4Voice Agent Training Train the AI on DME Express workflows: hospice terminology, common order status query patterns, authentication dialogue, and escalation triggers. QA review with sample call scenarios.
-
Week 5Internal QA & Edge Case Testing Simulate bad data (missing delivery, patient not found), ambiguous requests, hostile callers, and out-of-scope queries. Validate escalation logic — every edge case must route cleanly to a human.
-
Week 6Live Pilot — 10–15% of Real Inbound Calls Route a small slice of actual inbound calls to the AI. Monitor in real time. Measure resolution rates, escalation accuracy, and agent feedback. All results reported to the board at the end of this week.
Phase 1 Go/No-Go Gate — Week 6 Board Checkpoint
≥30% of routed calls resolved without human transfer | <5% bad escalations | Zero HIPAA incidents
If criteria are not met → STOP. No Phase 2 capital released. Full findings presented to board.
If criteria are not met → STOP. No Phase 2 capital released. Full findings presented to board.
8
Weeks
~230 hrs
Ark Fortune Engineering
Both Pillars
Status + Intake Live
100%
Inbound Calls Routed to AI First
Fixed Fee
No Overruns
-
Week 7–8Intake Flow Design Map every required data point for a valid DME order (patient name, DOB, diagnosis, address, insurance, equipment list, hospice agency). Design the AI dialogue to collect all fields in a single call, gracefully handling incomplete information.
-
Week 9–10Core System Write Integration — Draft Order Creation Build secure write access to Core System. AI creates a Draft Order with all fields pre-filled — no data entry for the human agent. All writes are logged and auditable. Human must click approve before any order goes to dispatch.
-
Week 11–12Full Inbound Routing All inbound calls route to AI first. AI classifies: Order Status → autonomous resolution. Order Intake → collect and draft. Everything else → immediate warm transfer. Build branch-aware routing so the AI knows which Core System instance and warehouse to query.
-
Week 13Staff Training & Change Management Train the 50-person support team on the new workflow. Their job shifts from data entry to clinical review. Show them the AI dashboard — resolution rates, escalations, draft accuracy. Address concerns directly.
-
Week 14Production Stabilization Monitor live production metrics for a full week. Tune escalation logic based on real call patterns. Review draft order accuracy with operations team. Confirm support staff workflow is stable.
Phase 2 Go/No-Go Gate — Week 14 Board Checkpoint
Draft order accuracy ≥90% | AI handles ≥40% of all inbound | Human review time per order reduced ≥40% | Zero compliance incidents
4
Months
~200 hrs
Ark Fortune Engineering
Outbound
AI Calls Coordinators
All Branches
Multi-Warehouse Core System Routing
Dashboard
Leadership Analytics
- Month 4–5Outbound AI CallsAI proactively calls hospice coordinators to confirm deliveries, follow up on unsigned orders, and collect missing intake data. Reduces inbound call volume further.
- Month 5–6Multi-Warehouse IntelligenceEach branch's Core System instance is connected. The AI knows which warehouse handles which territory and routes queries accordingly.
- Month 6–7Leadership Analytics DashboardReal-time visibility into AI resolution rates, cost-per-call by branch, escalation patterns, and draft accuracy. Visible to COO and CFO.
- Month 7–8Acquisition Auto-OnboardingEvery new branch acquired by Palladium is connected to the AI system from Day 1. No additional Ark Fortune time per acquisition — it is a configuration, not a rebuild.
∞
Per Month
- MonthlyPrompt Tuning from Failed CallsReview calls that escalated unexpectedly. Identify patterns. Update AI logic to handle those cases next month.
- MonthlyPlatform & Vendor ManagementManage Retell AI account, RingCentral integration, Core System API health, and BAA compliance. DME never interfaces with AI vendors directly.
- MonthlyNew Workflow AdditionAs DME Express adds products or services, new call types can be added to the AI's scope — on a per-workflow basis, with a separate scoping discussion.
- MonthlyCompliance MonitoringReview audit logs for any PHI handling anomalies. Report to compliance team. Ensure BAA obligations are met.
Why Ark Fortune — not an outside AI firm. The AI's entire intelligence depends on how deeply it integrates with DME Express's Core System architecture. Ark Fortune built that architecture. No outside voice AI vendor understands DME's Core System data model, workflow logic, or warehouse routing rules without 3–6 months of discovery. We are not starting from zero. We are extending work already in progress.