B
Bowers Enterprises
Prepared for Paul Williams
Direct PT/DX  ·  Direct Pay Provider Network, LLC
May 20, 2026
Engagement Proposal · Confidential

AI Document Automation
for workers' comp
medical scheduling.

A phased program to automate the manual document triage and bill-to-notes reconciliation work that today consumes 50–60 hours of staff time per week — without compromising the security posture you've spent fifteen years building.

Engagement
Phased delivery, milestone-billed
Scope
6 phases
Core Investment
$27,500
Section Guide

What's in this document.

01 Executive Summary The opportunity in one page. p. 3
02 Current State What I heard from the May 18 conversation. p. 4
03 Proposed Solution Architecture, approach, and key technology decisions. p. 5
04 Phase Breakdown Six phases — objective, scope, deliverables, acceptance. p. 6
05 Timeline & Milestones Phase sequencing and key decision points. p. 10
06 Investment Pricing, payment terms, and ongoing run-rate costs. p. 11
07 Assumptions & Dependencies What needs to be true for this plan to hold. p. 12
08 Risks & Mitigations Where this gets hard, and how we handle it. p. 13
09 Next Steps How we begin. p. 14
01 · Executive Summary

The opportunity in one page.

Direct PT/DX receives thousands of physical therapy reports each week from hundreds of clinics across dozens of software platforms. Two of the most time-consuming workflows on the team today — document triage and bill-to-notes reconciliation — are still done by hand. They don't have to be.

~2,000
Documents processed per week
50–60
Staff hours / week on this work today
70–85%
Target automation rate
~18 mo
Projected payback on direct labor savings

What we'll build

A self-contained AI document processing service that lives inside the Direct PT/DX AWS perimeter, integrates directly with Phoenix, and applies enterprise-grade large language models to the two specific workflows your team identified.

  • Document Triage — automatically identifies, deduplicates, classifies, and routes incoming PDFs.
  • Bill-to-Notes Reconciliation — verifies CPT-coded billing against documented work, with conservative pass/fail logic.
  • Human-in-the-Loop Console — surfaces the 15–30% that require judgment, with drafted provider outreach.

What this means for your team

The 50–60 hours per week your team spends opening PDFs and matching them to referrals is recovered. The reconciliation work currently limited to flagged providers expands to every bill — catching billing leakage that today goes undetected.

  • Staff capacity returns to higher-judgment work.
  • Coverage expands from "worst offenders only" to 100% of files.
  • Provider feedback loops accelerate — issues flagged within hours, not weeks.
  • Foundation for future Phoenix automations is in place.
"We could have used this five years ago. So what's another three or four months?" — Paul Williams · May 18 discovery conversation
02 · Current State

What I heard from the conversation.

Direct PT/DX runs a mature, custom-built PHP application — Phoenix — that's been refined over fifteen years to handle a complex, EDI-driven workers' comp scheduling business. The system works. The gap is in the manual judgment layer that sits on top of it.

Workflow One

Document Triage

PDFs land in two places: the SFTP "leftover" folder (files that didn't auto-match their EDI 837 bill) and the Zoho inbox (faxes and email submissions). A staff member must open each file and answer:

  • Whose patient is this? (name + DOB)
  • What date of service?
  • Which clinic and provider?
  • Is this a duplicate, or is there an addendum?
  • Which Phoenix referral does it attach to?

Then the file is archived, deleted, or uploaded with corresponding database updates.

Workflow Two

Bill-to-Notes Reconciliation

For flagged providers, the team verifies that billed CPT codes and units are actually supported by the clinical notes. A typical PT visit is four fifteen-minute units across one or more procedure codes.

  • Bills include line-item CPT codes with unit counts.
  • Notes describe work performed — but format varies wildly by clinic software.
  • Mismatches require outreach to providers.
  • Currently limited to a subset of providers because the work doesn't scale.

The bias is conservative: anything ambiguous is failed for human review.

Constraints that shape the solution

Security perimeter

HIPAA-grade posture, white-hat tested by the cyber carrier. Files must not leave the Direct PT/DX environment unless via a vetted, BAA-covered API.

AWS migration in flight

Production database, file server, and application servers are migrating from Cloud Eleven to AWS. New work should deploy directly into the AWS target environment.

Phoenix integration

Several hundred thousand lines of PHP 5.7. Integration via MySQL connector and the existing file server — not via Phoenix code changes.

03 · Proposed Solution

Build the toolkit first. Then the use cases.

A dedicated AI document-processing service runs alongside Phoenix in AWS, inside the Direct PT/DX perimeter. It uses an enterprise LLM (Claude or OpenAI under a Business Associate Agreement) for the parts of the problem that require judgment, and deterministic Python for everything else.

Architecture at a glance

Inputs

SFTP leftover folder · Zoho mailbox (fax + email) · Phoenix referral & bill data via MySQL · Provider CPT format library

Processing layer

Python service on an AWS EC2 instance · PDF text and vision extraction · Enterprise LLM calls (Claude/OpenAI w/ BAA) · Vector store for provider format memory · Deterministic matching logic

Outputs

File server uploads · Phoenix database updates · Review queue for ambiguous cases · Drafted provider outreach emails · Audit log of every decision

Key technology choices

  • Enterprise LLM, not self-hosted. Claude or OpenAI under a signed BAA gives the best accuracy and lowest operational burden. Both providers have HIPAA-eligible offerings with no data retention. We can swap to self-hosted later if posture requires it; the abstraction layer is designed for that.
  • Python service, not PHP. The LLM, OCR, and vector ecosystem is overwhelmingly Python-native. Phoenix stays in PHP. The two systems talk over the file server and MySQL — no PHP code is touched.
  • AWS-native deployment. EC2 for the service, S3 for staging, RDS or pgvector for the format library. Aligns with the migration target rather than the Cloud Eleven environment that's being retired.
  • Conservative bias by default. When the model isn't confident, the document goes to the review queue. We tune toward "miss less" over "automate more."

Why this works

Most document automation failures come from two mistakes: trying to use rules where you need judgment, and trying to use AI where you need precision. This architecture splits the work cleanly.

The LLM handles the unstructured judgment work — reading a faxed handwritten note, recognizing that Therapy South phrases 97110 differently than Encore, deciding whether a duplicate has new content.

Deterministic Python handles everything else — file movement, database updates, audit trails, integration with Phoenix. This is where reliability lives.

The result: an accuracy ceiling much higher than rule-based automation, with the operational reliability of traditional software.

04 · Phase Breakdown

Six phases, sequenced for safety.

Each phase has a clear objective, defined scope, named deliverables, and explicit acceptance criteria.

00Phase

Discovery & Foundation Design

Ground-truth before code.

The foundation phase. Before any code is written, we sign the BAA and MSA, get SFTP credentials and a representative sample of 500+ real documents covering the major providers and the long tail, catalog the format universe, define exactly which data fields we extract and how, and build the ground-truth test set every later phase will be measured against. We also finalize the LLM provider choice (Claude vs. OpenAI), the AWS deployment target, and the security review process.

Deliverables

  • Signed BAA and Master Services Agreement
  • Format catalog (provider × format matrix)
  • Field extraction specification
  • 500-document ground-truth test set
  • Architecture decision record

Acceptance criteria

  • BAA executed and on file
  • Sample documents accessible via Direct PT/DX SFTP
  • Phase 1 scope, sequence, and AWS target environment approved by Paul Williams
01Phase

Toolkit & Infrastructure

The reusable foundation everything else sits on.

Provision the AWS environment, stand up the Python service, build the integration layer between the LLM provider and Phoenix, and put logging, audit trails, error handling, and retries in place. By the end of this phase, we can ingest a PDF from any source, route it through an LLM, and write a result somewhere — generically. Phases 2 and 3 then specialize this foundation to the two specific workflows.

Deliverables

  • AWS EC2 service deployed inside the Direct PT/DX perimeter
  • LLM provider integration with BAA-covered endpoint
  • SFTP ingestion + Zoho IMAP ingestion modules
  • Phoenix MySQL read/write connector
  • Audit logging, error handling, retry framework
  • Deployment and operations documentation

Acceptance criteria

  • End-to-end test: a single document moves from SFTP → LLM → result with full audit trail
  • Service passes Direct PT/DX internal security review
  • Operations runbook approved by Mike
02Phase

Document Triage Automation

The easier of the two. Largest near-term labor recovery.

Built and shipped in six slices: identity extraction (patient name, DOB, date of service), provider classification, Phoenix referral matching, duplicate detection with addendum awareness, the action layer (archive / delete / upload + database update), and the Zoho mailbox handler that pulls fax PDFs out of email before they hit the leftover folder. Each slice is tested against the ground-truth set; the phase isn't complete until aggregate accuracy clears the agreed threshold.

Deliverables

  • Identity extraction across major provider formats
  • Phoenix referral matching with confidence scoring
  • Duplicate / addendum detection logic
  • File action automation (archive / delete / upload)
  • Zoho mailbox ingestion for emailed faxes
  • Production deployment with monitoring

Acceptance criteria

  • ≥80% of test-set documents correctly auto-routed
  • Zero false-positive uploads (precision over recall)
  • Two-week production observation period with no critical failures
  • Review queue averages <30% of daily volume
03Phase

Bill-to-Notes Reconciliation

The harder problem. Expands coverage beyond flagged providers.

The CPT-to-narrative reconciliation problem. Build the procedure-code dictionary from the AMA reference, then layer per-provider format mappings — how Therapy South phrases 97110, how Encore phrases 97140, how Cora phrases re-evaluations. The LLM reasoning layer compares billed line items against documented work and returns pass / fail / uncertain. Conservative bias throughout: uncertainty always becomes a failure routed to review. Provider-specific overrides for the largest providers. By the end of this phase, every bill in the system can be reconciled — not just the flagged-provider subset that the team can manually handle today.

Deliverables

  • CPT code dictionary with AMA reference data
  • Per-provider format mapping library
  • LLM reconciliation engine with confidence scoring
  • Pass / fail / uncertain decision logic
  • Provider-specific override modules for top providers
  • Integration with Phoenix bill and referral records

Acceptance criteria

  • Reconciliation accuracy ≥85% on test set
  • False-pass rate <2% (i.e., very rarely lets a real mismatch through)
  • Coverage expanded from "flagged providers only" to 100% of incoming bills
  • Production validation period with no regression on the triage workflow
04Phase

Human-in-the-Loop Workflow

For the 15–30% the system doesn't auto-resolve.

A lightweight review console for Debbie's team to handle the cases the system flags. Each item arrives pre-analyzed with the system's reasoning visible, drafted provider outreach (when relevant), and one-click actions to confirm, override, or escalate. The feedback channel Paul described — "the system found an eighth thing I didn't train it on" — becomes a structured loop where novel patterns surface to a human reviewer rather than disappearing.

Deliverables

  • Web-based review queue UI (internal, behind VPN)
  • One-click confirm / override / escalate actions
  • Drafted provider outreach emails for review
  • Novel pattern surfacing for ongoing tuning
  • Reviewer training session for Debbie's team

Acceptance criteria

  • Reviewers can clear daily queue in <1 hr per 100 items
  • All actions write back to Phoenix correctly
  • Feedback loop captures novel patterns to a structured log
05Phase

Hardening, Documentation & Handoff

A system the internal team can own.

Security review aligned with the standards your cyber carrier already applies. Runbooks for every failure mode. Documentation written for the PHP journeyman being onboarded — not for engineers who already know the system. Training sessions with Paul, Mike, and Debbie. After Phase 5, the internal team has the documentation, access, and confidence to run the system without ongoing reliance on Bowers Enterprises.

Deliverables

  • Security review documentation
  • Failure-mode runbooks
  • Architecture & operations documentation
  • Training sessions for internal team (Paul, Mike, Debbie)
  • Source code repository transferred to Direct PT/DX

Acceptance criteria

  • Cyber carrier review passed
  • Internal team can resolve top three failure modes without external help
  • Codebase, credentials, and infrastructure access fully transferred

Optional Add-On · Phase 6 — AI Enablement for the Phoenix Team

A separate engagement to get Paul, Mike, and Debbie operating with modern AI tooling: Claude Code inside VS Code, structured prompting for Phoenix support tasks, and AI-assisted code review for the PHP journeyman. Structured as a kickoff workshop followed by four follow-up coaching sessions. Available as a separate statement of work.

05 · Timeline & Milestones

Phased delivery over 22–28 weeks.

Sequencing is deliberate: Phase 0 de-risks everything that follows. Phase 1 builds the foundation. Phases 2 and 3 ship the two use cases. Phase 4 wraps the human layer. Phase 5 hands the keys over. Each phase has a clean handoff to the next, with go/no-go decision points between them.

Phase
Mo 1
Mo 2
Mo 3
Mo 4
Mo 5
Mo 6
Mo 7
0 · Discovery 2 wks
1 · Toolkit 3–4 wks
2 · Triage 5–7 wks
3 · Reconciliation 7–9 wks
4 · Human-in-Loop 3–4 wks
5 · Handoff 2 wks

Legend. Solid teal = active development. Outlined cells = overlap weeks with the prior phase (testing, handoff, parallel work). Bars assume a smooth path through acceptance criteria; phases can flex ±2 weeks within the total envelope.

Key milestones

Week 2

Discovery complete. BAA signed. Ground-truth test set in hand. Phase 1 approved to start.

Week 12

Document triage in production. First measurable labor recovery on the team begins.

Week 26

Reconciliation live across all providers. Full system handed off to the internal team.

06 · Investment

Pricing, terms, and run-rate costs.

# Phase
00 Discovery & Foundation Design
BAA, sample data, ground-truth test set, architecture spec
01 Toolkit & Infrastructure
AWS service, LLM integration, ingestion, Phoenix connector
02 Document Triage Automation
Identity extraction, referral matching, dedup, action layer, Zoho
03 Bill-to-Notes Reconciliation
CPT dictionary, provider mappings, reasoning engine, overrides
04 Human-in-the-Loop Workflow
Review console, drafted outreach, feedback loop
05 Hardening, Documentation & Handoff
Security review, runbooks, training, source transfer
Total Project Investment
All six phases, fixed price, end-to-end.
$27,500
07 · Assumptions & Dependencies

What needs to be true for this plan to hold.

Every estimate above assumes the items below are in place. If any of them shift materially, we'll re-plan together before scope or timeline changes are locked in.

From Direct PT/DX

  • Executed BAA and MSA before Phase 0 work begins.
  • Access to production-representative document samples via SFTP under BAA cover.
  • AWS account access for deployment of the AI service, sized appropriately for the workload.
  • Phoenix database access (read/write, scoped to relevant tables) for the service account used by this system.
  • Paul Williams available ~2 hours per week for review and decision cycles. Mike available for environment-related questions.
  • Test-set labeling: Direct PT/DX staff identifies the correct answer for the 500-document ground-truth set during Phase 0.
  • LLM provider account (Claude or OpenAI enterprise) provisioned under a Direct PT/DX BAA, billed directly.

From Bowers Enterprises

  • Weekly written status updates for the duration of every active phase.
  • All code, prompts, and documentation delivered into a Direct PT/DX-owned Git repository.
  • No data retention outside the Direct PT/DX AWS perimeter. Bowers Enterprises does not store PHI on its own systems.
  • Standard liability and confidentiality terms in the MSA, plus HIPAA-aligned obligations in the BAA.

One scope boundary worth naming

This proposal does not include changes to Phoenix's PHP code. The AI service integrates via the file server and MySQL — the same way Phoenix already exchanges data with its own automations. If a future phase requires Phoenix changes, that's separate scope and we'd quote it as such.

08 · Risks & Mitigations

Where this gets hard, and how we handle it.

No project of this size lands without friction. The honest version: here are the things most likely to go sideways, and how the plan absorbs them.

Risk · Accuracy ramp

Hitting the 80% triage threshold takes longer than expected.

Mitigation. Phase 0 builds the ground-truth set before we code. Phase 2 is sliced into six independent deliverables so we can measure accuracy incrementally and tune. If we miss the threshold at end of Phase 2, we extend by up to two weeks at no additional cost before re-scoping.

Risk · Format edge cases

Handwritten and low-quality faxed notes resist OCR.

Mitigation. These cases are explicitly out of the auto-routing target. They route to the review queue with the system's best guess and the original PDF attached — same workflow your team uses today, but consolidated in one place rather than scattered across folders.

Risk · AWS migration slippage

The broader Direct PT/DX migration runs longer than 60 days.

Mitigation. Phase 1 deploys into the AWS target regardless. If a Phoenix component is still on Cloud Eleven by Phase 2 kickoff, the service connects across — this is normal for hybrid environments. No phase blocks on the migration completing.

Risk · Security review surprises

The cyber carrier's review flags something during Phase 5.

Mitigation. Security posture is designed in from Phase 1, not bolted on at the end. Phase 0 includes a checkpoint conversation with whoever runs your security review so requirements are known before we build. Remediation work for any review finding falls inside the Phase 5 envelope unless it's a wholesale architecture change.

Risk · LLM provider change

Claude / OpenAI changes pricing or capabilities mid-project.

Mitigation. The LLM integration is built behind an abstraction layer. Swapping providers (or migrating to self-hosted later) is a configuration change, not a rewrite. Both major enterprise providers offer HIPAA-eligible tiers with similar capabilities, so a forced switch is recoverable.

Risk · Reconciliation complexity

Phase 3 turns out harder than the discovery suggests.

Mitigation. Phase 3 explicitly budgets per-provider override modules for the largest providers. If a provider's format defies general reasoning, we ship a targeted rule layer for that provider rather than expanding generic LLM work. The architecture supports this; the timeline absorbs up to three such override builds.

09 · Next Steps

How we begin.

If this proposal aligns with what you have in mind, three things move us into Phase 0 within a week.

01Step

Acceptance of this proposal

Signed acceptance below, or a counter-proposal on any line item we should adjust. Either is welcome.

02Step

Execute BAA and Master Services Agreement

Bowers Enterprises will deliver draft documents within 48 hours of acceptance. Standard HIPAA-aligned BAA, standard professional services MSA. Direct PT/DX legal review and execution.

03Step

Phase 0 kickoff

A working session with Paul Williams to map the document landscape, identify the test-set sources, and align on the architecture decisions that drive everything downstream. Phase 0 invoice issued at kickoff.

Accepted for Direct PT/DX
Paul Williams
Date
Bowers Enterprises
Chase Bowers
Date
B
Bowers Enterprises
Chase Bowers · Principal
chasebowers@gmail.com
Birmingham, AL