Work

Client names are kept private. These stories describe real problems and real solutions with identifying details removed.

Overnight Medicaid eligibility checksMulti-state Personal Care and Home Health operator
Problem
Staff checked client eligibility by hand in the state portal. Lapses were often found only after claims were denied.
Solution
An automated weekly batch that submits eligibility requests, parses the responses and routes each lapse or change to the right coordinator.
Result
[X] staff hours saved per week and [X%] fewer eligibility denials
AI outbound call agentHome-based care call management
Problem
Coordinators spent hours each day on routine confirmation and follow-up calls.
Solution
A voice agent built into the agency's call management app. It places routine calls, records outcomes and hands anything unusual to a person.
Result
[X] calls per week handled without staff time
One version of the truthMulti-line home-based care operator
Problem
Each line of business reported from a different system, and leaders couldn't agree on basic numbers.
Solution
A unified data platform and Power BI models with shared KPI definitions, plus reconciliation reports comparing manual and system-captured data.
Result
Executive reporting moved from [X days] of manual prep to automatic daily refresh
Clinical forms rebuilt as web appsNursing documentation
Problem
Nursing forms lived in a rigid e-sign platform that was hard to change and expensive per use.
Solution
Standalone fill-and-sign web forms the agency owns and can update in-house.
Result
[X] forms moved, with lower per-form cost and faster changes

Sample operational dashboard

A working copy of a weekly operations dashboard I build for multi-line agencies. Every agency, branch and number here is synthetic.

Operational performance

Home Health, Hospice and Personal Care. Sample data through the week ending . Current period is the last 4 weeks compared with the prior 4.

Census trend and year-end projection

Weekly average daily census for the last 52 weeks, with a linear trend projected to the end of the year.

Admissions and discharges

Last 13 weeks. Discharges above admissions in 3 of the last 4 weeks is flagged as a shrinking pipeline.

Home health growth projection model

The census and revenue model I use for growth planning. Change any input to see 12 months of census, admissions, referrals and revenue. Starting values are illustrative.

Census and mix

Flow
Episode length (days)
Revenue per period ($)

Periods per month = 30 / episode length. Admissions needed = net census growth + discharges. Referrals needed = admissions / conversion rate.

Solutions I can demo

Each one runs on sample data. Ask for a walkthrough on a free consult.

  • Eligibility agent

    Batch eligibility checks with response parsing and exception routing.

    Python, Azure Functions, Azure SQL

  • Voice call agent

    Outbound visit confirmations and shift-fill calls with live handoff to staff.

    Voice AI, .NET, Angular

  • Referral intake summarizer

    Turns a faxed referral packet into a structured intake summary with missing items flagged.

    Azure AI Foundry, document AI, Python

  • Agency operations dashboard

    Census, visits, authorizations and revenue in one Power BI model.

    Microsoft Fabric, Power BI

  • Portal automation

    Reliable browser automation for payer and state portals without APIs.

    Playwright, Python

  • Codebase knowledge agent

    An AI assistant that answers developer questions about a large healthcare codebase.

    MCP, knowledge graph, Claude

Why utilization rates lie

An analysis I use with Personal Care and private duty leaders. Authorized, scheduled, worked and billed hours net out against each other in a rollup, so a perfect-looking rate can hide a program that serves almost no one correctly.

100%Scheduled to authorized
100%Worked to authorized
90%Billed to authorized

Every gate passes. Five clients, 100 authorized hours, 100 delivered. Now look client by client.

  • Client B

    Scheduled: optimalWorked: optimal

    20Auth20Sched20Worked20Billed

    All 20 authorized hours delivered and billed.

  • Client C

    Scheduled: optimalWorked: under

    20Auth20Sched15Worked15Billed

    Root cause: attendance. Scheduled right, but 5 hours of care never happened.

  • Client D

    Scheduled: optimalWorked: over

    20Auth20Sched25Worked20Billed

    Root cause: attendance. Worked 25, only 20 billable, 5 hours given away.

  • Client E

    Scheduled: underWorked: under

    20Auth15Sched15Worked15Billed

    Root cause: scheduling. 5 hours were never scheduled, so never delivered.

  • Client F

    Scheduled: overWorked: over

    20Auth25Sched25Worked20Billed

    Root cause: scheduling. Scheduled and worked 25, billed only 20.

Client A, also authorized for 20 hours, couldn't be staffed and received nothing. Because the rollup measures percentages of delivered work, clients with zero service drop out of it entirely.

Client-level distribution

HoursUnderOverOptimal
Scheduled113
Worked221

1 of 5 clients received exactly the hours they were authorized.

Net vs. absolute variance

0Net variance, what the rollup measures
20Absolute variance, what clients experienced

10 hours of authorized care never delivered, and 10 hours worked beyond authorization. Same numerator, same denominator, opposite conclusions.

Why the direction matters

  • Under-served is a care gap. Outcomes and compliance suffer, and unused authorizations draw scrutiny at renewal.
  • Over-served is margin leakage. The caregiver is paid for hours the payer won't cover.
  • The cause assigns the owner. Scheduling gaps belong to the office. Attendance gaps belong to the field.

Find the first AI win in your agency.

A free 30-minute call. Bring your messiest workflow.

Book a free consult