How-To Guide
How to Use AI for a Home Care Agency in 2026: Intake, OASIS-E1, PDGM, Caregiver Matching & Owner Scorecard
Published May 21, 2026 · 17 min read · For home-care-agency owners of 20-300 caregiver agencies — Medicare-certified home health, Medicaid personal care, private-duty, and hybrid.
TL;DR
- Two AI layers for a home-care agency in 2026: a clinical layer (intake + LOC triage, OASIS-E1 SOC, PDGM case-mix + LUPA risk, medication reconciliation, denial appeals) and an operations layer (caregiver hire + background + matching, EVV exceptions, telephony rebook, scorecard).
- Ten prompts below: client intake synthesis, OASIS-E1 drafter, PDGM + LUPA projection, caregiver-match ranker, EVV-exception triage, DOL companionship-exemption check, denial appeal, family update, caregiver sentiment, owner scorecard.
- AI drafts; the RN signs OASIS; the administrator signs operations. HIPAA 45 CFR 164 minimum-necessary duty, CMS CoP 42 CFR 484 accountability, and state AL / Medicaid aggregator compliance sit with the agency — AI does not absorb any of them.
- PHI goes only into BAA-covered tenants. Caregiver-matching AI must rank on skills and non-protected preferences only — Title VII and state fair-employment laws still apply.
- ROI is real: same-day fill rate +15-30 pts, OASIS rework -40-60%, PDGM per-episode revenue +3-6%, caregiver turnover -10-20% — but only if fill-rate, median-days-to-fill, voluntary-turnover, and LUPA % are tracked weekly.
The 2026 home-care agency AI stack
| Layer | Tool | Use |
|---|---|---|
| EHR + agency | WellSky Home Health, HCHB Homecare Homebase, MatrixCare, Netsmart myUnity, PointClickCare Home Health, Axxess, Alora, Kantime | Clinical + operations |
| Personal-care agency | AxisCare, ClearCare (WellSky Personal Care), Alayacare, Smartcare, Generations, Rosemark, AdaCare | Scheduling + billing |
| OASIS-E1 assistance | Axxess OASIS AI, WellSky OASIS Analyzer, Forcura, Medalogix Pulse, Muse Healthcare, Trella Health | SOC drafting + QA |
| EVV | Sandata, HHAeXchange, Tellus, CareBridge, DataLogic, Netsmart CareAssurance | Visit verification + exception |
| Caregiver matching | AxisCare AI Match, Smartcare Scheduler, Paradigm, CareAcademy, Honor, Nevvon | Skill + proximity ranking |
| Ambient + dictation | DeepScribe, Abridge, Suki Assistant, Nuance DAX Copilot, Heidi, Freed, Sunoh.ai | Visit-note drafts |
| Telephony + recall | CallRail CI, Avoca AI, Weave AI, Doctible, Klara, Phreesia AI, BirdEye, Podium | Rebook + family updates |
| Denial + PA | Waystar, Availity, Change Healthcare, CoverMyMeds, Glidian, Myndshft | Appeal + ADR response |
| Quality + star | HHCAHPS + Care Compare Star Rating, CMS iQIES, CASPER, Medalogix Muse, Trella Marketscape | Star + referral intel |
10 copy-paste prompts for your home-care agency
1. Client intake + level-of-care triage
You are a home-care intake assistant. From the referral source + caller intake + prior hospital discharge summary + medication list, produce a 1-page synthesis:
1. Referral source — hospital case manager, SNF, ACO, MA plan, private pay
2. Primary diagnosis + ICD-10 + homebound status (for Medicare-certified home health: confined to home + normal inability to leave + requires taxing effort)
3. Service type needed — Medicare home health (skilled intermittent), Medicaid personal care (ADL), private-duty (shift), hospice referral if applicable
4. Insurance verification — Medicare FFS, MA plan contract, Medicaid state + managed-care plan, long-term care policy, VA, private-pay
5. Homebound qualification evidence (Medicare CoP 42 CFR 409.42)
6. Service intensity — visit frequency, PT/OT/SLP/MSW/HHA needs
7. Red flags — fall history, wound stage 3-4, behavioral-health crisis risk, caregiver-burnout risk, food or medication insecurity
8. Estimated PDGM clinical grouping (MMTA-Cardiac, Neuro-Rehab, Wounds, Behavioral-Health, MS-Rehab, etc.)
Do NOT certify homebound status or LOC. The RN on SOC signs.
2. OASIS-E1 start-of-care drafter
You are a home-health OASIS assistant (BAA-covered). From the SOC visit dictation + hospital DC summary + medication reconciliation, draft OASIS-E1 M-items:
1. M0-series — patient + episode demographics
2. M1000 — inpatient facility discharge within 14 days (admission source for PDGM)
3. M1021-M1023 — primary and secondary diagnoses ICD-10-CM (PDGM clinical group driver)
4. M1800-M1890 — ADLs + IADLs (functional-impairment level — low / medium / high)
5. M1900 — prior functional status 14 days before SOC
6. M2001-M2020 — medication reconciliation and management
7. M2100 — types and sources of assistance
8. M2310 — reason for emergency department use
9. M2410 — prior inpatient admission
10. GG-items — standardized functional assessment (self-care, mobility)
11. SDOH items — housing, food, transportation, utilities, health literacy
12. Transfer-of-health information (TOH) items — medication reconciliation to next provider
Cross-check internal consistency:
- M1800 ADLs vs M2020 medication management
- GG-items vs M1800-M1890
- Homebound status vs ADLs
Cite CMS OASIS-E1 Data Set Version effective Jan 1 2025.
RN reviews + signs before iQIES transmission. Do NOT upcode for PDGM.
3. PDGM case-mix + LUPA projection
You are a PDGM case-mix analyst. From the draft OASIS-E1 + plan of care + referral timing, project the PDGM HIPPS code:
1. Admission source — community vs institutional (hospital / SNF / IRF / LTCH in 14 days)
2. Timing — early (1st 30-day period in home-health stay) vs late (2nd+)
3. Clinical grouping (12 categories: MMTA-Other, MMTA-Cardiac, MMTA-Endocrine, MMTA-GI-GU, MMTA-Infectious, MMTA-Respiratory, MMTA-Surgical-Aftercare, Behavioral-Health, Complex-Nursing, MS-Rehab, Neuro-Rehab, Wounds)
4. Functional-impairment level — low / medium / high (from OASIS GG + M1800 + M1900)
5. Comorbidity adjustment — none / low / high (secondary diagnoses interaction)
6. Projected HIPPS code + case-mix weight + base payment
LUPA risk check:
- LUPA threshold for the clinical group (varies by group, 2-6 visits)
- Projected visit count in the 30-day period
- If near threshold — flag to clinical manager. Do NOT add visits that are not clinically indicated.
Documentation gaps:
- Missing secondary diagnoses that would shift comorbidity tier
- OASIS M-items not aligned with plan of care
Clinical manager + RN review. Coder signs. Truthful under False Claims Act — no upcoding.
4. Caregiver match ranker (Title VII compliant)
You are a caregiver-match assistant. From the open visit request + caregiver roster + client preferences, produce a ranked top-5 caregiver list:
Rank on (permitted factors only):
1. Required skills — CNA / HHA / LVN certification, state-specific competency, Hoyer lift, tracheostomy care, G-tube feeding, wound care, dementia care, insulin administration
2. Distance — miles / drive-time from caregiver home to client home
3. Availability — day / time / shift length matches open block
4. Language — if client has limited English, match to caregiver who speaks the client's primary language
5. Prior positive history with this client
6. Non-protected preferences — pet-friendly, smoker-free, CPR-certified, TB-test current, flu-vaccinated per agency policy
7. Gender (for personal-care intimate tasks where EEOC BFOQ-adjacent applies — bathing, toileting) — only when clinically documented
Exclude from ranking (Title VII + state fair-employment):
- Race
- National origin
- Religion
- Age (ADEA)
- Disability unrelated to caregiving competency
If a client directs exclusion based on a protected class, flag to administrator. Do NOT execute the exclusion. Document the refusal per agency non-discrimination policy.
Scheduler reviews before assignment. Consent from caregiver to take assignment required.
5. EVV exception triage
You are an EVV exception assistant. From the Sandata / HHAeXchange / Tellus / CareBridge / DataLogic / Netsmart CareAssurance EVV export, triage exceptions:
1. Parse exception type — missed clock-in, missed clock-out, duration outside expected range, GPS outside approved address, caregiver-client mismatch, manual entry, phone type mismatch
2. Categorize by Medicaid managed-care plan + state aggregator (open vs closed model)
3. State-specific — CA, TX, NY, FL, OH, PA, IL, MI, NC, GA EVV aggregator
4. Documentation required for each exception type
5. Draft caregiver follow-up + client follow-up (HIPAA-safe — no PHI in SMS subject)
6. 21st Century Cures Act EVV required data elements — type, individual receiving, individual providing, date, location, time start / end
7. Audit-sample flag — payer ADR likely
8. DOL Home Care Final Rule — travel-time between clients is compensable; make sure exception resolution does not alter timesheet
Supervisor reviews before Medicaid billing. Never alter EVV without documented reason + audit trail.
6. DOL companionship exemption check
You are a wage-hour compliance assistant. From the caregiver role + duty breakdown + timesheet, assess DOL Home Care Final Rule (29 CFR 552) status:
1. Employer type — third-party employer (agency) vs household individual — third-party employer may NOT claim companionship / live-in exemption
2. If third-party employer — caregiver entitled to federal minimum wage + overtime after 40 hours — per 2015 DOL rule and Home Care Ass'n v. Weil 2015 DC Cir affirmation
3. Travel time between clients on same day — compensable
4. Waiting time engaged to wait — compensable
5. Sleep time — for shifts 24 hours+ in employer home, up to 8 hours may be excluded only with written agreement + adequate sleeping facilities + uninterrupted
6. State wage-hour — state minimum wage + state overtime (daily OT in CA / AK / CO / NV; domestic-worker bill of rights in CA / NY / IL / MA / NV / OR / CT)
7. State sick leave + PTO accrual
8. Caregiver classification — W-2 employee vs 1099 — misclassification is IRS + DOL + state audit risk
Payroll administrator reviews. If in doubt — classify as W-2 + pay OT.
7. Denial + ADR appeal drafter
You are a home-health denial appeal drafter. From the denial letter + OASIS + visit notes + plan of care + physician certification, draft a Level 1 appeal:
1. Denial reason parsing — homebound not established, medical necessity, skilled service, face-to-face encounter, plan of care, OASIS inconsistencies, PDGM downcoding
2. Medicare CoP 42 CFR 484 citation (homebound 42 CFR 409.42, plan of care 484.60, face-to-face 42 CFR 424.22)
3. MAC LCD + NCD references (Medicare Administrative Contractor — CGS, Palmetto, NGS, Noridian, WPS)
4. Evidence — homebound narrative with specific examples of taxing effort, skilled service documented per visit (not every visit, but consistency), MD orders + face-to-face encounter within 90 days prior / 30 days after SOC
5. PDGM appeal — if downcoded clinical group or comorbidity tier, provide secondary-diagnosis documentation
6. Timeliness — 120 days from MSN (Medicare) or plan-specific for MA
7. Redetermination → Reconsideration (QIC) → OMHA ALJ → Appeals Council → Federal Court — ladder
8. MA plan internal appeal path + peer-to-peer
9. ADR (Additional Documentation Request) — 45-day response + full episode record
DON manager + administrator reviews + signs. Truthful under False Claims Act.
8. Family update letter (HIPAA-safe)
You are a family-communication assistant. From the visit notes + HIPAA authorization on file, draft a weekly family update to the authorized representative:
1. Greet by authorized family contact name (verified on HIPAA authorization 45 CFR 164.508)
2. Week's visits — date + caregiver + service type
3. Observations — mood, appetite, ADL performance, medication adherence, fall events, skin integrity
4. Care-plan updates — PT progression, wound-care status, behavioral-health check-ins
5. Upcoming — MD appointment, medication refill, lab draw, scheduled team meeting
6. Caregiver change if any (no reasoning beyond "schedule")
7. Action items for family — pharmacy pick-up, groceries, equipment
8. Private channel for questions — agency office line + on-call RN
9. Never disclose PHI to non-authorized contacts; verify every sender of incoming family email
10. State DV / APS / protected-address flag — suppress address + caregiver identity for clients with ACP
Care manager reviews before send. HIPAA-safe.
9. Caregiver sentiment + retention trigger
You are a caregiver retention assistant. From caregiver texts to scheduler + call transcripts (BAA-covered) + EVV patterns + shift-decline rate, produce a weekly at-risk list:
1. Sentiment signals — frustration, burnout language, negative references to clients / supervisors (BAA AI sentiment only — never on personal devices)
2. Behavioral signals — late clock-in trending, shift-decline rate 2x baseline, missed EVV rising, PTO pattern changes
3. Tenure bucket — 0-30 day, 31-90, 91-180, 6-12mo, 12mo+ — different retention plays per bucket
4. Reason-hypothesis — pay, commute, client mismatch, supervisor conflict, scheduling fit, benefits, PTO
5. Retention intervention suggestion — 1:1 with scheduler (not manager-skip-level first), pay-grade check, commute re-match, client re-assignment
6. Escalation path — supervisor → DON / administrator → owner
7. Do NOT disclose individual caregiver sentiment data to other caregivers or clients — HR confidential
Agency administrator reviews weekly. At-risk list is HR-confidential. Coaching, not discipline.
10. Owner monthly scorecard
You are a home-care-agency analyst. From the EHR + scheduler + EVV + billing + HHCAHPS, produce a 1-page owner scorecard:
1. Census — SOC, ROC, discharges, active clients by payer (Medicare FFS, MA, Medicaid, private-pay)
2. Visit volume — by discipline (RN, LVN, PT, OT, SLP, MSW, HHA)
3. PDGM case-mix — HIPPS distribution + LUPA rate by clinical grouping
4. Star rating trend — Care Compare HHCAHPS + quality measures (Acute Care Hospitalization, ED Use, Timely Initiation of Care, Medication Education, Breathing / Moving / Bathing / Pain improvement)
5. Fill rate — same-day, 24hr, 48hr; median days-to-fill; unfilled hours by day-of-week
6. Caregiver metrics — voluntary turnover by tenure bucket, new-hire-to-first-shift days, retention call execution rate
7. Revenue cycle — AR aging (current, 30-60, 60-90, 90+ days), denial rate by payer + reason, ADR response timeliness
8. Operations — EVV exception rate + resolution SLA, no-show rate, complaint log
9. Regulatory — OASIS accuracy rate, state survey deficiencies open, CMS CoP compliance
10. Top 3 actions for next month with owner
Owner + administrator reviews. No PHI.
Compliance floor — the guardrails the prompts assume
- HIPAA 45 CFR 160 + 164 — BAA required for every AI vendor touching PHI; minimum-necessary; §164.514 de-identification; §164.502(a)(5)(iii) Dec 2024 reproductive-health final rule; §164.508 family authorization; HHS OCR enforcement.
- CMS Medicare Home Health CoP 42 CFR Part 484 — plan of care 484.60, comprehensive assessment 484.55, QAPI 484.65, personnel qualifications 484.80, infection prevention 484.70; 42 CFR 409.42 homebound; 42 CFR 424.22 face-to-face encounter.
- CMS OASIS-E1 Data Set — effective January 1 2025 replaces OASIS-E; SDOH items, TOH items, medication reconciliation; iQIES transmission; CASPER reports.
- PDGM (Patient-Driven Groupings Model) — 30-day payment periods; 432 case-mix categories; LUPA per clinical-group threshold; admission source + timing + clinical group + functional-impairment + comorbidity adjustment.
- Electronic Visit Verification (EVV) — 21st Century Cures Act — mandatory for Medicaid personal-care; six required data elements; state aggregator model (open vs closed); Sandata, HHAeXchange, Tellus, CareBridge, DataLogic, Netsmart CareAssurance.
- DOL Home Care Final Rule 29 CFR 552 — third-party employer cannot claim companionship or live-in exemption; Home Care Ass'n v. Weil DC Cir 2015 affirmed; travel-time compensable; state OT rules (CA, AK, CO, NV daily OT; domestic-worker bill of rights CA, NY, IL, MA, NV, OR, CT).
- Title VII + ADEA + ADA + state fair-employment — caregiver matching cannot exclude on race, national origin, religion, age, disability unrelated to competency; EEOC BFOQ for intimate personal-care tasks documented carefully.
- State home-care agency licensure — varies by state (CA HCSB + HHS, FL AHCA, TX HHSC, NY DOH, OH ODH, IL DPH, GA DCH, NC DHSR, PA DOH, MI LARA); Medicaid managed-care contracts; private-pay registration.
- False Claims Act + Anti-Kickback + Stark — truthful billing, no upcoding PDGM, no marketing kickbacks, no sham MSO; OIG home-health fraud advisories.
- FTC Endorsement Guides 2023 + Fake Reviews Rule 16 CFR 465 — $51,744 / violation FY 2026; no fake Care Compare / Google / Caring.com reviews; state UDAP.
- TCPA 47 U.S.C. §227 + state mini-TCPA — CA, FL, MA, WA, PA, IL, MT, NH, CT, MD; FCC 2024 one-to-one consent; quiet hours 8am-9pm local.
- 21st Century Cures information blocking 45 CFR 171 — patient access to records; TOH interoperability; USCDI v4.
60-day rollout plan
- Days 1-10: Sign BAAs with every AI vendor. Stand up AI-use policy covering OASIS assistance, caregiver matching, EVV triage, family communication. Patient-and-caregiver consent verbiage in intake + hire.
- Days 11-20: Baseline same-day fill rate, voluntary turnover by tenure, PDGM HIPPS distribution, LUPA %, OASIS rework rate. Train schedulers on match-ranker exclusion rules.
- Days 21-30: Pilot ambient scribe with 3 RNs on SOC visits. Activate OASIS-E1 AI QA in shadow mode for 2 weeks before trusting.
- Days 31-40: Flip match ranker on for 2 scheduler pods. Turn on EVV-exception AI. Track median-days-to-fill + exception-resolution SLA.
- Days 41-50: Denial appeal drafter + PDGM projection in full. Caregiver sentiment weekly at-risk list to administrators.
- Days 51-60: Owner review — fill rate +15 pts or more, voluntary turnover -5 pts, PDGM revenue per-episode +2-4%, OASIS rework -30-50%. Keep what moves the number.
8 common mistakes
- Letting AI answer a caller about homebound qualification without RN sign-off — CoP violation.
- Excluding caregivers from a client assignment on race, national origin, or religion — even when the client requests it — Title VII + state law exposure.
- Upcoding PDGM clinical grouping to hit a higher case-mix weight — False Claims Act.
- Altering EVV clock-in or clock-out without a documented reason — Medicaid managed-care audit finding.
- Treating caregivers as 1099 to avoid OT — DOL + IRS + state audit.
- Sending PHI via non-BAA AI (consumer ChatGPT, non-tenant Copilot) — HIPAA breach per §164.
- Replying to a public Care Compare or Caring.com review in a way that confirms a reviewer is a patient — HIPAA violation.
- Sampling OASIS-E1 M-items from a stale template without the Jan 1 2025 update (SDOH, TOH, medication reconciliation) — data-set version mismatch and payer downcoding.
FAQ
Can AI draft OASIS-E1 start-of-care assessments?
Only as decision support. The clinician (RN, PT, OT, or SLP doing the SOC visit) still signs the OASIS-E1 and carries the M-item responsibility. AI assistants inside WellSky Home Health, HCHB Homecare Homebase, MatrixCare, Netsmart myUnity, PointClickCare Home Health, and Axxess can draft M-items from the visit note, cross-check for internal consistency (M1800 ADLs vs M2020 medication management), and flag PDGM case-mix category mismatches before export to iQIES. The CMS 2026 OASIS-E1 replaced OASIS-E on January 1 2025 — social-determinants-of-health items, transfer-of-health information, and medication reconciliation are now part of the assessment. An AI that cannot cite the current data-set version number is stale. The supervising RN remains responsible under 42 CFR 484.55 for a complete and accurate comprehensive assessment.
Is AI safe for HIPAA-covered caregiver matching?
Yes, under BAA-covered enterprise deployment. The match engines in AxisCare, ClearCare (WellSky Personal Care), CareAcademy, Paradigm, Alayacare, and Smartcare sign BAAs and keep PHI inside the tenant. The risk is not the match itself — it is federal and state anti-discrimination exposure. Title VII and state fair-employment laws prohibit matching caregivers or refusing assignments on the basis of protected class (race, national origin, religion) even when a client requests it. The 2026 practical rule: clients can express preferences on language, gender (for personal-care tasks the EEOC has recognized as BFOQ-adjacent), cultural-food familiarity, and pet-friendliness. They cannot direct the agency to exclude protected-class caregivers. A compliant AI ranks on skills, certifications, distance, availability, and non-protected preferences only.
How does PDGM case-mix affect AI decision support in 2026?
Under Patient-Driven Groupings Model (PDGM), every 30-day home-health payment period is classified into one of 432 case-mix categories based on admission source (community vs institutional), timing (early vs late), clinical grouping (12 groups including MMTA-Cardiac, Neuro-Rehab, Wounds, Behavioral-Health, MS-Rehab), functional-impairment level (low/medium/high from OASIS), and comorbidity adjustment. A 2026 AI should project the PDGM HIPPS code from the draft OASIS and flag LUPA (Low-Utilization Payment Adjustment) risk — if projected visits are near the LUPA threshold for the case-mix group, either the plan of care needs clinical justification for more visits or the finance team needs to know the period will pay per-visit. The AI never changes the clinical plan to hit a payment tier — that is upcoding exposure under the False Claims Act.
What about Electronic Visit Verification (EVV)?
EVV is mandatory for Medicaid personal-care services under the 21st Century Cures Act, and most states have activated it for home-health services as well. Every visit needs (1) type of service, (2) individual receiving service, (3) individual providing service, (4) date, (5) location, (6) time the service begins and ends. The 2026 AI adds value at the exception queue — missed clock-in, GPS outside approved address, duration outside expected range, caregiver and client mismatch. Sandata, HHAeXchange, Tellus, CareBridge, DataLogic, and Netsmart CareAssurance all have EVV exception AI. The state aggregator (open EVV vs closed vendor-based model) varies — CA, TX, NY, FL, OH, PA, IL, MI, NC, GA each have their own. Never override an EVV exception without a documented reason; Medicaid managed-care auditors sample them.
What is the ROI for a 20-300 caregiver home-care agency?
Three concentrated wins in 2026: (1) caregiver fill rate — AI-assisted matching lifts same-day fill rate 15-30 percentage points by widening the candidate pool to qualified near-by caregivers, directly recapturing canceled visits; (2) OASIS accuracy — AI consistency checks reduce SOC/ROC rework by 40-60% and improve PDGM case-mix capture (not upcoding — capturing the right category the documentation supports), lifting per-episode revenue 3-6%; (3) caregiver retention — conversational AI (Paradigm, CareAcademy) surfaces frustrated caregivers via sentiment scoring before they quit, enabling same-week retention calls; turnover-attributed recruiting costs drop. Agencies that capture the margin are the ones that measure fill-rate by day-of-week, median-days-to-fill, voluntary-turnover by caregiver-tenure bucket, and LUPA % by clinical grouping — not agencies that buy the most AI tools.
Sources & further reading
- CMS Medicare Home Health Conditions of Participation 42 CFR Part 484
- CMS OASIS-E1 Data Set + Guidance Manual (effective January 1 2025)
- CMS PDGM (Patient-Driven Groupings Model) + CY 2026 Home Health PPS rule
- 21st Century Cures Act Electronic Visit Verification §12006
- DOL Home Care Final Rule 29 CFR 552 + Home Care Ass'n v. Weil (DC Cir 2015)
- HIPAA 45 CFR 160, 162, 164 + 2024 Reproductive Health Privacy final rule
- FTC Endorsement Guides 2023 + Fake Reviews Rule 16 CFR 465
- Care Compare HHCAHPS + Quality Measures + Star Rating methodology