Workload and staffing estimates

Status: PROVISIONAL WORKFORCE MODEL — explicit planning scenarios, not observed demand, CMS time standards, staffing approval or a price. Date: 2026-10-04 Unit: active work per constant-tier beneficiary-year, separately from scheduled paid hours, initial setup, standing coverage and fees. Scope: Proxi staffs/furnishes all applicable GUIDE services except respite. Household choices and independent outside care are not Proxi payroll; Proxi-contracted fulfillment is paid.

Direct answer#

The estimate now follows actual episodes rather than task counts or one unexplained annual allowance.

• Puerto Rico L2: the illustrative established-year scenarios produce 105.6 / 247.0 / 686.3 active minutes for the explicitly modeled one-caregiver known component. Add the unresolved RCC caregiver portion below; this is not a complete population forecast. The ordinary known component is 20.6 active minutes/month, not required talk time. One annual planning contact replaces a qualified scheduled contact; only genuinely separate work adds minutes.
• Clinical work: the ordinary selected split-team scenario produces 47.8 RN + 7.6 LCSW/behavioral + 98.1 medical-clinician active minutes/year. Annual assessment work is real, but the RN/medical split is a Proxi design—not a CMS requirement for a nurse to prereview every chart. A prepared medical question goes directly to its authorized clinician.
• Philippines first answer: one continuously available primary seat, with the stated shrinkage/reserve/lead inputs, requires 6.677 / 7.419 / 8.275 paid FTE equivalents. At 5,000 actually shared lives, that is 166.7 / 185.2 / 206.5 paid minutes/life-year. These are annual coverage calculations, not proof that the queue meets its response promise.
• Payroll: Leon confirmed $22 per scheduled paid L2 hour, including downtime/training (D-029). Retain the existing 160 paid hours/month × 70% productive load = 6,720 working minutes. The ordinary known component of 247.0 active minutes/year requires 352.8 scheduled paid-minute equivalents at 70% load, about $10.78 PMPM known-component L2 payroll before unmodeled additional work.
• No complete service price yet: actual PH/clinical/contractor rates and payout units, receiving availability, program work, field travel/fees, interpreters and other paid dependencies remain visible below. Unknown does not mean free.

All scenario inputs below can be changed as planning assumptions under O-039. Low/high are contrasting demand cases, not confidence intervals or limits on care. Actual necessary work continues even when it exceeds a budget. The eight service SOPs still control authority and completion; this calculator never selects urgency, treatment, eligibility, a clinician or a bill.

What the percentages mean#

One actual event/time is counted once even if it separately furnishes several services. A service label, reminder, queue, attempted call, transmission or outside appointment is not an additional event or completed service. D-023 Daily Companion input is not clinical monitoring, human contact or payment evidence. A billed month still needs actual qualifying care and current contact cadence; do not manufacture outreach to fill it.

Role and geography boundary#

Puerto Rico is U.S.-based, but its L2 operation remains nonclinical under D-015. Professional title, geography or certification does not grant an act. Every licensed act identifies its actual subject/current physical location, applicable provider-location rules, scope/modality and accepting professional. Caregiver-as-reporter is different from caregiver-own care; dual subjects retain separate results. A regular RN does not satisfy the dementia-proficient Part B E/M slot merely by holding an RN license.

Settled staffing design for this estimate#

Proxi uses PR nonclinical navigation, PH nonclinical human first answer and a separate appropriately authorized U.S. clinical workforce. Required GUIDE capability is not an optional customer-supplied staffing package. Outsourced clinical/onsite/trainer work is paid when it fulfills Proxi's promise. Independent PCP/pharmacy/facility/emergency/community acts remain their own acts; Proxi owns the applicable coordination. Actual Partner/offshore/care-team acceptance remains O-001/O-PA-001. Formal navigator contact is not automatically supplied by L1 pickup.

CMS-fixed navigator frequency#

The CMS RFA specifies the original tier cadence; the current PMP adds RCC bimonthly contacts. CMS does not prescribe contact minutes.

These are not bankable annual quotas. Use actual alignment/tier/caregiver/residence dates and the owning contact window; clustered contacts cannot replace a later due window. For low dyad/RCC, calendar-month windows reset after a touchpoint. Required interactions may be in person, telephone or audio-visual; SMS/AI is not the human contact. Actual person-specific value—not billing or target talk time—sets the agenda and endpoint.

Navigator minutes by tier#

Use Ongoing's total-active episode table, not direct talk time alone.

The episode does not have to last the estimate. Zero promises means zero promise oversight. Later distinct work is booked once to its actual owner; routine L1 chasing is not silently charged as L2 time. Screens inside the conversation allocate that conversation, not another contact per topic. Annual planning may serve as one qualified contact only when actual actor/modality/content/date satisfy it; unrelated work does not become free merely because it happened the same day.

Reproducible scheduled-contact core#

For this simple full-year scenario, one primary contact is the annual planning milestone. Its budget is the larger of the selected primary episode and source planning episode, because that meeting is the contact, not because all unrelated work can be costed by taking a maximum. Additional actual agenda/work remains additional.

Per tier: (primary contacts − 1) × P + A + focused contacts × Q.

This is an illustrative case-mix bound, not observed distribution. A more precise model weights stable/ordinary/extended episodes within each tier from data; it must still account for annual planning and second contacts once.

Distinct nonclinical work outside those contacts#

Every row below means actual L2 work not already performed in a parent contact or receiving service. A clear digital choice or administrative transaction has 0 new L2 minutes. Reach and bundle fractions are O-039 assumptions; they do not decide whether a particular person gets help.

Formula: reach × events × (1 − already-bundled share) × actual human minutes. Failed attempts consume labor but furnish no contact/service by themselves. D-007's 1 / 1 / 2 ordinary release/initiation is not blanket clinical or other-class PHI authority. Actual other release classes remain separately mapped/unpriced—not presumed to fit this row. A clinical question skips unnecessary L2 and uses its own professional event.

D-033's two accepted clean provider-sending classes add 0 extra human release minutes; do not multiply a release-check band across every routine request or approved recommendation. Actual clinical authorship, mandatory restricted-case review, manual D-031 fallback, preparation/repair and receiving availability remain their own paid work, counted once. This changes the conditional task contribution, not the incidence, hourly-rate, scheduled-capacity or portfolio assumptions above; actual class/occurrence calibration remains O-008/O-021/O-030/O-039.

Caregiver work without a flat allowance#

The known non-RCC dyad mix supplies 70% caregiver-present exposure in this illustration. RCC caregiver status is unknown, not absent. For one distinctly served caregiver-equivalent per life, add 0.077 × r /0.78 × r /8.64 × r active minutes/year to the low/ordinary/high known L2 component, where r is the actual RCC caregiver-present share. Thus ordinary L2 is 246.98 + 0.78 × r, not an unconditional 246.98 total. If r is unknown, the complete total stays unknown. Actual additional separately furnished caregiver events add their time; joint/shared work is not multiplied. The table below contains the 70% known non-RCC component only.

These are established-year repeat/need assumptions, not limits on required initial education, training availability or accepted support. Foundation caregiver assessment/plan work stays in its assessment/planning event. Initial diagnosis information and initial training are entry occasions below. Training uses one D-019 application touchpoint, not nine lectures. A group allocates one actual facilitator session across actual participants; do not pay its hourly time and the same full-session fee twice. CG26 entered via 24/7 keeps its service ownership in Caregiver Support and only actual L1 intake in Access. Contracted trainer/interpreter/provider fees are Proxi cost when contracted for our service.

Established-year human involvement#

The ordinary four modeled known components total 400.6 active minutes/year. Within these four components only, L2 is 61.7%, RN 11.9%, behavioral 1.9% and medical 24.5%. This excludes L1 pooled payroll, RCC caregiver additions, field/setup/availability, real program work and other paid fees; it is not a total-cost or whole-workforce percentage.

Clinical workload across all non-respite GUIDE services#

Annual 20 / 35 / 50 nursing and 60 / 80 / 120 medical minutes are the selected split-team scenario bands, not CMS time requirements. The current reference initial assessment is 65 clinical-support + 110 medical minutes in Assessment. Assign the actual nursing/behavioral/medical domains; do not buy a generic chart prereview. AI/L1 prepares available sources, while the clinician personally elicits/examines what is clinically necessary and performs judgment.

Book one clinical event owner, separately from its entry channel: assessment first; transition; medication outside assessment; behavioral/caregiver/safeguarding; genuinely plan-owned new question; residual unclassified 24/7 clinical issue; other contact question. This is accounting precedence, not clinical routing policy. A medication problem entering through a navigator/24-hour call stays medication-owned. Current applicable professional products are reused without another blessing; new/due/conflicting work is not waived.

Each incidence is a scenario assumption, not observed frequency or clinical policy. The residual other-contact counts are net inputs after removing issues already owned by assessment, transition, medication, behavioral, plan or residual 24/7 care; they are not 5% / 10% / 20% raw triggers that can be added again. Actual calibration assigns one owning family to each distinct clinical issue before summing its real professional contributions; an unrelated second issue may have a separate event. A medical-only case has 0 mandatory RN/L2 gateway minutes. Required specialty/pharmacy/disagreement work adds its actual authorized component; contractor fees/capacity remain separate.

The selected assessment split is not the only staffing option. If one professional is qualified and authorized to perform all assigned domains, replace the split with that actual professional's work; do not add a second RN approval. Until measured, a conservative no-time-saving comparison retains the combined 115 annual / 175 initial active-minute envelopes and applies the actual professional's payout basis. These are summed planning envelopes, not a clinical equivalence claim, required duration or permission to perform another profession's protected act. The mandatory dementia-proficient E/M capability and any additional physician-medical-director requirement still apply.

Care Plan clinical labor attribution#

Source contributions already provided by assessment/medication/transition/behavioral care stay charged there. Incremental plan-owned questions use these stated assumptions:

Between-event plan-owned questions: 0.05 / 0.15 / 0.50 per beneficiary-year,60% lower-scope (80% RN/20% behavioral) and 40% medical. This gives ordinary 1.872 RN + 0.588 behavioral + 2.88 medical minutes. Routine gathering/organizing of evidence is software/L1, not another 3-minute nurse charge. Actual clinical disagreement needing judgment is a separate scoped return, not eliminated or budgeted as clerical work. Zero incremental planning time never means no annual clinical work.

Pooled labor that case minutes miss#

24/7 human availability#

At 365 days, one primary seat needs 525,600 on-duty clock minutes. Under the existing PH convention, one FTE has 2,080 scheduled paid hours/year. Paid coverage FTE = clock minutes ÷ (124,800 × (1 − shrinkage)) × (1 + reserve) + 0.5 lead FTE.

These figures use the stated formula; no hidden extra seat is in the high row. Shrinkage and reserve must describe different needs, not count the same training/absence twice. PH's 2,080-hour convention is not PR's retained 1,920-hour convention. Change either to actual contract hours when known; do not compare them as one undefined FTE.

The ordinary illustrative activity is 25% caller reach × 1.5 calls among reached people = 0.375 calls/life-year × 10 active minutes = 3.75 call minutes, and 60% admin reach × 2.7 events × 13 = 21.06 admin minutes. At 5,000 lives, calls consume 18,750 minutes, 3.57% of one clock seat. Average occupancy does not prove answerability. Noninterruptible outbound calls, transfer holds, after-call work, simultaneous callers, language needs and outages require actual interval/backup capacity. Only work that truly fits without losing answerability is absorbed by the paid roster; it is not added to payroll twice. Overflow/another seat is separately paid. Two continuous primary seats with the same ordinary assumptions would require 14.338 FTE, not merely the 0.5 lead.

L1 first answer is nonclinical, not navigator contact or clinical service. An approved PH direct care-team support capacity would be attributed by that actual act under O-001, not nursing title. Clinical/caregiver receiving hours, response promise, acceptance, backup and fees are separate O-054/O-001 dependencies; no automatic 24 h U.S. clinician or PR shift is inferred. Failed transfer is unfinished; L1/L2 never decide waiting is clinically safe.

Shift rotation/handover also consumes paid time. For an illustrative 8-hour shift schedule, three changes/day × 10 overlap minutes × 365 days × one additional incoming worker = 10,950 paid minutes (182.5 hours/year). If the stated reserve already funds this overlap, earmark it there without a second charge; otherwise add it once to the actual schedule. Actual handover content remains owned by outgoing/incoming L1 staff and must preserve live answer/backup coverage. An overlap budget does not itself prove that another person is available during a noninterruptible handover.

Program and compliance overhead#

There is no generic program/compliance case performer. Software applies approved policy. Actual routine correction returns to its original worker/lead; real technical, HR, Billing/Revenue Cycle, Medicaid, business-commitment, nursing/medical-leadership or legal work uses that real job and its exact trigger.

Model program hours by drivers: shared portfolio work + sum(per-Participant work) + per-employee/per-life/per-claim/per-partner work + actual scale steps/incidents. The prior 300 hours at a 500-life Participant is only a calibration assumption, not a constant all-scale budget. Its rate and actual job mix remain unquoted. O-045 defines those jobs/coverage, not a universal mandatory-review matrix. Release/policy/clinical/billing/legal requirements use their own controlling authority; no four-minute anonymous case worker is priced. Keep units explicit: per-life allocation = (all actual pooled fixed and scale-step minutes) ÷ covered lives + actual variable minutes/life. Per-Participant fixed blocks enter that pooled numerator with their actual multiplicity; convert per-employee/claim/partner work to its actual total or per-life amount before combining it.

Panel-size-specific Proxi full-service subtotal#

Use the actual covered group and multiplicity of each cost pool, not one universal numeric denominator. Example: one PH roster really covers 5,000 lives across ten Participants, each 500 lives. Its 185.177 minutes/life differs legitimately from each Participant's 300 hours ÷ 500 = 36 program minutes/life. Ten program blocks mean 180,000 minutes ÷ 5,000 = 36; one genuinely shared 300-hour block across 5,000 means 3.6. The scope facts select the case. Neither proves 300 hours sufficient at scale.

Patients who do not actually share the same accepted roster/contract/hours/language/response route cannot use that divisor. One Participant may have its own fixed block even while PH and some clinical capacity are shared. Explicit variable/step work does not dilute toward zero.

Rate and payout basis#

For PR: active hours ÷ productive load = scheduled paid-hour equivalents. At ordinary 246.98 minutes/year and 70% load: 352.83 paid-minute equivalents/year; 352.83 × 22 ÷ 60 ÷ 12 = $10.78 PMPM. A full 160-hour scheduled month costs $3,520 and supplies 6,720 working minutes before any learning time excluded from that allowance.

Illustration only: if the $64/$140 clinical inputs buy delivered active work hours, the ordinary known subset at 5,000 shared lives is PR $10.78 + RN/behavioral $4.93 + medical $19.08 + PH $1.41 = about $36.21 PMPM. If clinical rates instead buy scheduled hours, apply their actual productive capacity/availability and minimum fees; do not present this as the same price. This is not a full-service quote or a revenue forecast.

Unpriced components are explicit: actual program jobs/rates/scale; clinical and approved caregiver receiving availability/backup; physician medical director if applicable; actual other release classes; field visit/travel/minimums; interpreter/accessibility and contracted training; actual workforce-learning/turnover not already included; software/telephony/other nonlabor costs and margin. Their absence cannot be read as 0. Optional enhanced commitments under O-052 have separate incremental occurrences/fees/capacity—not borrowed annual clinical minutes or 106 extra appointments.

First-year alignment add-on#

Do not use one universal add-on. Separate three clocks: initial service/setup date; successful alignment start (CMS uses initial staging date, not final BAR-notification date); and the first six alignment payment months. The annual assessment window is 306–425 days after the previous comprehensive assessment. Submit by the earlier of 60 days after starting that annual assessment or day 425 after the previous comprehensive assessment. Actual annual work falling within the cost horizon is another event, not erased by the initial visit.

The reference nonclinical entry rows sum 95.72 known-component L2 minutes before subtracting qualifying shared work; RCC/multiple-caregiver additions remain unresolved, and this is not a mandatory duration or an add-on to every existing contact. Initial skills application, enrollment exceptions and actual extra early support still use their own accepted occurrence/share. A welcome, plan or assessment replaces a scheduled contact only if actual dates, actor, modality and content qualify; otherwise it is separate. Initial source routes and low/high cases remain in their detailed SOPs, not all forced into this typical bundle.

Define the first-year workload horizon as 365 days beginning on the initial assessment's dementia-staging date (day 0 through day 364). This differs from a calendar year, completed-assessment date, final BAR notice and the first six payment months. The illustrated initial-plus-first-year cost view includes the full initial episode, including preparation before day 0, plus other actual work within the interval. Do not charge the same initial work again inside the interval or acquisition allocation. Count unfinished annual work by its actual incurred role minutes.

For selected whole annual bundles fully inside the horizon, initial event 65 / 110 + f × annual event 35 / 80 gives average nursing/medical minutes: f = 0 →65 / 110; f = .5 →82.5 / 150; f = 1 →100 / 190. A half-cohort scenario is not a half-completed clinical service. When roles work across the boundary, instead use 65 + actual annual nursing minutes/life and 110 + actual annual medical minutes/life incurred in the horizon. Example: 35 nursing minutes before the boundary and 0 medical minutes before it yields 100 / 110, not 100 / 190. Partial/extra work retains actual incurred minutes without a service-completion claim or a cap at the modeled envelope. No accounting choice authorizes delay beyond the annual 306–425/earlier 60-day/day 425 deadline. The 30 / 30 difference describes one initial event instead of one annual event only; it is not universal first-year workload.

Initial onsite reach = moderate/high non-RCC tier share + low-tier share × actual elected-onsite rate + RCC share × the actual controlling applicability. Current non-RCC mix gives 50% before elective low visits; at 40 reference minutes that known part is 20 expected onsite minutes. RCC applicability is unknown pending the controlling Agreement term—not assumed exempt—and low election adds work. Field fee/travel/failed visit/extra documentation and clinically inadequate telehealth follow-up are actual separate paid work. D-028 does not add a second full clinical assessment merely because the clinician joins remotely.

Prealignment candidate costs that never produce final alignment belong in acquisition/failure cost, not vanished labor or furnished GUIDE claims. Per-successful-alignment acquisition allocation = total included candidate costs ÷ actual successful alignments. Revenue timing and patient/worker clinical need are separate; a partial first alignment month consumes payment month one, not a full workload month.

What each candidate workforce actually replaces#

L0 prepares/retrieves/drafts approved information and rule-constrained work; it never substitutes for the person's choice or protected judgment. L1 replaces manual administrative pursuit and supplies nonclinical human first answer. L2 supplies actual needed/permitted navigation/coaching, not generic factual approval. RN supplies actual nursing assessment; behavioral clinician supplies their licensed component; the medical clinician supplies medical decisions/reconciliation/orders. Qualified contractors supply actual field/specialty work when funded, otherwise Proxi supplies honest coordination around outside care. One employment category may perform many permitted duties; it does not become a new role for every task.

Capacity implication#

Keep the existing PR convention: 160 paid hours/month = 1,920/year; at 70% productive utilization this is 80,640 working minutes/year. The ordinary known component of 246.98 active minutes/year gives an arithmetic 326.5 lives/FTE before unresolved RCC caregiver additions and other work excluded from the model. Low/high give 763.6 / 117.5. These are mathematical capacities, not staffing promises, guarantees or a recommended pilot caseload. Actual failed attempts, spikes, language, clinician handoff, leave, supervisor work and all excluded duties must fit the accepted load.

Learning and replacement without double counting#

CMS requires at least 20 initial navigator hours (at least 10 didactic and 10 live experiential), comprehension assessment and 2 additional hours annually. Software can remove instructor time for approved web didactic content, not paid learner hours. Live shadowing/instruction counts learner time and only actual incremental trainer time; do not rebill the experienced navigator's already-counted patient service.

• Included in 70% productive allowance: earmark actual training/leave/meetings/turnover components within its 576 nonservice hours/year; do not add learner payroll again. Verify the allowance actually covers them.
• Not included: deduct actual paid learning/replacement time from 80,640 service minutes. An example new hire's 20 initial + 2 annual hours is 1,320 minutes, leaving 79,320 usable minutes; actual assessment/trainer/handover time adds only once.
• Replacement/turnover: count actual hires, vacancy/overlap, recruit/credential work by HR, learner/trainer time and coverage while unavailable. A percentage assumption must name its hires and hours; it is not an anonymous beneficiary allowance.

The $22 paid-hour cost is unchanged by these accounting choices. Paid capacity and cost per served life change; a true contracted fee already including training is not charged twice. Medical-director/clinical learning and PH roster training have their own actual hours/payout basis, not automatically the navigator allowance.

For a concrete replacement sensitivity, assume the following—not measured turnover statistics:

The ordinary case earmarks 7.6 of the existing 576 nonservice hours if included; it does not raise payroll. A newly opened seat separately needs its initial 20 + annual 2 hours, not both this full new-seat training bundle and another identical initial-training charge. Incremental trainer time, recruitment, paid overlap/coverage and lost vacancy capacity use their actual workers and schedule; they are not hidden in the learner formula. This makes replacement sensitivity explicit without claiming an observed navigator turnover rate or deducting it twice.

What is not yet in the estimate#

Rates/fee bases and actual paid availability listed above; Participant-specific RCC/offshore/Partner/release acceptance; clinical domain/scope/modality coverage; caregiver-present/acceptance and group economics; changing tier/exposure; actual participant/claim/partner/employee scale work; first-horizon annual work and failed alignment; noninterruptible PH demand/peak queues; extra fees/travel/interpreters; enhanced package selection. These are missing evidence or inputs, not new invented specialists or 0-cost obligations.

Calibration data needed from the first operating cohort#

Capture actual role/accepted capacity, supplying/payee organization, subject/current location and clinical act where relevant; event owner and entry channel; contact tier/window/exposure; direct/prep/own-record/follow-through/failed-attempt/travel minutes; shared parent work already counted; actual service/result/non-success; caregiver participation and group actual attendance; paid roster hours/rate basis/minimums/fees; hires/learning/replacement; pool membership/multiplicity and interval availability/arrival/concurrency. These are measurement questions, not an implementation schema.

Replace assumptions with source-auditable cohort values once enough examples exist; do not declare stability merely at 100 people/90 days. Recompute totals from actual exposure and distributions, preserve unknowns and rare expensive cases, and show sensitivity to caregiver/tier/role/queue changes. The actual worker completes clinically necessary care, not a modeled minute quota.

Source and decision boundary#

Primary current anchors: CMS GUIDE RFA ↗, PMP v3.0 ↗, current FAQ ↗, and HHS actual subject/provider location licensing ↗. Current CMS facts were checked 2026-10-04. The source register and executed Agreement hierarchy still control; no time/rate/provider/geography assumption is a CMS mandate.

D-012/D-015/D-016/D-019/D-021/D-026/D-028/D-029 control service ownership, clinical separation, count-once work, actual human training, preparation, first answer, hybrid field and scheduled-hour cost. O-039 accepts/replaces quantitative assumptions; O-045 assigns real enabling jobs; class-specific release decisions remain separate; O-051 transition work, O-052 extra promises, O-054 receiving availability, O-060 actual pooled cost scope and O-PA-001 controlling terms remain open. No quote or operational staffing approval is claimed.