$1,667 per referral: why closure rates use the wrong endpoint

In August 2023, SeniorsPlus — one of Maine's three agencies that pair residents with in-home nursing care — stopped accepting new clients under MaineCare's Section 96 program, telling the Portland Press Herald that its per-member-per-minute reimbursement formula was 'not workable' after costing the agency about $600,000 the year before. By December 2024, two of the three agencies had closed themselves to new referrals.

Value-based care runs on the same trap. Linear Health finds that referrals which never complete, plus referrals whose outcome is never tracked, make up 20 to 40 percent of referral volume — and of the three possible outcomes, only the documented completed visit produces the quality scores, shared savings, and risk-adjusted revenue these contracts pay for. The other two outcomes, in Linear Health's phrasing, produce cost without revenue.

That is the endpoint the market refuses to price. Referral-management products span $3 to $12 per member per month, and the upgrade path mostly buys dashboards layered around what is really a staffing-and-workflow problem: getting a placed referral converted into a finished visit. The invoice scales with attributed lives; the closure of any single loop is, in most contracts, nobody's guaranteed job.

Anatomy of the Invoice

A referral loop is not closed when the appointment is booked, the reminder fires, or the visit happens. It closes only when the specialist's consult note is filed back into the referring clinician's chart — in practice, a FHIR DocumentReference or CCD attachment landing in the originating EHR. Closed-loop referral rate is strict arithmetic: notes returned divided by referrals sent. Vendors blur this definition because the loose version flatters the product. The pre-platform baseline is unforgiving: according to O'Malley and Reschovsky, writing in Archives of Internal Medicine in 2011, primary care physicians routinely failed to receive specialist findings for roughly half of referred patients — a closed-loop rate sitting near 50% before anyone sold software against it.

The ladder maps cleanly to function. At $3-$5 per member per month you buy a referral directory plus electronic or fax send, with status entered by hand. The $6-$8 band adds automated patient outreach — SMS and IVR nudges, self-scheduling links, the Luma Health and Artera pattern. Only the $9-$12 band layers risk stratification, bidirectional ADT feeds, analytics dashboards, and a named customer-success manager, the Arcadia/Astrana-style stack. Note what the top shelf lacks: nothing on it files a note.

Closure fails across two gaps, and the tiers hit them unevenly. Outreach engines work the first — between appointment-scheduled and visit-completed — recovering no-shows and patients who forgot prep instructions. The second gap, visit-completed to note-returned, is touched only by EHR integration work, which vendors bill separately as a project. That is the mechanical reason paying a higher tier does not automatically raise audited closure: the $9-$12 dashboard watches the second gap; it does not close it.

The billing base compounds the problem. Per-member-per-month bills against attributed member count, not referral volume — capitation, as Diligize's April 2026 guidance puts it, buys a predictable payment per patient regardless of activity. A group sending 40 referrals a month pays identically to one sending 400, so effective cost per referral swings from roughly $15 to $150+ across panel sizes: the identical tier is a bargain for one group and a structural loss for another. Maine ran the natural experiment in public — according to the Portland Press Herald, Section 96 reimburses agencies per member per minute while Section 19 pays roughly $35 per member per month. The reimbursement unit, not the headline number, sets the economics.

Three line items hide below the headline figure. Per-message pass-through fees run approximately $0.02-$0.05 per SMS at carrier rates and scale with outreach volume, not membership. One-time EHR integration projects, quoted at $5,000-$25,000, remain the only spend that attacks note-return. Navigator staffing gets called included in sales decks but capped at a fraction of an FTE in scoping documents — price that cap against the going wage, about $22-$23 per hour for care coordinators at SeniorsPlus per the Portland Press Herald, and force the cap to be stated in hours, not adjectives.

Before signature, require a scoping exhibit stating the closure denominator (notes returned ÷ referrals sent), the per-SMS pass-through rate, the integration quote, and the navigator cap in hours. What appears only in the deck does not exist at audit time.

The Evidence

Kim-Hwang et al., publishing in the Journal of General Internal Medicine in 2010, audited a safety-net delivery system and found that 35% of primary-care referrals were never completed. Sixteen years later, that number remains the load-bearing wall of the entire closed-loop category: every platform on the market positions itself as the fix for precisely this deficit. Read it as a baseline rather than a worst case — it was measured in a system with no systematic outreach layer at all, which is exactly the condition a referral platform is purchased to correct.

Now grade the evidence honestly, because this is where buyers get sold. The strongest controlled signals come from e-referral and e-consult deployments: according to RAND's evaluations of Hawaii Pacific Health's electronic referral system, specialty waits collapsed from months toward days. Note what did the work — routing, tracking, and closed communication between primary and specialty care. Workflow, not predictive analytics. Against that, the 20-to-30-point closure lifts advertised in Luma Health and Artera case studies are uncontrolled vendor self-reports. In health-services terms, that is the difference between an efficacy signal and a testimonial: directionally interesting, causally worthless until replicated on your own panel.

The practical takeaway for the contract conversation: ask every vendor for the control-group design behind its headline lift number. If the answer is a before-after chart, you have learned nothing about their product and everything about their marketing — which is precisely why the decision rule above anchors payment to a verified closure outcome rather than to a dashboard.

One number governs this entire market: $1,667 — the value a 5,000-life group must place on each recovered referral loop before the middle price band pays for itself. Every other conclusion in the tier comparison follows from it. The architecture is fixed: three rows, one per price band ($3-$5, $6-$8, and $9-$12 per member per month), scored on five columns — annual all-in cost at 5,000 attributed lives, credible closure lift, integration burden, required staffing, and the panel profile where that row wins.

The bottom row first. At 5,000 lives, the $3-$5 band bills $180,000-$300,000 a year for what is functionally directory-plus-send: an accurate specialist roster and dependable transmission of the referral order, with closure left entirely manual. That is the product design, not a defect, and it makes the band the rational pick in exactly two situations — panels under roughly 2,500 lives, where volume is too thin to justify workflow spend, and groups that already employ a referral coordinator who needs transport of referrals rather than transformation of workflow. Never buy this tier expecting the software to chase patients; it will not.

The top row disqualifies most readers of this guide. At 5,000 lives, $9-$12 PMM costs $540,000-$720,000 a year, and the premium purchases risk stratification — an engine that ranks patients by predicted utilization and is supposed to redirect tomorrow's care-management effort. That engine repays itself only when two conditions hold at once: a panel above roughly 7,500 lives and downside-risk contracts such as MSSP Enhanced Track or ACO REACH, where each avoided admission is worth thousands and risk scores feed day-to-day management decisions. A 2,500-life group paid fee-for-service under an upside-only shared-savings contract is buying an actuarial instrument for a book of business that never settles the account.

Tier Math

So the explicit winner, for the modal small group — 2,500 to 7,500 lives, mixed fee-for-service and upside-only value contracts — is the $6-$8 mid-tier, with one contractual amendment that outweighs any feature checklist: convert part of the flat fee into a per-closed-loop payment, measured against the fourteen-day consult-note standard defined earlier, so the vendor's revenue tracks the metric that actually pays. This converges with the signing rule above: a $6 PMM mid-tier contract carrying that conversion sits inside the $3-$6 closure-linked zone, while the same platform signed flat at $8 with no linkage behaves like the top band — fully prepaid, nothing at risk. Run your own version of the math before any demo: divide the quoted annual fee by your expected additional closures; if the quotient exceeds what a completed referral is worth in your contract mix, drop a band.

Every closure rate you have been shown so far shares one flaw: the endpoint is wrong. Vendor dashboards mark a loop closed when the appointment status flips to "completed" — Linear Health's own materials confirm dashboards report referral completion as the visibility metric. Clinically, the loop closes only when the consult note is filed back into the referring chart, and audited note-return rates run materially lower than the dashboard figure. As of the 2026 procurement cycle, no major vendor publishes a third-party-audited note-return rate beside its marketing closure rate. Demand both numbers in writing and watch how many replies go quiet.

The second problem is who produced the evidence. Published lift numbers come overwhelmingly from organizations that simultaneously funded referral navigators and workflow redesign, so the software's isolated contribution is statistically unidentified. The celebrated UCSF safety-net eReferral gains credited to the portal depended on centralized coordinators working the queue every day, as Chen and colleagues' Health Affairs account of the program makes explicit. Strip out the humans and the residual software effect has never been cleanly measured anywhere.

Third, the denominator you pay on moves underneath you. Attribution rosters shift monthly as patients switch Medicare Advantage plans and providers reassign panels, so a group billed on January's census routinely pays PMM on members gone by June. Realized cost can run 10-20% above sticker with no compensating service reduction offered. The countermeasure is contractual, not analytical: bill against the current-month roster and reconcile quarterly.

Fourth, reminders plateau. Reminder-only interventions reliably lift simple scheduling but stall on complex pathways — gastroenterology and endoscopy no-shows persist under SMS campaigns because preparation burden, transportation, and language barriers, not forgetfulness, drive the missed appointments. For procedure-heavy panels, a reminder module buys activity, not closures; the written >=70%-in-14-days guarantee described earlier is what forces the vendor to fund the human follow-up those pathways actually require.

What the Data Doesn't Tell You

Fifth, the noise floor. A three-provider group sending 60 referrals per month faces +/-10-point month-to-month closure swings from sampling variation alone. At that volume it is statistically impossible to verify a vendor's promised lift within a single contract year without pooling multiple quarters of data — and even then, benchmark against your own baseline, never the vendor's case-study cohort.

None of this overturns the tier logic above; it hardens it. Each blind spot is an edge case that collapses toward the same resolution: where the data cannot vouch for the software, the contract must — audited note-based endpoint, current-month denominator, multi-quarter evaluation window. A premium tier repays its price only at the scale and risk profile already specified; everywhere else, the guarantee clause, not the analytics layer, is doing the work.

Fifty-three closed loops a month is the entire asset a four-physician internal-medicine group brings to a referral-management negotiation. Picture a composite built to mirror a common 2026 profile: 3,200 attributed lives (62% Medicare Advantage, 38% MSSP-aligned), 140 referrals sent per month, and an internal EHR audit — not a vendor dashboard — showing 38% of loops actually closing. Priced honestly, both flat-tier offers on the market lose this group money; the winning contract is one almost no vendor demos unprompted.

Option C is not a product; it is a renegotiated contract: $3 PMM ($115,200 a year) plus $12 for every loop closed above a 60% floor. At the modeled 56%, the vendor misses its own floor, earns no kicker, and the group still nets +$20,700. If the vendor over-delivers to 65%, the kicker is a trivial $1,008 — 84 above-floor loops at $12 — and the group nets roughly +$89,000. The structure wins in both states of the world because the fee scales with the value event, not the panel.

Option C wins every row it appears in — including the row where the vendor misses its own floor. The lesson generalizes across the sub-5,000-life spectrum: a flat per-member fee is a population tax, while closure value scales with referral volume times lift, and 140 referrals a month cannot carry the premium tiers' fixed costs. Negotiate in sequence: install the closure-linked structure first, then ratchet the floor toward the seventy-percent, consult-note-filed, fourteen-day benchmark this guide's decision rule requires — each floor point becomes negotiable currency only after the vendor has agreed to be paid on closures, so Option C as signed is the entering position, not the destination. Finally, apply the case's rejection test verbatim: refuse any flat-tier quote whose annual fee exceeds the dollar value of that vendor's own claimed lift. Both losing options here fail it; the winning contract passes it in every scenario modeled.

The signature page, not the demo environment, is where referral-loop economics get decided. Vendors price features; your group gets paid on closures, so every clause below exists to move money out of the feature column and into the outcome column. Kill the comfortable assumption first: the premium tier is not the "safe" choice. Software modules do not call patients — outreach workflow and human follow-up close loops — which is why a hard-nosed contract on the base tier beats a generous one on the top tier anywhere below the risk-bearing threshold.

Worked Case

Rule 1 — Buy the floor, not the feature list. Sign only against a written guarantee that at least 70% of referrals close within 14 days, with "closed" meaning the consult note filed in the referring chart — the definition this guide has used throughout — audited quarterly against extracts pulled from your EHR, with automatic fee credits whenever the floor is missed. Two drafting details separate a real floor from a marketing sentence: fix the denominator as referrals sent, not referrals the vendor "accepted," and require the audit sample to come from your EHR extract rather than the vendor's own scorecard. A guarantee without a credit mechanism is a brochure.

Rule 3 — Reject status-flip metrics in writing. The endpoint flaw covered earlier has a contractual fix: require note-return rate — documents filed divided by referrals sent — as the primary KPI, and demote dashboard completed-statuses to secondary inside the reporting annex itself, not a side letter, so the demotion survives account-manager turnover. Vendors engineer toward whichever metric drives the renewal conversation; make that metric the artifact clinicians actually read.

Rule 4 — Bill on the minimum, true-up on the maximum. Attributed census moves month to month, and invoicing on peak census bills you for phantom lives. Negotiate PMM invoicing against the lowest monthly attributed census in the trailing quarter, with an annual reconciliation that trues up to actual membership — capping churn exposure at zero. Before signing, verify how the vendor defines "attributed": payer-roster dates and EHR panel counts disagree, and the gap between them is exactly where phantom lives accumulate.

Rule 5 — Pilot one specialty for 90 days before any tier upgrade. Run the base tier on a single high-volume pathway — cardiology or endocrinology, at least 30 referrals per month so the read carries enough volume — and authorize movement to the $9–$12 band only if audited closure rises at least 15 points over a pre-pilot chart-audit baseline AND at least 25% of the panel sits in downside-risk contracts that monetize the difference. Demand both conditions in writing before the upgrade meeting; a closure lift without risk-bearing revenue is a prettier dashboard, not a return.

For groups under 5,000 lives — most readers of this guide — rows one and two are the entire decision. Everything above the base tier is someone else's contract.

Five Rules Before You Sign

The signature page, not the demo environment, is where referral-loop economics get decided. Vendors price features; your group gets paid on closures, so every clause below exists to move money out of the feature column and into the outcome column. Kill the comfortable assumption first: the premium tier is not the "safe" choice. Software modules do not call patients — outreach workflow and human follow-up close loops — which is why a hard-nosed contract on the base tier beats a generous one on the top tier anywhere below the risk-bearing threshold.

Rule 1 — Buy the floor, not the feature list. Sign only against a written guarantee that at least 70% of referrals close within 14 days, with "closed" meaning the consult note filed in the referring chart — the definition this guide has used throughout — audited quarterly against extracts pulled from your EHR, with automatic fee credits whenever the floor is missed. Two drafting details separate a real floor from a marketing sentence: fix the denominator as referrals sent, not referrals the vendor "accepted," and require the audit sample to come from your EHR extract rather than the vendor's own scorecard. A guarantee without a credit mechanism is a brochure.

Rule 2 — Compute all-in cost per life before comparing tiers. Integration fees, per-message pass-throughs, and mandated staffing lines routinely sit outside the sticker PMM; add them before any comparison. If the all-in load exceeds a $6-per-member-per-month equivalent and your panel is under 5,000 attributed lives, decline and run closure through a 0.4-FTE coordinator — about $26,000 per year in salary and benefits — working the EHR referral work queue directly. That arrangement buys the same mechanism the mid-tiers resell at a markup: phone calls and queue hygiene, not analytics.

Rule 3 — Reject status-flip metrics in writing. The endpoint flaw covered earlier has a contractual fix: require note-return rate — documents filed divided by referrals sent — as the primary KPI, and demote dashboard completed-statuses to secondary inside the reporting annex itself, not a side letter, so the demotion survives account-manager turnover. Vendors engineer toward whichever metric drives the renewal conversation; make that metric the artifact clinicians actually read.

Rule 4 — Bill on the minimum, true-up on the maximum. Attributed census moves month to month, and invoicing on peak census bills you for phantom lives. Negotiate PMM invoicing against the lowest monthly attributed census in the trailing quarter, with an annual reconciliation that trues up to actual membership — capping churn exposure at zero. Before signing, verify how the vendor defines "attributed": payer-roster dates and EHR panel counts disagree, and the gap between them is exactly where phantom lives accumulate.

Rule 5 — Pilot one specialty for 90 days before any tier upgrade. Run the base tier on a single high-volume pathway — cardiology or endocrinology, at least 30 referrals per month so the read carries enough volume — and authorize movement to the $9–$12 band only if audited closure rises at least 15 points over a pre-pilot chart-audit baseline AND at least 25% of the panel sits in downside-risk contracts that monetize the difference. Demand both conditions in writing before the upgrade meeting; a closure lift without risk-bearing revenue is a prettier dashboard, not a return.

For groups under 5,000 lives — most readers of this guide — rows one and two are the entire decision. Everything above the base tier is someone else's contract.

What to do next

Frequently Asked Questions

At what dollar value per recovered referral does the middle price band start paying for itself?

A 5,000-life group must place $1,667 of value on each recovered referral loop before the middle price band pays for itself.

What share of referrals never generate revenue under value-based contracts?

Incomplete plus untracked referrals make up 20 to 40 percent of referral volume, and of the three possible outcomes only the documented completed visit feeds quality scores, shared savings, and risk-adjusted revenue.

Why did Maine's Section 96 agencies stop taking new clients?

SeniorsPlus stopped accepting new Section 96 clients in August 2023, calling MaineCare's per-member-per-minute formula 'not workable' after roughly $600,000 in prior-year losses, and by December 2024 two of the state's three agencies had closed themselves to new referrals.

How bad was referral closure before any software existed?

According to O'Malley and Reschovsky writing in Archives of Internal Medicine in 2011, primary care physicians routinely failed to receive specialist findings for roughly half of referred patients, a closed-loop rate sitting near 50%.

Does paying for the top $9–$12 tier get my specialists' notes filed back?

No — the gap between visit-completed and note-returned is touched only by EHR integration work, which vendors bill separately as a project quoted at $5,000–$25,000.

Can I trust the 20-to-30-point closure lifts vendors advertise?

The 20-to-30-point closure lifts advertised in Luma Health and Artera case studies are uncontrolled vendor self-reports, whereas the strongest controlled signal comes from RAND's evaluations of Hawaii Pacific Health, where specialty waits collapsed from months toward days through routing, tracking, and closed communication.

Quick answers