What is AI Workforce Management for Telehealth?

TL;DR: AI workforce management for telehealth means forecasting demand and building schedules at the level where your constraints actually live: state, payer, service line, hour. We work with virtual care teams running hundreds of clinicians, and the pattern repeats: it is almost never a capacity problem. It is a matching problem. Paid clinician hours sit idle in one state while patients wait three weeks in another, and nobody can see it until the quarter closes. Closing that gap is what we built Untether Labs to do.

What makes workforce management different in virtual care?

Your unit of supply is not a clinician. It is a clinician, in a state, on a payer panel, for a service type, at an hour. A psychiatric NP licensed in Texas, Florida, and New York, enrolled with two payers in Texas but one in Florida, at panel capacity for follow-ups but open for intakes on Tuesday and Thursday evenings, is not one unit of capacity. She is dozens of cells in a matrix, and most of them are empty at any given moment.

That is why "do we have enough capacity?" has no answer at the network level. The real question is whether you have intake capacity in that state on that panel at the hours patients actually book, which in consumer telehealth means evenings and Sunday nights, exactly when clinicians least want to work. And the stakes are set by your cost structure: the U.S. spends $4.9 trillion a year on healthcare, and in a virtual care company nearly all delivery cost is one line, regulated clinical labor. There is no facility absorbing the slack. Every paid hour that does not end in a completed visit comes straight out of gross margin.

The seven-question test

When we start working with a new team, we ask seven questions. Each should be answerable from a screen, not a meeting:

• How many bookable intake hours do you have next week, by state and payer, against forecast?
• What is your median time-to-first-appointment in your five most strategic states, and which payer access standard are you closest to breaching?
• What share of contracted clinical hours actually got published as bookable availability last month?
• What is completed utilization, after no-shows and late cancels, by service line?
• How many clinicians are at each stage of credentialing, and how many billable hours come online in each of the next three months?
• If demand doubled in one state next month, what flexes first, and at what cost per additional visit?
• How many appointments this week are booked against a clinician who cannot bill for them?

Nobody passes all seven on day one. That is not a criticism; it is the gap between the systems you have (an HRIS, an EHR, a credentialing tracker, spreadsheets) and the question they were never designed to answer together. The teams that close that gap run differently within a quarter.

The three loops every telehealth ops team runs

The quarterly capacity plan. Forecast demand by state, payer, and service line two to four quarters out, then commit hiring, cross-licensing, payer enrollment, and flex depth against it. The loop exists because of lead time: a hire takes a quarter, a license and an enrollment can each take another. Capacity decided today is revenue in six months, so your marketing calendar and payer go-lives belong in the same plan. In consumer telehealth a capacity shortfall surfaces first as wasted acquisition spend: every extra day between signup and first available appointment bleeds conversion you already paid for.

The schedule build, two to six weeks out. Forecasts become published calendars. The hard part is the intake-to-follow-up mix: every intake commits a stream of future follow-ups, so booking intakes greedily in a slack month buries you in follow-up demand sixty days later, mid-care, where continuity breaks cost outcomes and payer quality scores. Licensure and panel status are hard rules here, and clinicians self-schedule inside guardrails. The best retention we see comes from clinicians choosing within rules the system enforces, not from top-down assignment.

Day-of. Forecasts miss daily in small ways. Late cancels release to automated waitlist backfill. Predicted gaps trigger targeted open-shift offers before the gap arrives. Slack hours move to documentation, outreach, or async work instead of evaporating. The instrument is a morning flash by state, not one heroic ops person and twelve browser tabs.

Where AI actually moves the numbers

We are specific about this because the category is full of vague claims. Four mechanisms do the work:

• Cell-level forecasting. Demand by state, payer, service type, and hour-of-week, with your marketing calendar and contract go-lives as inputs. A spreadsheet can forecast a network total. It cannot keep three hundred state-payer cells current. A model can.
• Constraint-native scheduling. Licenses, panels, panel caps, modalities, preferences, and access standards solved as one problem, instead of five spreadsheets reconciled by the one person who knows where everything is.
• Targeted incentives. Flat premiums overpay for shifts that would have filled anyway. Predicting which specific shifts go short means incentive dollars land only where they change the outcome.
• Reconciliation that stops existing. Swap approvals, license checks at booking, payroll-to-schedule matching. This is where our customers' admin savings come from, and where the payroll errors clinicians remember forever stop happening.

The mechanism is always the same: shrink the gap between hours you pay for and hours that end in a completed visit, and see it weekly instead of at quarter close.

Why this is the 2026 problem

Payer mix is tipping toward fee-for-service and value-based contracts, where idle capacity is pure cost and attributed volume arrives on the payer's schedule. Consolidation keeps merging clinician networks, and the admin load scales with the headcount you absorb, right when the deal model promised leverage. Multi-state expansion keeps widening the matrix. And the regulatory floor keeps moving: Medicare telehealth flexibilities live on short-term extensions and virtual controlled-substance prescribing rules keep shifting, so Medicare-heavy or psychiatry-heavy capacity plans have to be scenario plans. The prize has not shrunk; McKinsey put up to $250 billion of U.S. healthcare spend in reach of virtual care. It goes to the operators whose workforce scales without the admin scaling with it.

This is what we build

Untether Labs is cell-level forecasting, licensure-aware scheduling, and day-of staffing in one system, integrated with the EHR and HRIS you already run. Take the seven-question test with your team this week. For every question that takes a meeting, we should talk.