AI Phone Agent Pricing: Compare Minutes, Calls and Completed Requests

The ledger below is an original operating-cost worksheet with explicitly fictional quotes. Linear Health's own published starting price appears separately as a vendor statement; the fictional rates are not market benchmarks.

Why phone agent quotes do not line up

A useful starting question is whether the vendor's billable event matches your operational event. Your team thinks in terms of a patient who called about a referral. The contract may think in terms of thirty-second increments of connected audio, or a session that ended in a transfer, or a request that reached a defined end state.

The Linear Health Operational AI Evaluation Kit lists this as an evidence check under commercial terms: model pricing at current, expected and high-volume usage, including implementation and support costs, and treat it as a red flag when the buyer cannot model growth before signing. That framing is the basis for the method below.

The AHRQ Workflow Assessment for Health IT Toolkit makes a related point from the implementation side: health IT changes both clinical and administrative workflow, and you need to understand that impact to plan well (AHRQ). Pricing is one of the administrative workflows that changes, because the invoice now depends on how calls are routed.

Step 1: Map every billable event in the proposed contract

Before you compare numbers, list every event a quote could charge for. Use the ledger below as the starting checklist and ask each vendor to confirm, in writing, how each row is treated. The right answer for a row might be "included", "billed as a minute", "billed as a call", "not billed" or "not applicable". The wrong answer is silence.

Fill the "your monthly count" column from your phone system reports for a typical month. If you do not have transfer or callback counts, that is itself a finding to record, because you cannot model cost for an event you cannot count.

Step 2: Separate connected minutes, calls, transfers and callbacks

Check four event definitions before applying any rate.

Connected minutes. With a fixed rate and no change in minimums or rounding, more billable minutes increase usage charges. A scheduling call that needs insurance details, a second appointment and a confirmation read-back will cost more than a simple hours check. Ask about the rounding increment, whether hold time counts and whether the clock keeps running after a transfer.

Calls. Per-call pricing flattens that variance, but you need to know what counts as a call. Does an abandoned call in the queue count? Does a two-second misdial count? Does a call that comes back after a dropped connection count twice?

Transfers. Ask which call legs continue after a transfer, which party supplies them and how each is charged. Do not assume the AI's conversation ending also ends every telephony charge. Record whether the quote counts a transferred request as completed or leaves it open for staff resolution.

Callbacks. Include outbound callbacks explicitly. Confirm whether attempts, connections or completions are billed, and whether callbacks sit under the same rate as inbound.

Step 3: Normalize fictional quotes to the same request mix

The matched-scenario worksheet is the core artifact. Build one synthetic month from your real call mix, then run every quote through it. The example below is entirely illustrative. The vendors are fictional, the rates are invented for arithmetic only and nothing here reflects any real vendor's price or any Linear customer's experience.

Synthetic month (labelled example)

• 2,000 inbound calls answered in total
• 1,500 not transferred, averaging 4 billable minutes each
• 500 transferred, averaging 2 billable minutes before handoff; the fictional quote excludes later staff-leg minutes
• 300 unanswered or abandoned
• 400 outbound callbacks, 250 connected at 3 minutes each
• 1,200 distinct requests reaching the agreed completed state across these contacts; calls and requests are different units

Fictional Quote X: per connected minute at $0.20, callbacks billed on connection only, no setup fee, 1,000-minute monthly minimum.

• Non-transferred inbound: 1,500 calls x 4 min x $0.20 = $1,200
• Transferred inbound: 500 calls x 2 min x $0.20 = $200
• Callbacks: 250 x 3 min x $0.20 = $150
• Unanswered: $0
• Monthly total: $1,550

Fictional Quote Y: per answered call at $0.90, transfers counted as calls, callback attempts counted as calls, $3,000 one-time setup.

• Inbound answered: 2,000 x $0.90 = $1,800
• Callback attempts: 400 x $0.90 = $360
• Monthly total: $2,160, plus $3,000 setup amortized over 12 months = $250 a month
• First-year monthly equivalent: $2,410

Fictional Quote Z: per completed request at $1.50, $500 a month support, no setup.

• Completed requests: 1,200 x $1.50 = $1,800
• Support: $500
• Monthly total: $2,300

Under these invented rules, first-year vendor totals are $18,600 for X, $28,920 for Y including setup, and $27,600 for Z. These exclude any unlisted cost; they are not a recommendation.

A changed mix does not necessarily change the ranking. Hold total answered inbound calls at 2,000 and increase transfers from 500 to 1,000, retaining the assumed 4-minute non-transfer and 2-minute transfer durations. X becomes (1,000 x 4 + 1,000 x 2 + 750) x $0.20 = $1,350 monthly. Y remains $2,160 before setup because its billable call count is unchanged. Z changes only if the billable completed-request count changes. Every scenario must state which inputs stay fixed.

For comparison, Linear Health describes its own model this way: From $2,000 a month. Flat monthly fee per workflow, sized to your volume. You pay nothing monthly until the workflow is live and processing real work. That is Linear's commercial statement, not an independent benchmark, and the worksheet should test it the same way: ask how "sized to your volume" is defined and what happens if volume changes. If you want to run your own activity through that model, you can book a demo and bring the ledger.

Step 4: Add setup, support and minimum commitments transparently

Usage lines are only part of the invoice. Add these to every scenario as explicit rows:

• Setup or implementation. One-time or recurring, and what it covers. Ask whether telephony, EHR connection and script configuration are separate lines.
• Support tier. What the base tier includes and the cost of the tier your team would need.
• Minimum commitment. Volume or dollar floor, and whether unused volume rolls over.
• Overage. The rate above the commitment and how it is calculated.
• Go-live timing. When billing starts relative to live operation. A quote that bills from signature and a quote that bills from go-live are not comparable without this row.
• Internal labor. Your staff time for configuration, testing and exception handling. Keep this separate from vendor fees so capacity is not counted as cash.

The evaluation kit's business-case check asks for recurring fees, usage, implementation effort, internal labor and downside scenarios in one model. The worksheet above is one way to meet that check.

Step 5: Reconcile the first invoice against the activity ledger

As editorial guidance, treat the first live invoice as a test of the quote rather than a bill to pay and file. Export the activity report for the billing period, map every line to a ledger row, and look for three things: events billed that you did not expect, events you expected that were not billed, and unexplained count differences. Reconcile time zones, billing-period boundaries, per-event rounding, transfer legs, credits and minimums before deciding a difference is an error. Record the differences, send them to the vendor and keep the reconciliation as part of the decision record. If the invoice cannot be mapped to the ledger, escalate before the second billing cycle.

This step is also where the pricing question meets benefit modeling. The voice AI ROI article covers how to model returns; this article only covers what you will be billed. Keep the two separate so that a favorable return model does not hide an unfavorable billing structure.

How this fits a broader vendor review

Pricing is one of seven decision areas in the Operational AI Evaluation Kit, alongside workflow coverage, integration, oversight, data responsibility, implementation and outcomes. The kit scores evidence from claimed to demonstrated to verified and treats privacy, security and safety issues as gates that a good price cannot offset. If you are comparing an AI phone agent against a human answering service rather than against another AI vendor, the category comparison covers that decision. Both sit under the patient engagement platform pillar.

If you have already built your ledger and want to see how a flat per-workflow model reconciles against it, book a demo and bring the completed worksheet.

Frequently asked questions

Sources

Sources accessed October 29, 2026.

• Linear Health, Operational AI Evaluation Kit.
• AHRQ Digital Healthcare Research, Workflow Assessment for Health IT Toolkit.
• Linear Health, Patient engagement platform.