Patient Conversation AI: Suite vs Point Tool vs Custom

By Jude Lee · · Comparison

Front-desk staff at an independent practice handling patient calls and check-in at a scheduling workstation

What the recent vendor consolidation actually signals

In September 2026, Healthcare IT Today reported that Hello Patient acquired Converse Health specifically to extend its AI agents across “the full patient conversation, from first call to follow-up.” The same week, Assort Health announced via PR Newswire that it now partners with more than 1,300 providers. Both items were accessed in November 2026, and both are company-reported — a press release and a vendor announcement, not audited disclosures. You may also see a valuation figure attached to this round in trade coverage; I can’t point you to a specific primary filing for it, so I’ve left the number out rather than repeat one I can’t source.

Strip away the funding news and one product truth is left: the hard part of front-office AI is not answering the phone — it’s remembering. A patient calls to reschedule, then gets a reminder text, then a balance notice, then a recall postcard. If four different systems generate those four touches, the patient experiences four strangers. Vendors are buying each other to solve that continuity problem, because it can’t be solved with better speech recognition.

The bottleneck in patient-communication AI isn’t the voice. It’s whether the agent knows what already happened.

Three ways to buy the patient conversation

Conversation suite (one vendor, end to end)

Fits when: you want one BAA, one escalation path, and one vendor accountable. Shared context across call → intake → reminder → follow-up comes free because it’s one database.

Costs you: flexibility. The scheduling logic is whatever the vendor built. If your suite is acquired or repriced, you renegotiate everything at once. Migration means replacing four workflows, not one.

Watch for: depth-vs-breadth. A suite assembled through acquisition may have one excellent module and three that are merely adequate. Demo the weakest one.

Point tools (best-of-breed per job)

Fits when: one job is clearly your bleeding wound — missed calls, or no-shows, or recall — and you’d rather fix it well than fix everything shallowly.

Costs you: the handoff. Your phone agent doesn’t know your reminder tool already texted the patient twice. Each tool needs its own BAA, its own audit review, its own escalation rules.

Watch for: stacking. Three point tools can exceed the cost of a suite while delivering a more fragmented patient experience — run the arithmetic with real quotes before assuming either direction.

The third path, and what it actually demands of you

A custom agent layer means an AI assistant (Claude, or another model your compliance posture allows) connected to your practice management system, scheduling, and messaging through MCP — the Model Context Protocol, an open standard for giving an assistant governed access to specific tools and data. You define skills — packaged, repeatable instructions for a job like “triage a reschedule request” — and the agent executes them the same way every time. The mechanics are covered in connecting Claude to your EHR via MCP.

Fits when: your workflow genuinely doesn’t match what vendors ship, or the data an agent needs lives in a system nobody integrates with.

Costs you: ownership of everything a vendor would otherwise carry. Someone has to be on call when the MCP server stops responding at 8 a.m. on a Monday. Someone has to re-test every skill when the model version changes or your PMS ships an API update — both happen without asking you. Someone has to write, store, and retain the audit logs your BAA and your privacy officer will expect, because no dashboard produces them for you. That’s ongoing engineering hours, not a one-time build, and a practice without either an internal technical owner or a contracted partner on retainer should treat this path as unavailable rather than merely harder.

Watch for: scope creep from “one skill” to “our whole front office” before the first skill has run clean for a quarter.

Where AI genuinely fits the patient conversation today

If you’re asking how to use AI in your practice without gambling, start where the task is high-volume, low-ambiguity, and reversible:

What should not run unsupervised: anything involving symptoms, triage urgency, medication questions, or clinical instructions. Route those to a clinician, and confirm with a qualified clinical professional before any patient-facing message on those topics goes out.

Partial automation is the right posture — the 30% figure isn’t a standard

There is no governing body behind the “30% rule for AI” you’ll see in vendor decks, and no study I can point you to that establishes 30% as a real threshold. The underlying instinct is sound, in my opinion: partial automation with a human reviewing the exceptions is usually the right starting posture. But treat the number itself as folklore. Measure your own containment rate — calls fully resolved without a human — after 30 days, and let that figure set your expansion pace.

Which front-office roles actually change

The roles most exposed are purely transactional: reading a script, taking a message, dialing a reminder list. The roles that get more valuable are judgment-heavy and relationship-heavy — the coordinator who knows which patients need a call rather than a text, the biller who can read a denial pattern, the office manager who can design and audit an agent’s rules. My observation from practices I’ve worked with is that the team size doesn’t change; what changes is that calls stop getting lost at 4:45 p.m. and phone time gets reassigned to eligibility, recall, and balances. If you’re weighing headcount against software, AI agents vs healthcare VAs vs a new front-desk hire lays out the tradeoffs.

There is no single best AI tool for medical professionals

General assistants like ChatGPT and Claude are large language models — generative, general-purpose AI, not medical devices and not purpose-built practice software. They’re excellent at drafting, summarizing, and reasoning over documents you give them, and they become operationally useful when connected to your systems through MCP. They are not, by default, configured to handle protected health information, and you should not paste PHI into a consumer account.

Purpose-built vendors — voice agents that integrate with major practice management platforms, or your EHR’s own AI features — win on integration and compliance scaffolding, and lose on flexibility. “Best” is per job, not per brand, which is why our job-by-job buyer’s map is organized around tasks rather than logos.

The HIPAA floor, before any demo

Model the economics with your own numbers

Ignore vendor ROI calculators. Build it yourself, starting from a baseline you actually measured:

2 weeks
Baseline to measure before you demo anything
Your own phone system and scheduling reports

  1. Count what you're losing

    Pull two weeks of call logs: abandoned calls, after-hours calls, average handle time. Pull no-show and unfilled-slot counts from your schedule.

  2. Decide what the recovered hours do

    Hours only pay off if they’re reassigned to revenue work — eligibility checks, recall, balance follow-up. Name the destination before you buy.

  3. Pilot one conversation segment

    Run 30 days on inbound scheduling only. Track containment, escalation accuracy, and complaints.

  4. Then choose your architecture

    If the vendor handled it well and your workflows are standard, expand into the suite. If one job is great and the rest are weak, stay point-tool. If the agent kept failing because it couldn’t see your data — and you have someone who can own uptime — that’s the signal to evaluate a custom MCP build.

As of 2026, this market is moving fast enough that the vendor list will look different in a year. The durable move is to own your baseline numbers, your escalation rules, and your data access design. Those survive any acquisition.

Not sure where to start?

Get a free automation audit: we map your scheduling, intake, insurance, billing, and patient communication and show you what's worth automating — before you spend a dollar.

Get a free automation audit

Related guides