How to Diagnose a Telehealth Consult Funnel with AI: Without Sending PHI Out

A telehealth consult funnel leaks in about five predictable places, and until recently diagnosing which one meant building an export you were not allowed to build. The stages are click to landing page, landing page to form start, form start to form complete, form complete to scheduled consult, and scheduled to attended. Each has its own failure signature, and each is now a question you can ask Curve AI Analyst in a sentence, because the data spanning all five already lives inside one HIPAA compliant boundary.

What follows is a diagnostic you can run this week, in order, with the question to ask at each stage and what a bad answer looks like.

The most measurement dependent business, with the least measurement

Telehealth is the sharpest version of the healthcare marketing squeeze. The entire business is digital acquisition, so the funnel is the unit economics. There is no walk in traffic to hide behind and no referral network absorbing a bad month.

It is also the most restricted. The landing pages name conditions, because they have to for search. The intake forms collect exactly the information that makes an export regulated. The booking step usually happens on a separate vendor's domain. And the outcome that matters, whether the patient showed up, lives in a system marketing has no login to.

So the standard diagnostic move, exporting the funnel and staring at it, is both the obvious thing to do and the thing you cannot do. That choice, and what sits between the two bad options, is covered in leak or go dark, and the third option for telehealth tracking.

The five stages and what breaks at each

Stage one: ad click to landing page

Losses here are invisible in most accounts, because a click that never becomes a session does not appear anywhere except as a gap between two systems that were never compared.

What breaks: click identifiers stripped by a redirect chain or link shortener, so the session arrives unattributed. Slow pages on mobile connections. Message mismatch, where the ad promises same day appointments and the page opens with a form. Geographic waste from states the practice is not licensed in, which in telehealth is a hard wall rather than a soft inefficiency.

Ask: how many sessions arrived from paid last month compared with the clicks the platforms reported. Which regions produce paid sessions, and do they match the states we are licensed in. Which paid landing pages bounce worst.

Stage two: landing page to form start

This is the persuasion stage, and it is usually the largest single loss in a telehealth funnel.

What breaks: the page does not answer the three questions every telehealth visitor has, which are whether you treat their condition, whether you can see them in their state, and what it costs or whether insurance applies. Missing trust signals. The form starting below the fold on mobile. A page written for a search engine rather than an anxious person comparing three tabs.

Ask: what share of visitors to the consult page start the form. Which entry pages produce the best rate through to form start. Is the rate different on mobile than desktop. Which channel sends traffic that lands and leaves immediately.

Stage three: form start to form complete

The mechanical stage, and the most fixable one, because every loss here is a specific field on a specific step.

What breaks: too many fields before any commitment is earned. Insurance carrier and member ID requested up front rather than after the appointment is held. A state of residence dropdown that quietly disqualifies people with no explanation. Date of birth pickers that are hostile on phones. Photo identification or insurance card uploads, a large drop point on mobile.

Ask: how many people started the intake form last month and how many finished. Which step loses the most people. Did completion change after we added the insurance question. Is completion worse on mobile, and by how much. Which sources produce form starts that never finish.

If the form is a Curve form, the funnel can be generated with it, so step level drop off is measured directly rather than inferred from page transitions.

Stage four: form complete to scheduled consult

The handoff stage, and the place where most telehealth attribution dies quietly.

The visitor finishes your form and moves to a booking tool, often on a different domain and sometimes a different vendor entirely, such as IntakeQ, Calendly, or Jane App. In a default setup the campaign context does not survive that trip, and everything after it is attributed to direct traffic. That is why so many telehealth accounts show an implausible volume of direct conversions.

What breaks: attribution lost at the click out. Scheduling availability that shows nothing for four days, which is a conversion killer nobody in marketing ever sees. Payment or eligibility checks introduced at the last step. A confirmation flow that requires an account.

The fix is structural rather than analytical. Bridge tokens carry attribution across the handoff, so when the booking system reports the completed appointment back through an incoming webhook, the event matches to the original session and campaign. Cross domain tracking covers the case where the next step is another property you own.

Ask: how many completed forms became scheduled appointments last month. Which campaigns produced scheduled appointments rather than form completions. Has that gap widened.

Stage five: scheduled to attended

The stage that decides whether the whole account was profitable, and the one almost no marketing team measures.

No show rates in telehealth are not a rounding error and are not evenly distributed. Different channels, creative promises, and appointment lead times produce very different attendance. A campaign with the best cost per booking can easily have the worst cost per attended consult, and if you only measure bookings you will scale it.

What breaks: the outcome exists only in a scheduling or clinical system. Reminder flows that fail. Long lead times between booking and appointment. Eligibility problems discovered after booking.

Getting this stage into your data is a setup task with two paths. A webhook from the scheduling system can report the attended outcome in real time, and Curve supports treating the earlier step as a milestone for journey context while the later event is the final conversion, so scheduling and attendance are not collapsed into one number. Otherwise, offline conversion uploads backfill outcomes from a CSV export, matched on bridge token, click ID, or email.

Ask: how many scheduled consults were attended last month, by channel. Which campaign has the best cost per attended consult rather than per booking. Does attendance differ by how far in advance the appointment was booked.

Run the diagnostic in this order

Order matters, because a loss at an early stage makes every later percentage look strange.

1. Check data health first. Ask whether every reporting connector synced this week. A stale connector explains more mysterious drops than any funnel problem does.
2. Size each stage. Ask for volume at each of the five steps for last month and the month before. Look for the biggest absolute loss, not the worst rate: a stage that loses many people from a large base beats a terrible rate on tiny volume.
3. Split the worst stage by device. Run every question again with "on mobile" appended. Telehealth traffic skews to phones more than teams expect, and the failures are phone specific: uploads, date pickers, long dropdowns, forms below the fold. A funnel that looks acceptable in aggregate is often two funnels averaged into a number that describes neither.
4. Split it again by channel. A stage that performs badly for one source and fine for the rest is a targeting or message problem, not a funnel problem, and the fix is in the ad account.
5. Compare with a period before the last change. If the drop coincides with a release, a new form field, or a scheduling policy change, you have your cause.
6. Go watch five sessions. Once you know the page and the step, session recordings and heatmaps tell you the why in about ten minutes. The number narrows five hundred pages to two.

Why asking beats exporting here specifically

This funnel spans four systems: the ad platform, the website, the booking tool, and whatever holds attendance. The export approach means four exports and a join, and the join key is exactly the thing you should not be emailing around. The compliance problem and the analytical problem share a root.

Inside Curve those systems already share a thread, because the events, the bridged handoff, the webhook outcomes, and the campaign reporting all landed in the same place. The question is answered where the data lives, so there is no moment where a marketer decides what is safe to move. That distinction is the argument in why you cannot paste a patient funnel into ChatGPT, and the architecture that makes the alternative possible is in how compliant tracking makes AI answers possible.

One caveat the assistant raises on its own: Curve sees site activity in real time while ad platforms report on a lag of hours or days and revise as attribution settles. When a comparison crosses those two clocks, the answer says so rather than handing you a confident number about yesterday.

What to fix, by where the leak is

• Stage one: audit the redirect chain for click identifier loss, tighten targeting to licensed states, fix mobile page speed on paid landing pages.
• Stage two: answer condition, state, and cost above the fold. Match the page to the ad promise.
• Stage three: cut fields to the minimum required to hold an appointment, and move insurance and uploads after the booking is secured.
• Stage four: implement bridge tokens and a return webhook, then treat scheduling availability as a conversion factor rather than an operations detail.
• Stage five: get attendance into the data, then optimize on cost per attended consult instead of cost per booking.

Stages four and five are setup work rather than analysis, and they pay the most, because they convert your reporting from measuring intent to measuring outcomes.

Setup, so the answers are real

1. Curve script on every domain in the journey, with cross domain tracking enabled and the relevant domains linked.
2. Bridge tokens created on the site and returned by the booking system in its final webhook.
3. Events named for the business: consult requested, intake completed, appointment scheduled, consult attended.
4. Those events mapped, and goals and funnels defined for the path you want to measure, since nothing can drop through a funnel that does not exist.
5. Reporting connectors authorized and syncing, which is a separate setup from conversion forwarding.
6. Attendance arriving through a webhook or periodic offline upload.

The setup detail lives in HIPAA compliant conversion tracking setup, and the reason this order is not optional is laid out in track, then attribute, then ask.

What this is not

Worth stating plainly, because the phrase invites the wrong reading.

• It is not clinical. It has no view of diagnoses, treatment, or outcomes of care, and no business having one.
• It is not a per patient lookup. This is aggregate marketing analytics, and identifier shaped values are redacted before anything reaches the model layer. Do not type patient details into a question.
• It does not write SQL or run free form queries against a database.
• It does not create your goals or funnels, or change your tracking configuration. You configure, it reads.
• It does not fix the form. It tells you which step to fix.

Sentinel, the name you may have seen on our social channels, is the same product described here.

Frequently asked questions

Can I see where people drop out of my intake form without sending PHI anywhere?

Yes. Step level drop off is aggregate behavioral data collected by Curve into HIPAA compliant infrastructure under the BAA included with every plan, and answering a question about it does not move that data anywhere. You are asking how many people left at step three, not who they were.

How do I track a booking that happens on a different vendor's site?

Bridge tokens. A token created during the website session travels through the handoff, and when the booking system reports the completed appointment back through an incoming webhook, Curve matches the outcome to the original session, campaign, and click. Webhooks can also match on click ID or email where a token is not available.

Can I measure attended consults and not just booked ones?

Yes, provided attendance gets into Curve. Real time is a webhook from the scheduling system, and the earlier step can be recorded as a milestone for journey context while attendance is the final conversion. Where real time is not available, offline conversion uploads backfill attendance from a CSV export, matched on bridge token, click ID, or email.

Why does my telehealth account show so much direct traffic?

Usually the booking handoff, occasionally a redirect stripping click identifiers. When campaign context does not survive the move to another domain, the conversion has nowhere to be attributed and lands in direct. If direct looks implausibly large, start at stage four.

Do I need session recordings to run this diagnostic?

No, but they shorten it. The analytics tell you which stage, page, and device. Recordings and heatmaps tell you why, and they are built with input masking and the ability to block sensitive regions from capture.

How often should this run?

The full pass monthly, the data health check weekly. That cadence is the point: once asking costs a sentence, you check the funnel on a Wednesday because you were curious, and you catch the broken upload step three weeks earlier than the monthly report would have.

What if my funnel has more or fewer than five stages?

Most do. Some add eligibility screening, insurance verification, or a payment step; some skip the form in favor of direct booking. The method is the same: define the stages you actually have as a funnel, size the loss at each, then segment the worst one by device and channel before changing anything.

The funnel you could never see

Telehealth operators have been running the most measurable business model in healthcare with the least measurement available to them. The stages were always there and the losses were always happening. What was missing was a legal way to look at all five at once, because looking meant exporting and exporting meant a disclosure.

Curve is the first platform to let you talk to your analytics, your marketing, and your campaign reporting in a HIPAA compliant way, and the best place for it because the click, the session, the form, the bridged booking, and the returned outcome all landed in the same system. You ask. It answers. PHI does not leave. For the full picture of what it reads, start with what Curve AI Analyst is.

Bring the stage you suspect is broken. See how it works at curvecompliance.com.