Ask Your Analytics: Questions Clinics Start With

The first questions clinics ask their analytics are almost always the same four: where did last month's patients come from, which campaign is actually producing bookings, where do people drop out of the funnel, and can I trust this number. Curve is the HIPAA-compliant tracking, attribution, and analytics platform for healthcare, and Curve AI Analyst lets a marketer put those questions in plain language against their own clinic data. It works because the website analytics and campaign reporting already live inside Curve, under a signed BAA on every plan, so nothing has to be exported to a third party to get an answer.

One product, two names

You may see this feature called Sentinel in social posts and product announcements, and Curve AI Analyst in documentation and search. They are the same thing: the conversational layer over your Curve analytics and campaign reporting.

Why in-house data is the whole point

Plenty of tools promise to let you chat with your marketing data. Most of them work by shipping your data somewhere else first. You connect an account, the tool copies rows into its own warehouse, and a model reads them there.

For a healthcare marketer that copy is the problem. Website behavior tied to a service line page, a session that ends on an intake form, a booking event attached to an email address, all of it can carry protected health information or something close enough that a privacy officer will treat it as such. Moving it to a vendor with no Business Associate Agreement creates a disclosure that no amount of dashboard polish makes acceptable.

Curve does not need to move anything. The event stream, the session data, the attribution, and the connected campaign reporting are already held in Curve's US-hosted infrastructure because that is how the tracking works in the first place. Events go to Curve rather than directly to ad platforms, which is the same architectural decision that makes server-side tracking safer than a client-side pixel. The chat layer reads what is already there.

The questions clinics actually start with

Across practices, telehealth companies, med spas, and multi-location groups, the opening questions cluster tightly. Here is what people ask in the first week, and what makes each one answerable.

Where last month's patients came from

The most common first question, and usually the one that exposes how bad the previous setup was. A good answer breaks traffic and conversions by channel and by UTM source, medium, and campaign, and separates paid from organic from direct.

The reason this question is hard without a unified layer is that the answer lives in four places: Google Ads reports one number, Meta reports another, Google Analytics reports a third, and the CRM reports a fourth. Asking one system that holds all of it gets you a single answer instead of four you then have to reconcile by hand.

Which campaigns produce bookings, not just leads

This is the question that separates marketers who are optimizing from marketers who are reporting. Lead volume is cheap to generate. Booked and attended appointments are the thing the practice is paid for.

Answerable only if downstream outcomes are getting back into the measurement layer, through an incoming webhook from the CRM or scheduling tool, or through an offline conversion upload. If your booked appointments never come back from the practice management system, no chat interface can conjure them.

Where people drop out between the ad and the booking

The funnel question. Landing page to form start, form start to submit, submit to consult booked, consult to appointment attended. Each step has a number and the biggest gap is your next project.

Curve holds page-level behavior alongside conversion events, so the drop-off question and the page performance question are the same dataset. That matters, because the answer to "where do they leave" is usually a specific page and a specific step, not a general observation about the funnel.

Why Google Ads shows more conversions than the dashboard

Very common, and it is a real question rather than a data quality complaint. Ad platforms count with their own attribution windows, include view-through conversions, apply modelling to fill gaps, and are self-reporting on their own performance. Your analytics counts sessions and events it observed directly.

Both numbers can be correct. Understanding the gap is a whole topic on its own, covered in our piece on setting up conversion tracking across Google, Meta, and Microsoft.

Which pages people actually read before converting

Entry pages, exit pages, and the pages that show up in converting paths. Clinics are usually surprised here. The service page everyone argues about in meetings is often not the page that precedes bookings. Pricing pages, insurance pages, and provider bio pages routinely outperform the pages marketing spent the most time on.

Whether the tracking is working right now

The trust question, and the one experienced marketers ask first. Are events arriving, are they mapped, are destinations receiving them, when did each connector last sync. A number you cannot date is a number you cannot use.

How to ask a question that gets a useful answer

Natural language does not mean vague language. The questions that return something worth acting on share a few properties.

Name the time period. "Last month" and "the last 30 days" are different windows and will return different numbers. Say which one you mean, and be consistent when you compare.

Name the conversion. A clinic tracks several: form submitted, call started, consult booked, appointment attended. "Conversions" without a definition is ambiguous, and the ambiguity is exactly where reporting arguments come from.

Ask for a comparison, not just a number. "How many bookings last month" is a fact. "How did bookings by channel compare to the month before" is a finding. The second one leads somewhere.

Follow the answer down. The value of a conversational interface is the second and third question, not the first. Channel, then campaign, then landing page, then device. Each step narrows the problem.

Ask what the number excludes. Good analysis is mostly about knowing what is missing. Sessions with no consent for analytics, conversions that happened on a booking tool you have not connected, phone calls that never touched the website. Ask.

What it will not do, and why that is deliberate

Being clear about the boundary is more useful than overselling the capability.

Curve AI Analyst answers questions about the analytics and campaign reporting data Curve holds. It is a reading and reasoning layer over your measurement, not a control panel. It does not reconfigure your account for you, and that separation is intentional: the configuration surface of a HIPAA-compliant tracking setup, particularly per-destination field mapping, is exactly where a careless change becomes a disclosure. Changes to what data leaves your platform stay a deliberate, human, auditable act.

It also cannot answer a question about data that was never collected. If click IDs were not captured at landing, attribution for those sessions is gone and no interface recovers it. If your booking tool sits on a separate domain with no bridge token, the chain is broken before the chat layer ever sees it. If a service line has no distinct event, you cannot ask about it by name.

And it is not a compliance opinion. It can tell you what your tracking is doing. It cannot tell you whether your privacy notice is adequate or whether a given disclosure was permitted. Those are questions for your privacy officer and counsel.

How Curve AI Analyst works

Curve is HIPAA-compliant ad tracking, attribution, and analytics for healthcare. The tracking script installs on the clinic site in place of the Meta Pixel and Google tag. Events go to Curve's US-hosted infrastructure, and only explicitly mapped fields forward to a given ad platform, with identifiers SHA-256 hashed to each platform's conversion API requirements. Nothing forwards by default.

That architecture produces a side effect worth naming. Because Curve sits in the middle rather than at the edge, it ends up holding a fuller picture of the clinic's marketing than any single connected platform does: first-party website behavior, source and UTM performance, page and entry and exit performance, device and geography, goals and funnels where they are configured, plus campaign reporting from connected ad and search and CRM sources.

Curve AI Analyst reads that. You ask in plain language, it answers from your own data, and the data does not leave. The practical consequences:

• No export step. There is no copy of your patient-adjacent behavioral data sitting in a third-party analytics vendor's warehouse for the sake of a chat feature.
• One dataset, not four. Paid media, organic search, website behavior, and CRM outcomes are answered from one place, so the answer does not require reconciliation first.
• Follow-up questions keep context. The narrowing sequence, channel to campaign to page, is the part that produces decisions.
• Under the same BAA. Curve includes a signed BAA on every plan, and the analytics layer is inside that scope rather than a bolt-on with separate terms.

The quality ceiling is your instrumentation, not the interface. Clinics that get the most out of it are the ones that captured click IDs at landing, used compliant lead routing into the CRM, wired outcomes back through webhooks or offline uploads, and gave their events consistent names. That work pays off twice: once in accurate reporting, and again in questions you can now ask conversationally.

A first-week question sequence

If you want a starting script, run this order. It moves from posture to performance to action.

1. Is data arriving and how fresh is it? Establish trust before you interpret anything.
2. What did traffic and conversions look like last month by channel? The baseline.
3. How does that compare to the prior month? Direction matters more than level.
4. Which campaigns drove the conversions in the strongest channel? Narrow once.
5. Which landing pages did those campaigns use, and how did they perform? Narrow twice.
6. Where in the funnel did those visitors stop? Find the leak.
7. What is different about the worst performing campaign? The question that turns into work.

Frequently asked questions

Is Sentinel the same as Curve AI Analyst?

Yes. Sentinel is the public and social name, Curve AI Analyst is the descriptive name used in documentation and search. One product, one feature, two labels.

Does my clinic data leave Curve to answer a question?

No. The analytics and campaign reporting are already held inside Curve because that is how the tracking is built, and the conversational layer reads that data in place. There is no export to a separate analytics vendor as a precondition of asking a question.

Can it change my tracking setup or my destinations?

No, and that is deliberate. It answers questions about your data. Configuration changes, especially per-destination field mapping that governs what leaves your platform, stay a deliberate human action so they remain reviewable.

What if the answer contradicts what Google Ads is showing?

That is normal and usually explainable rather than a fault. Ad platforms use their own attribution windows, count view-through conversions, apply modelling, and report on their own performance. Your analytics counts what it directly observed. Treat the gap as information about methodology rather than as an error to eliminate.

Do we need every connector wired up before this is useful?

No, but coverage sets the ceiling. Website behavior and conversion questions work as soon as the script is installed and events are mapped. Spend and campaign questions need the reporting connectors. Booked and attended outcome questions need the CRM or scheduling system returning results through a webhook or offline upload.

Is it safe to ask about a specific patient?

Do not. Curve's analytics layer is built for aggregate marketing measurement, not individual patient lookup, and individual-level clinical questions belong in your BAA-covered clinical systems where access controls and audit trails are designed for that purpose.

What is the single biggest thing that limits answer quality?

Missing outcome data. Most clinics can answer questions about clicks and forms and cannot answer questions about attended appointments, because the practice management system never sends outcomes back. Fix that first and every later question gets better.

Where to start

Start by writing down the five questions your team argues about most in reporting meetings. Those are the ones worth instrumenting for, and they are almost always about outcomes rather than clicks.

Curve gives healthcare marketers a compliant measurement layer that holds website analytics, attribution, and campaign reporting in one place, with per-destination field mapping, hashed identifiers, neutral event aliases, bridge-token attribution, and a signed BAA on every plan. Curve AI Analyst sits on top of that so those five questions get answered in plain language without your data leaving. Run our free compliance scanner against your clinic site to see what your current tracking is exposing, or visit curvecompliance.com to talk through what your data can already answer.

Reviewed August 2026. Product capabilities, ad platform reporting behavior, and healthcare advertising policies change frequently. Verify current requirements before implementation.