Introduction

AI is not going to fix your marketing numbers.

AI is going to move faster in whichever direction they were already pointing.

Most healthcare marketing dashboards track forty things. A multi-location group is actually run by five: booked appointments, cost per appointment, show rate, capacity, and revenue per patient. Everyone is racing to put AI on top of everything measured. But are the few that truly matter clean yet?

If the numbers are clean, that’s a competitive advantage. If they aren’t, you’re about to misallocate budget at a speed you couldn’t have managed by hand.

1. The Five Numbers That Actually Run the Group

Forty KPIs on a dashboard is a signal that no one has decided what matters. In a multi-location healthcare platform, five numbers do the actual work of running the business. Everything else is context.

  1. 01
    New patient appointments booked.

    Not leads. Not form fills. Not calls. Appointments on the calendar, with returning patients pulled out. If your dashboard can’t separate the two, it isn’t tracking what you think it’s tracking.

  2. 02
    Cost per new patient appointment.

    Not CPL. Not CPA on a form fill. The real number tied to a real patient on the schedule. It is almost never the number on the paid media report.

  3. 03
    Show rate.

    A patient who books and does not show is a $0 patient. Win the booking, lose the revenue. Marketing gets credit for the appointment. The P&L doesn’t.

  4. 04
    Capacity by location.

    The ceiling on what marketing can actually convert. You can spend perfectly and still lose the patient because the next new-patient slot is eight weeks out.

  5. 05
    Revenue per new patient.

    First visit revenue, average first-year revenue, and lifetime value. Is the patient you bought worth what you paid.

Get those five clean. Then put AI on top of them with the underlying data that supports how those measures are shaped. This is the same failure mode I wrote about in the Attribution issue — the dashboard is a screen, the definition under it is the actual system.

2. Example: A Dental Group That Was Winning on the Dashboard

I worked with a dental group this year that was spending about $85 per form fill and reporting a clean, healthy paid search program to the board. Number on the slide, up-and-to-the-right trend, everyone nodding.

Once we ran the numbers back through the practice management system, 28% of those forms became kept appointments. Real cost per new patient was north of $300. The dashboard was working. The definition under the dashboard was wrong.

“An AI layer on that same data would have raised bids on the campaigns generating the most forms. Which were the campaigns generating the fewest patients.”

The model would have been correct. The outcome would have been worse.

That is the shape of the whole risk. AI is faithful to whatever objective you point it at. If the objective is form fills, you get more form fills — whether or not they become patients. If the objective is kept appointments tied to revenue by service line, you get more revenue. Same tools, opposite outcomes.

3. What AI Is Actually Good For Here

Once the core KPIs are clean, AI really does accelerate three specific areas — and none of them require a new AI tool. All of them require the KPI numbers to be clean first.

  1. 01
    Reconciliation across systems.

    Matching a form fill to a call to an appointment to a kept visit to a revenue event, across systems that were never designed to talk. This is really hard work and it is where the money is. Ninety percent of healthcare attribution problems come down to identity resolution, and this is the exact task where a model earns its keep.

  2. 02
    Pattern detection at the location level.

    Which locations are converting patients efficiently, which are leaking at the front desk, which are capacity-constrained. A human looks at forty locations and sees noise. A model looks at forty locations and sees the few problems worth fixing for the biggest impact.

  3. 03
    Bidding against lifetime value.

    Once you know what a new patient is actually worth by service line and by location, you can bid to that number instead of to a form fill. This is where the 20–40% CAC improvements come from. Not from a new tool. From bidding to the right target.

Notice what’s not on this list: no new dashboards, no new agents, no new SaaS. Each of these is a use of AI on top of infrastructure you already have — but only if the KPI definitions underneath it are the right ones.

4. The Simple Test

Pull your last board deck. Find the marketing slide.

If the number the board is being shown is cost per lead, cost per form fill, or a blended CAC that mixes new and returning patients — you don’t have an AI problem. You have a definition problem. AI will make it worse before it makes it better.

If the number is cost per kept appointment, tied to revenue per new patient by service line, you’re ready to put AI on top. Most groups aren’t there yet.

“What gets measured against the wrong definition gets aggressively misallocated. The dashboard is not the problem. The definition under the dashboard is.”

AI does not fix broken numbers. It scales them. The advantage goes to whoever cleaned up their core KPI measurement before their competitors did.

If your board is being shown cost per lead instead of cost per kept appointment, we can help you fix the definition before you buy the model. That’s the sequence. Numbers first. AI on top.

— Matt