HIPAA-aware AI · ADA / WCAG-compliant builds · trusted by practices nationwide HIPAA-aware · ADA-compliant
[ORACLE.001]
12 MODELS · SIGNALING
// CONFIDENCE 92%
HORIZON 90D //
Predictive analytics

Your next 90 days, before they happen

Revenue forecasts. No-show predictions. Churn alerts. Segment LTV scores. All built from your practice data — and updated every hour with confidence intervals, not just point estimates.

Your data Confidence intervals Fully explainable
Insights
12 models · 3 alerts
LIVE
Weekly bookings · forecast Next 30 days
214 ↑ 18% vs prior 30d
Actual · 60d Forecast · 30d 95% confidence
Anomaly · Tuesday bookings12% below 60d avg · investigate schedule
87%
Opportunity · Cosmetic segment24 patients ready for follow-up · LTV $3.2K
92%
Models 12 Signals 47 Confidence 92%
+18%
Revenue forecastnext 30d · 92% confidence
3 alerts flaggedauto-detected today
12 models livehourly refresh

Every drop in bookings, every lost patient — the pattern was in your data weeks earlier

What it predicts

Six forecasts, running right now against your data

Not academic. These are the predictions that actually move dollars — and the ones your gut currently makes without help.

Revenue forecasting
Next 30 · 60 · 90 days

Point forecast with a confidence band. Never just a number — always a range you can plan against.

Time-series

No-show prediction
48h before the appointment

Every scheduled visit scored on no-show risk. High-risk ones get an extra reminder — or a courtesy call.

Per-visit

Patient churn alerts
Before they slip away

Flag patients whose booking cadence has shifted. Reactivation outreach fires while there's still a relationship.

Weekly

Insurance denial prediction
Before the claim ships

Score claims on denial risk before submission. Flagged ones get a coding review; the rest go through untouched.

Pre-submit

Capacity forecasting
For smarter staffing

Predict busy vs slow days two weeks out. Shift capacity, adjust hours, add hygienists — before demand hits.

Rolling

Segment LTV scoring
Where the value actually is

Score patient segments on projected lifetime value. Marketing dollars, retention effort, referral rewards — targeted at the right cohorts.

Segment-level
The model lifecycle

Every prediction, the same five stages

Ingest → train → predict → alert → explain. Once an hour, per model, per account.

01 · INGEST

Signals in

Pull from your PMS, marketing, financial, and communication data — every hour.

AI-driven
02 · TRAIN

Models fit

Models trained on your practice's unique patterns, not industry averages.

AI-driven
03 · PREDICT

Forecasts out

Point estimates plus confidence intervals — never just a single number.

AI-driven
04 · ALERT

Anomalies flagged

When a metric drifts from its prediction, your team gets pinged — with reasoning.

AI-driven
05 · EXPLAIN

Traced to source

Every prediction linked to the signals that drove it. No black boxes.

AI-driven
Architecture

One engine, every signal

Your PMS, ads, payments, communications — all flowing into one forecast surface, all traceable back to source.

DATA IN MODELS LIVE SIGNALS OUT EXPLAINABLE CCA Insights 1H REFRESH Data sourcesPMS · ADS · PAYMENT ModelsFORECAST · SCORE OutputsALERTS · DASHBOARDS ExplainabilityFEATURE IMPORTANCE
Signals in, forecasts out — every hour Every prediction traceable to source
Guardrails

What the models will never do

Six commitments enforced in the training pipeline — not policy alone.

Never predicts individual health outcomes

Predictions cover business metrics — bookings, revenue, churn. Never diagnostic or clinical outcomes. That's medicine, not analytics.

Never uses protected class attributes

Race, religion, gender identity, disability status — excluded from every model as inputs and outputs. Enforced in the training pipeline.

Never presents predictions without confidence

No point estimates alone, ever. Every forecast comes with a range, and every alert comes with a confidence score. Uncertainty is a feature.

Never trains across practices without consent

Your data trains your models. If we want to use aggregated data for a benchmark model, we ask — opt-in only, aggregated only.

Never black-boxes a decision

Every prediction is traceable to the signals that drove it. If a patient is flagged high-risk, you see exactly why. No “the model said so.”

Never replaces clinical judgment

Predictions inform decisions — they don't make them. A no-show risk score is a nudge to send an extra reminder, not a reason to overbook the slot.

Data sources

Every signal your practice already generates

We connect to the systems your practice already uses. No new logging, no manual entry — the signals are already there.

PMS · EHRBookings · visits · cancels
Marketing analyticsGA4 · Meta Pixel
Ad platformsGoogle · Meta · TikTok
FinancialStripe · Square · payer
CommunicationsCall · SMS metadata
Web analyticsSessions · funnels
Review platformsGoogle · Yelp · Zocdoc
External signalsWeather · seasonality

Custom data source? If it has a query interface or export, we can ingest it. More data = tighter confidence intervals.

Governance

The commitments, every prediction inherits

Predictions without accountability are guesses. Every forecast, every alert, versioned and reviewable.

Trust artifacts

Predictions are only useful when they're accountable

Every model version is logged. Every alert carries feature importance. Every forecast comes with a confidence interval. Nothing runs in a black box — not the models, not the anomaly detectors, nothing.

Model versioningEvery prediction tied to a specific model version — auditable and rollback-able.
Enforced
Confidence intervals, alwaysNever a single number — always a range. Uncertainty is explicit, not hidden.
Enforced
Feature importance, per alertEvery prediction traces back to the signals that drove it. Explainable, always.
Yours
Kill-switch, per modelPause any model instantly. Predictions stop — the audit log stays.
One-click
The numbers

What we hold ourselves accountable to

Operational commitments — not marketing numbers.

0%
Avg forecast confidence
Median across active models
0
Models per practice
Median across our accounts
0
Signals per prediction
Across all model inputs
0hr
Refresh cycle
Every model, every hour
FAQ

The predictions, answered

The questions we get from practice owners, IT leads, and skeptical clinicians.

01

Where does our patient data live — do you take it?

Compliance

It stays in your PMS. We pull only the fields we need for prediction (bookings, cancels, spend, demographics — never diagnosis or treatment content) through your existing HIPAA-compliant connections. We aggregate and model — we don't stockpile.

02

What happens when the forecast is wrong?

Accuracy

Every prediction comes with a confidence interval, not a single number. If the actual falls outside the interval, we retrain and flag it as a model-quality issue. Forecasts are wrong all the time in the tails — that's what the interval is for.

03

Can we see how the model makes a prediction?

Ownership

Yes — every alert comes with feature importance. If we flag a patient at high no-show risk, you see exactly which signals drove it (previous no-shows, appointment lead time, day of week, distance, etc.). No black boxes.

04

How much historical data do we need?

Setup

Minimum 90 days for basic forecasts, 12 months for reliable ones. Less data = wider confidence intervals, not worse predictions — the model tells you what it doesn't know. If you have less, we start with simpler predictions and grow into the harder ones.

05

Which predictions are most reliable?

Accuracy

Aggregate ones (30-day revenue, weekly bookings) beat individual ones (this-patient no-show risk) — that's always true. We're transparent about which is which. If a prediction is under 70% confidence, we don't alert on it.

06

What if we're a brand-new practice with no data?

Setup

We use industry-benchmark models to bootstrap, then swap them out for your data-trained models as history accumulates — usually within 4–6 months. You'll see it happen in the model versioning log.

Want us to forecast your next quarter?

Share read access to your PMS. Within 48 hours, we'll come back with a 30-day booking forecast, top-3 no-show risks flagged in your schedule, and one anomaly worth investigating — free, no obligation.

[FORECAST.REQUEST]
12 MODELS · SIGNALING
// 92% CONFIDENCE
HORIZON 90D //
Ready to see next quarter

See what your practice will do next.

Share view-only PMS access. Within 48 hours, we'll walk you through a live 30-day forecast for your practice, the top-3 no-show risks in your current schedule, and one anomaly worth investigating — free, no commitment.

Your data Confidence intervals Fully explainable BAA signed Kill-switch anytime