An automation layer over Dentally that pulls intake and referral documents straight into the record, and reads clinical notes to surface and personalize the follow-up that actually books treatment, with the front desk approving every step.
The group runs on Dentally, and two things leaked around it. New-patient intake and referral letters were re-keyed by hand, slow and error-prone, with medical-history flags occasionally missed. And accepted treatment plans went unbooked while patients drifted past their recall, because chasing them was manual and untargeted. The patients were already there. The follow-up was the gap.
01 The Challenge
Two leaks, both in the unstructured edges of the workflow. At the front, every new patient's intake form, medical history, and any referral letter arrived as paper, a phone photo, or a PDF, and someone re-typed it into Dentally. It was slow, it introduced errors, and a buried medical-history flag could be missed. At the back, the real money: treatment plans that were presented and accepted but never booked, and patients overdue for recall, both sitting in the data while the front desk had no time to work them, and no way to know which were worth chasing first.
The patients were already there. The follow-up wasn't.
02 The Approach
The governing rule: the LLM reads the unstructured information and drafts the action; the practice team verifies and acts. Clinical data stays inside the practice's environment, and nothing is written to a patient's record or sent to a patient without a human approving it.
Two decisions shaped it. First, structured extraction with provenance: every field pulled from an intake or referral document carries the source it came from, so the front desk approves an extraction in seconds instead of re-typing it. Second, prioritize follow-up by value, not by list order: combine Dentally's structured data (recall due, unbooked treatment, last visit) with the model reading the associated clinical notes and treatment plans, so the team works the highest-value reactivations first, with outreach already drafted and personalized for approval.
What we deliberately did not do: no auto-writing to clinical records, no auto-sending patient messages, no clinical advice generated by the model, and no second practice system. This is built on Dentally.
03 The Build
Intake forms, ID and insurance documents, and referral letters land in S3. Claude Haiku 4.5 reads each one (vision for photos and scans), extracts the structured patient data, medical history, and plan details, and returns the source for every field. The front desk approves in a TypeScript console, and approved records write to Dentally through its API. For retention, a Python job reads Dentally for recall-due and unbooked-treatment patients; Claude Opus 4.8 reads the associated clinical notes and treatment plans to rank them by value and draft personalized outreach. Temporal runs the outreach sequence over SMS and email, send, await, follow up, so a patient is never double-contacted, and every message is approved before it goes out. A natural-language query lets the team ask things like "unbooked treatment over £500 not contacted in 60 days."
It's right-sized for a practice group and built entirely on top of Dentally rather than replacing it.
04 The Results
Intake is now captured straight into Dentally with no re-keying, removing roughly five hours a day of front-desk admin and stopping medical-history flags from slipping through. On the retention side, the team works a prioritized worklist of the highest-value lapsed and unbooked-treatment patients with outreach already drafted, and treatment-plan bookings rose about 22%. Reactivation went from a someday task to a daily, approved routine.
05 What's Next
This pairs with the attribution build: acquisition brings new patients in, and the lifecycle layer keeps them and converts the treatment they have already accepted. The same structured patient spine supports multi-location rollout and group-level reporting, and the worklist can extend to hygiene recall and treatment-plan reminders, all built once on Dentally and reused per practice.
An LLM reads supplier invoices, freight and customs bills, 3PL fee schedules, and processor statements into the cost model, so contribution margin reflects true landed cost without manual data entry, even as tariffs and freight move.
An agent that reads every cryptic retailer deduction, checks it against your order-to-cash record, and assembles the dispute, so the invalid ones stop being silently written off.
Numbers you can trust, operations that run themselves, all built into your stack and owned by you. Tell us where you want the edge, and we'll build it.
You'll hear from an engineer the same working day, not a sales sequence.
A short call, then a plan, and we start with Proof Week or a full build.
A fixed fee agreed before anything starts, for Proof Week or the full build.