Skip to content

Conversation Factory (DME)

Source: 2/18 Meeting (2026-02-18)
Priority: Tier 2 — learning loop
Status: Planned


Job To Be Done

OrthoXpress staff need to teach the system their most common equipment conversations without invasive call recording. The knowledge that makes DME support effective lives in field staff heads.

"The part that's going to make a difference is whether it's equipment or whether it's questions coming in to the PA at a doc's office or whether it's PT, it's the conversations that are sort of on the line." — 2/18 Meeting

Pain

  • Real troubleshooting knowledge is experiential, not documented
  • Recording phone calls feels invasive to staff and patients
  • Equipment manuals are generic — they don't reflect how staff actually help patients
  • Knowledge walks out the door when experienced staff leave

Value

Staff review example AI-generated chats, compare to what they would say, and correct. Same pattern as the doctor-side Conversation Factory — but focused on equipment scenarios instead of clinical ones.

"He was finding out in real time, right in front of us, that his staff was wasting 70% of their time." — 2/18 Meeting

"One of the things we realized at dinner last night... the true North Star is that there are 70% percentage drops... And some conversations are valid, that's the 30%." — 2/18 Meeting


The Steve Example (Why This Works)

"This guy Steve who at day 8 has a free go, just had his knee done eight days ago and in the protocol if you read the seven page handout... it'll tell you that the bruising continues to get worse and worse over the first eight days, right? And so he gets on the app and kind of says basically WTF, right? And it says all right hold on, is the pain in your cap or in your knee... is it a sharp pain or is it a dull pain... in the protocol, oh, we know this happens. Like you will have increasing bruising, swelling that goes right down to your ankle. You're fine. And it talks them down from the clip." — 2/18 Meeting

This is the Conversation Factory in action: a real scenario where the protocol knowledge, delivered conversationally, prevents a phone call.

"Here's how this thing could save everybody time and money and anxiousness and whatever else." — 2/18 Meeting


Requirements

4 seed conversations as concrete next step

Start with 4 hand-crafted example conversations covering the most common equipment calls:

  1. CPM pain/discomfort during use
  2. Cold therapy "leak" (condensation education)
  3. Walker height adjustment
  4. EMS pad placement

These become the templates that staff review and improve.

Staff periodic review

"And I think a guy like Pete, maybe, you know what I mean, right? Or somebody that's been doing it like a... what are all the questions that we're dealing with? What are all the callbacks?" — 2/18 Meeting

Staff compare AI-generated responses to their own answers on a recurring basis. The AI gets better because experienced staff tell it when it's wrong.

ChatGPT-generated complaint lists as accelerator

Use LLM-generated lists of common equipment complaints to seed scenarios faster. Staff validate rather than create from scratch.

Note: Jeff's TKA DME Complaint Research (3 ChatGPT conversations) already provides comprehensive complaint catalogs per device — these are ready to use as seed material.

No call recording

Explicit requirement: the Conversation Factory approach is specifically designed to avoid recording patient calls. Staff teach the system through collaborative authoring, not surveillance.


How It Works

1. System generates example conversation (equipment complaint → AI response)
2. Staff reviews: "Would I say this?" / "Here's what I'd actually say"
3. Staff corrections become training data
4. System updates its responses
5. Next conversation incorporates the correction
6. Repeat — system gets smarter with each review cycle

Connection to Doctor-Side Factory

The doctor-side Conversation Factory handles clinical scenarios — swelling, pain, medication questions, activity restrictions. The DME factory handles equipment scenarios — CPM, cold therapy, EMS, walkers.

Same architecture, different domain knowledge. A patient might interact with both: clinical questions route to the doctor-side protocol, equipment questions route to the DME-side knowledge base.

"Even gets us to a point where we wanna have an unbelievably great experience for the patient, but if we're creating this great experience and saving money to the doctor by alignment of certain things, the experience of that ride might only have to be 70% because they're getting so much benefit of the other side. But the goal is to be experienced at 100% goal." — 2/18 Meeting