Why we built Signal: An EHR for the session, not the billing department
May 13, 2026 · 7 min read · By the Signal team
Most therapists spend more hours documenting than they do with clients.
That sentence still sounds wrong every time we say it out loud. But it is the lived reality of nearly every clinician we have spoken to in the last two years. A 50-minute session generates an hour of paperwork once you count the note, the billing claim, the message to the patient, the calendar update, and the inbox triage. That is the problem we set out to solve.
The math is upside down
A therapist sees a client for 50 minutes. The session ends. Then begins the second shift: a SOAP or DAP note that takes 20 to 40 minutes if you are fast and 60 if the session was hard. A claim to file. A reminder to send. A treatment plan to update. A balance to chase. Multiply that by six clients a day, five days a week, and you have a clinician who is doing two jobs and only getting paid for one.
The other half of that math is what gets lost. By the time a therapist sits down to write a note three hours after the session, the texture is gone. The exact phrasing the client used. The pause before they answered. The shift in tone when they mentioned their mother. The note that gets written is a sketch of a sketch. It is good enough for billing. It is not good enough for clinical work.
We started Signal because that was unacceptable to us. The note should be a clinical artifact, not a billing receipt.
Legacy EHRs were built for the billing layer
SimplePractice, Jane, TherapyNotes — these are good products. They solved a real problem. Before they existed, therapists were running their practices out of paper calendars and Excel spreadsheets and getting their CMS-1500 forms rejected for reasons nobody could explain. Those tools made it possible to run a small practice without hiring a full-time biller.
But here is the thing nobody says out loud: those products were built for the billing layer. The clinical layer was an afterthought. The note editor is a glorified text box. The progress tracking is a checkbox. The treatment plan is a PDF template you fill out once and forget. The system has no idea what happened in the room. It only knows that 50 minutes elapsed and a CPT code needs to be attached to it.
That is not a criticism of the people who built those tools. It is a description of what was technically possible in 2012. You could not put a microphone in a therapy room and ask a computer to understand what was being said. So the EHR became a filing cabinet with a payment processor stapled to it.
That constraint is gone now.
What changed: the models got good enough
Three things converged in the last 24 months that made Signal possible.
First, speech-to-text got good enough to actually hear a therapy session. We use Deepgram, and the word error rate on natural conversation, including accents and emotional speech, dropped below the threshold where you have to babysit the transcript. It is not perfect. But it is accurate enough that what comes out the other end is something a clinician can use.
Second, speaker diarization — the ability to tell who is talking — became reliable enough for two-person and small-group clinical conversations. This sounds like a small thing. It is not. Without diarization, you get a wall of text. With it, you get a transcript that knows the difference between what the client said and what the therapist said, which is the entire ballgame for clinical analysis.
Third, language models got accurate enough to draft clinical notes that a therapist would actually sign. Not summaries. Not bullet points. Real notes in DAP, SOAP, BIRP, EMDR, IFS, CBT and a dozen other formats — written in a way that reflects how the therapist thinks about the work, not how a generic AI thinks about the work.
Once those three things were true, the question was no longer whether you could build an EHR around the session. The question was whether anyone would.
Three things we believed when we started
Every product is the answer to a few stubborn beliefs the founders refused to let go of. Here are ours.
The note should write itself, not the other way around
The therapist is not in the room to feed the EHR. The EHR is supposed to feed the therapist. If you have to think about your note while you are with a client, the product has failed. By the time the session ends, a draft should be sitting there waiting for you to read it, edit one or two things, and sign it.
The AI should help during the session, not just after
Note-generation is the obvious win. It is also the easiest win. The harder, more interesting question is what an AI can do for a clinician while the session is actually happening. Real-time emotion analysis. Pattern recognition across sessions. A quiet signal that the alliance is shifting, three sessions before the client cancels. That is where the product is going.
The front desk should not be a separate hire
A solo practitioner cannot afford a receptionist. A small group practice can barely afford one. So the work of the front desk — answering calls, booking appointments, sending reminders, handling cancellations, chasing balances — falls back on the clinician. We built Amelia so it does not have to. Amelia is the AI staff member who answers the phone, books the appointment, and sends the invoice. She is included in the price.
What we built: a brief tour
Signal is one platform. It runs the clinical work and the practice work in the same system, because pretending those two things are separate is what got us into this mess in the first place.
- Session capture. You hit record. Signal listens. We use Deepgram for transcription, run our own diarization pipeline, and pass the result to a clinical analysis engine that we built ourselves. By the time the session ends, you have a transcript, a draft note in your preferred format, an emotion timeline, and a set of session highlights.
- Clinical notes in 16 formats. DAP, SOAP, BIRP, EMDR, IFS, CBT, DBT, ACT, narrative, group, couples — whichever way you write, Signal writes that way. You read it, edit what you want, sign it.
- Amelia. A chat receptionist that lives on your website and a voice receptionist that answers your phone. She books appointments, handles rescheduling, sends reminders, and matches new patients to the right therapist using six factors including specialty, insurance, and clinical urgency.
- Insurance billing. Unlimited claims, included in the subscription. Eligibility checks, claim submission, ERA reconciliation, denial work — all of it. We did not want to be the EHR that nickel-and-dimes you per claim.
- Practice management. Calendar, intake forms, telehealth, secure messaging, treatment plans, outcome measures, document storage. The boring stuff is in there. It just is not the point.
The whole thing is $199 per practitioner per month. Admins and receptionists are free seats — we only charge for the people doing the clinical work. There is no per-claim fee, no per-message fee, no setup fee, no annual contract.
What we will not do
A few promises that matter more than the feature list.
We will not train models on your session data.
Not on transcripts, not on notes, not on anything that came out of a clinical encounter. The models we use to draft your notes do not retain your data after the request is finished.
We will not sell your data.
Not to advertisers, not to pharma, not to insurers. Not in aggregate, not anonymized, not under any other label that means the same thing. PHI is yours and your client's. Full stop.
We will not lock you in.
Your notes, your client records, your billing history — exportable in standard formats, on demand, without a phone call. If we ever stop being the right tool for you, you should be able to leave with everything you put in.
Where we are going
Signal today is what we needed to ship to make the case that an EHR can be built around the clinical work. The next two years are about what becomes possible once that foundation is in place.
Outcomes tracking that is actually useful — not a PHQ-9 score in a chart, but a longitudinal picture of how a client is changing across the dimensions that matter for the work you are doing with them. Pattern detection across sessions, so the system can quietly flag that a client's alliance ratings have been drifting for a month before they ghost. A clinical understanding of each patient that gets richer with every conversation, so when you walk into session 47, the prep card knows what session 12 was about.
The bigger bet is that the EHR stops being a place where you do paperwork and becomes a place where you do clinical thinking. Where the system is a quiet collaborator that makes you a better clinician — not a slower one.
We are not done. We do not pretend to be. But we know what we are pointed at, and we know why.
Try Signal free for 14 days
Real-time session capture. AI clinical notes in 16 formats. Amelia AI receptionist. Unlimited insurance claims. Full practice management. $199/practitioner/month after the trial. No credit card required to start.