Billing without getting in the way

The note is done. The session is over. And somewhere there's still a billing tail: a CPT code to pick, an ICD-10 to confirm, a modifier to remember, a timesheet to reconcile.

In behavioral health, the billing tail is often longer than the note itself. It's also where most solo and small-practice clinicians lose an hour or two each week. Not to the coding itself, but to the cognitive switching between clinical and administrative thinking.

This post is about where Aimé steps in and, just as importantly, where we explicitly stop.

What Aimé does on the billing side

When a note is drafted, Aimé also produces a billing suggestion:

• Suggested CPT code, based on session length, modality, and the content of the session (e.g., crisis intervention vs. routine follow-up).
• Confirmed ICD-10 diagnoses pulled from your intake and updated if the session shifted the working diagnosis.
• Session duration, computed from transcript timestamps rather than a separate timer.
• Modifiers when telehealth, interpreter, or out-of-session work applies.

Everything is a suggestion. Nothing is submitted anywhere. The clinician approves, edits, or overrides before the billing info gets written back to the EHR.

Where we stop, deliberately

A question that's come up on every enterprise call we've had:

"Can Aimé just submit the claim for me?"

No. Not now, not in the current product, probably not ever as a default. Here is why.

Submission means ownership. The moment software submits a claim, the question of who is responsible for a miscoding gets murky. Insurers and regulators treat the clinician or practice as responsible regardless, but the workflow of "I pressed a button and it was gone" creates a practical gap between the decision and the accountability. We'd rather the clinician press the button themselves, even if it's a two-second confirmation.

Auto-coding is the place where AI hallucinations become fraud. A miscoded CPT is a billing error. A systematically miscoded CPT across hundreds of sessions is, at some threshold, a compliance problem. We don't want to be the layer that introduces a systematic bias into someone's billing.

Suggestion + confirm is fast enough. The clinicians who've asked for auto-submit have all, after using the confirm flow for a month, stopped asking. Two seconds to confirm a suggestion isn't the bottleneck they thought it was.

What good billing automation looks like

Our shortlist for what's worth automating and what isn't:

| Task | Automate | Why / Why not | | -------------------------------------------- | -------- | ---------------------------------------- | | Session duration from the transcript | ✓ | It's a measurement, not a judgment | | CPT suggestion based on modality and length | ✓ | With clinician confirmation | | ICD-10 suggestion from note content | ✓ | Diagnosis is the clinician's call | | Superbill generation | ✓ | Reduces paperwork without making claims | | Claim submission to payer | ✗ | Clinician responsibility, not software's | | Insurance verification | ✓ | Pure lookup, no clinical judgment |

The meta-point

There's a version of AI in healthcare that tries to automate everything that used to require a clinician's signature. We don't think that version is good for patients, and we don't think it's good for clinicians either.

The useful version of AI in a practice is the one that does the clerical, measurable, low-judgment parts of the work and leaves the decisions in the hands of the person whose name is on the chart. The billing tail, for us, is exactly that.