Guide

Clinical documentation automation

The phrase covers at least five different technologies solving different parts of the same problem, which is why evaluating it is confusing: two vendors can both say "we automate your documentation" and mean almost nothing in common.

This is what the category actually contains, which benefits are worth counting, how to roll it out without a dead pilot, and — the part most pages skip — what it does not automate.

Written for independent practices · Physician-authored

What it actually means

Documentation automation is a category, not a product. Most practices end up running more than one layer, and the layers have very different maturity.

Layer	What it does	Where it falls short
Dictation	Speech to text. Mature, accurate, and still genuinely useful.	You still compose the note. It converts typing into talking.
Templates & macros	Pre-built structure and phrases you drop in.	The main engine of note bloat — text that is present because it was easy to insert, not because it was true.
Ambient AI scribes	Listen to the visit and draft the note from the conversation.	Output quality varies by specialty and accent; every draft needs review.
Document intake	Read inbound faxes, referrals and outside records; extract what matters.	Far less mature than scribing, and rarely included in "AI scribe" pricing.
Coding support	Suggest a level or flag missing documentation.	Only as good as the note beneath it, and easy to misuse.

Most of the market conversation is about the third row. The first and second have existed for decades. The fourth is where most of the unglamorous daily work actually sits, and it is the least served.

The benefits worth counting

Time outside clinic hours

The widely cited 2016 Annals of Internal Medicine time-motion study found physicians spending roughly two hours on EHR and desk work for every hour of direct clinical face time, with further documentation after hours. Automation attacks that second figure most directly, because after-hours charting is largely composition work rather than clinical judgment.

Be sceptical of vendor time-savings numbers, including favourable ones. They are usually self-reported, measured over a short pilot with motivated early adopters, and rarely net of the review time the tool adds back.

Documentation that reflects the visit's real complexity

Under the 2021 E/M guidelines the level is driven by medical decision-making, and MDM has to be visible in the record. A complex visit documented thinly gets coded at the safe, lower level — not because anyone intended to undercode, but because the note could not carry the argument. Automation helps here only when it captures reasoning rather than padding the note with normal findings. We wrote that argument out in full in The Defensible Visit.

Retrieval, which almost nobody counts

This is the benefit that compounds. A record structured as it is created is one you can find things in later. If every visit produces another dated document in a pile, you have automated the production of documentation without improving the record — and the six-month problem of "what happened with this patient's kidney function" is exactly as hard as it was before.

What it does not automate

How to automate clinical documentation

Most failed rollouts fail the same few ways: they start with the hardest cases, they skip the baseline, or they add a tool without removing the workflow it was meant to replace.

1. Measure the baseline firstFor two weeks, record time spent charting after clinic. Without this you cannot tell whether anything improved, and you will be left arguing from impressions.
2. Start with the highest-volume repetitive workNot the complex consult. The routine follow-up you do thirty times a week is where automation compounds and where failure is cheap.
3. Run one clinician, not the whole practiceA single motivated user for three weeks surfaces the real objections before you have trained everyone on something that does not fit.
4. Keep review in the loop, explicitlyDecide who reviews what before go-live rather than discovering the answer after a bad note is signed.
5. Remove the thing it replacedIf the template and the scribe both stay, you have added a step. This is the most common reason a pilot shows no time saved.
6. Re-measure at 30 daysAgainst the baseline, on the same metric. If it has not moved, the tool is wrong or the workflow around it is.

What to ask vendors

Questions whose answers actually distinguish products, including ones we do not win:

• Does it write back into my EHR? Many products, ours included, do not. Ask for a demonstration in your EHR rather than a general claim — coverage varies enormously by system.
• What happens to inbound faxes and outside records? Usually nothing. If document intake matters to you, ask early; it is rarely in the base price.
• Are you SOC 2 certified? A fair question with a real answer. We are not — we are HIPAA compliant and sign a BAA. If SOC 2 is required to sign, ask upfront rather than late.
• What does the chart look like at six months? Ask to see one, not a single note.
• What is the real monthly cost at my volume? Per-encounter fees and usage caps are where advertised pricing and actual pricing diverge. Ours is on the pricing page.
• Is there a trial that does not need a card? The answer tells you how confident they are.

If you are comparing specific products rather than deciding whether to automate at all, we keep a physician-written roundup of AI scribes for independent primary care — including the two that cost nothing, and the visit type that separates them.

Common questions