CDI software and an ambient AI scribe touch the same chart at different times. The scribe drafts what happened during the encounter. Clinical documentation integrity work asks whether the signed record supports the codes, quality measures, and claim that follow. Buying one to do the other’s job creates a very polished gap.
Takeaways:
- An ambient scribe starts with the visit; CDI starts with the record and its downstream use.
- CMS tells clinicians to make the record support the codes reported on the claim.
- A scribe can capture a stated diagnosis, but it cannot decide which diagnosis is supported.
- The clean handoff is a signed note plus a traceable queue for questions, not one blended score.
One chart, two work queues
Picture a Tuesday afternoon in an internal-medicine group. The clinician says, “Shortness of breath is worse, weight is up six pounds, increase the diuretic, recheck Friday.” An ambient scribe can turn that conversation into a structured draft. The clinician corrects it and signs.
The CDI question begins after that. Does the record contain the assessment that supports the diagnosis selected for the claim? Is a condition merely listed in history, or was it evaluated today? If the documentation conflicts, who asks the clinician for clarification without steering the answer?
CMS’s current E/M guidance says the medical record should support the CPT, HCPCS, and ICD-10-CM codes reported on the claim. It also lists the encounter elements the record should carry, including the reason for the visit, relevant findings, assessment, plan, and the identity of the observer. That is a documentation standard. It is not permission for software to invent the missing bridge.
The handoff map
This is the simplest useful artifact for separating the two purchases.
| Moment | Ambient scribe owns | CDI workflow owns | Human decision that remains |
|---|---|---|---|
| Visit | Capture the words spoken in the encounter | Nothing yet | What the clinician asks, observes, and concludes |
| Draft | Arrange the stated history, assessment, and plan | Define which signed records enter review | Whether the draft is faithful and complete |
| Signature | Return the draft for clinician review | Wait for the authenticated record | Corrections, diagnosis choice, and signature |
| Review | No new clinical facts | Detect a documented gap or conflict | Whether a query is warranted |
| Query | No role | Route a neutral clarification under policy | The clinician’s answer |
| Coding and claim | Offer ICD-10 suggestions only | Check whether documentation supports downstream coding | Final code and claim responsibility |
The dangerous design is a shared green check that claims the note is both fast and compliant. Those are two tests. Keep them separate in the user interface and in the pilot report.
CDI is a program before it is a product
CMS’s ICD-10 implementation handbook tells large practices to audit current documentation, focus on common clinical conditions, identify recurring gaps, monitor documentation quality, and create templates that guide required information. The handbook was written for the ICD-10 transition, but the operating lesson still holds: you need a target condition, a review sample, an owner, and a feedback loop before a dashboard means much.
Start with 50 to 100 encounters from one service line. Record why each case entered review. “Missing specificity” is too vague. Write the missing element: laterality, acute versus chronic status, the link between two conditions, the work performed today, or the rationale for an order. Then count repeats by reason and by workflow stage.
That baseline tells you where the gap forms. If the clinician stated the detail but the draft lost it, test the scribe. If nobody stated it and the signed note lacks it, that is a documentation and query problem. If the note contains it but the code does not, the issue sits farther downstream. One chart can fail in three different places.
What the scribe changes
An ambient scribe removes the blank page. It can preserve the assessment and plan the clinician actually said, which gives the signed record a cleaner starting point. It can also produce more prose than the encounter needs. A long note with a copied problem list is not automatically useful to CDI.
The Peterson Health Technology Institute’s 2025 report on early ambient-scribe adoption found that health systems were still defining success across purchasing rationale, vendor selection, rollout, and impact measurement. The report explicitly asks whether organizations want a custom tool or an off-the-shelf product and how much IT and operations effort they will contribute. That uncertainty is useful. It argues for a measured pilot, not a claim that a scribe fixes the revenue cycle.
Patient Square is an AI clinical platform. Practice Copilot brings the whole practice under one AI copilot: an ambient AI Medical Scribe that hands back a structured SOAP note, ICD-10 suggestions, and a prescription draft minutes after the visit, plus a bundled AI EHR, scheduling, and messaging as you move up the plan. Hospitals get Hospital Copilot.
The operative word is “suggestions.” The clinician confirms the diagnosis and code. Patient Square does not claim autonomous coding, compliant-query generation, denial prediction, or a replacement for a CDI team. A practice can use the bundled EHR from the Copilot tier or keep its current EHR and use Practice Copilot alongside it. Named write-back integrations are not part of that claim.
A pilot that does not confuse speed with integrity
Run the pilot in two lanes. Lane one measures note creation: minutes to draft, minutes of clinician editing, completion by end of day, and abandoned drafts. Lane two measures the signed record: review rate, query reason, answer rate, coding change after clarification, and cases escalated for human review.
Do not merge those measures into one “documentation quality” percentage. A scribe could cut draft time while leaving the CDI queue unchanged. CDI work could reduce recurring gaps while adding a few minutes to a complex chart. Both outcomes can be rational.
We think the best buying sequence is boring: map the current failures, test the scribe against spoken-detail loss, and test CDI against signed-record gaps. Software demos come after the map. The small-clinic implementation guide has a narrow rollout pattern, while the claim-denial documentation guide keeps the claim boundary visible.
If note creation is the first broken step,