SmarterDx

One clinical AI layer sits behind every SmarterDx product. Trained on more than 21 million encounters, it reads the full record, about 30,000 data points a chart across notes, labs, vitals, orders, imaging and codes, and surfaces missing or incorrect diagnoses, charge opportunities, medical necessity evidence and denial root causes.

People validate every finding, which settles who decides rather than who does the work: no coding department reads every data point of every chart, and without the model there would be nothing to validate.

Nothing goes to a payer from SmarterDx. The hospital's documentation and coding teams validate every finding before final billing, and the company never submits a claim.

The product pages pair that design with completeness promises: the platform page offers to 'capture every diagnosis before final billing', the home page repeats 'capture every diagnosis', and SmarterCharges promises to 'capture every charge'. Validation can only catch errors in what the engine surfaces, so a team that takes the promise at face value stops looking for what it missed.

Training data is described in some detail: more than 21 million clinically validated encounters of real EHR data, from which the models recreate clinical reasoning across the roughly 30,000 data points in a chart, and they keep learning from every validated finding.

No foundation model, model class, version, change practice or precision and recall figure is given.

AWS appears as a hosting badge, and the training corpus is described. No foundation model provider, model class or subprocessor is named, and there is no subprocessor list.

SmarterNotes came from the acquisition of Pieces Technologies, announced in September 2025, and combines SmarterDx's clinical AI with Pieces' note automation; the components of that combined stack are not described separately.

Named clients include Novant Health, McLaren Health, UCHealth, OHSU, UAMS, Universal Health Services, Franciscan, Baptist Health Arkansas, Baxter Health and Carle Health, and SmarterDx reports more than 85 health systems and 300 hospital sites. McLaren's chief financial officer credits the platform with more than $11 million in annualized net new revenue, one health system recovered $1.5 million from a denials backlog, and an academic health system reports a 91 percent win rate on complex denials.

Net new revenue is a counterfactual, money the organization says it would otherwise have missed, and it depends on assumptions about the prior process. The 5 to 1 return and the $3.5 million average per 10,000 DRG discharges rest on the same construction, and no outside evaluation exists.

Every chart reviewed and every validated finding feeds a live learning loop, according to the platform page, so customers' validated findings train the shared model. SOC 2, TX-RAMP and HIPAA badges and AWS hosting are the other assurances shown.

Retention, de identification before training, and whether a customer can keep its charts out of the shared corpus are not stated, and guardrails against wrong findings are described only through human validation.

SmarterDx's privacy policy, last updated July 16, 2026, refers to protected health information it holds as a business associate under HIPAA, and the home page shows a HIPAA badge. That policy covers the website rather than the service.

No business associate agreement, service privacy notice or subcontractor terms are available, although the platform processes complete inpatient records across more than 85 health systems.

The home page shows a SOC 2 badge without a report type and a TX-RAMP certification badge. TX-RAMP is issued by a Texas state authority for a defined cloud service and a fixed term after review of its security controls, and it is required to serve Texas public institutions.

There is no trust center, report request route, audit period or penetration testing statement.

Finding documentation gaps and revenue opportunities in a clinical record is an administrative function, and no device pathway applies or is claimed.

The regulatory weight sits in payment. SmarterDx surfaces missing diagnoses on inpatient charts, and added diagnoses are what shift payment weight under DRG grouping; documentation improvement that moves DRGs is among the most examined areas of federal program integrity work, where the question is whether an added diagnosis reflects care delivered and documented. The company's answer is design: each finding is tied to chart evidence and validated by the hospital's clinical staff before billing. The documentation standard applied to a surfaced diagnosis, audit sampling of accepted findings, and the response when a payer challenges one are not described.

Every finding is attributable to chart evidence for the customer's team to validate and defend, and the engine reads every data point of every chart rather than a subset, which removes one quiet source of bias.

No governance framework, monitoring result, acceptance or reversal rate, or breakdown by specialty, payer or physician is available. The product exists to add diagnoses and charges and learns from each validated finding, so drift toward higher documented severity would not show in any single chart.

Each finding arrives with its chart evidence for the hospital's documentation and coding teams to validate before final billing, so a person decides and SmarterDx never submits a claim.

No precision, false positive or acceptance rate, confidence threshold, warranty or indemnity is offered. Because the hospital validates and bills, an accepted but unsupported diagnosis is the hospital's False Claims Act exposure.

Reading notes, labs, vitals, orders, imaging and codes for every chart, with implementation in under eight weeks, requires deep access to the hospital's record system.

No record system, standard, interface or marketplace listing is named, and nothing says whether findings write back into the EHR or stay in SmarterDx's own work queue. Named clients run enterprise EHRs, but a customer list does not show how the connection works.

AWS hosting is shown through a badge, and complete inpatient records for every discharge move to the platform.

No processing region, tenancy model or residency commitment is stated, and the TX-RAMP badge does not say which services or which certification level it covers.

SmarterDx says its fees are tied to the value it delivers, and its return calculator sizes that value: annual DRG discharges divided by 10,000, multiplied by $3.5 million, with conservative and liberal cases at $3.0 million and $4.0 million. Implementation takes under eight weeks.

No price, percentage or unit of charge is stated. A fee tied to identified or realized revenue carries different budget and audit consequences from a subscription, and whether the fee differs across the seven products is not explained.

Inpatient hospital revenue cycle is the setting, from prior authorization and medical necessity before care, through documentation, diagnosis and charge capture, to denials and underpayments after the claim. Named health systems and academic medical centers back it, and the return calculator is keyed to DRG discharges.

Ambulatory and professional coding are not covered, and coverage is described by revenue cycle stage rather than clinical specialty, so performance in complex surgical or oncology documentation cannot be told apart from general medicine.