Candid Health

Artificial intelligence is a feature layer on a rules engine whose value stands without it. Candid's Automation and AI page describes an advanced rules engine carrying reverse engineered payer rules, kept current with payer requirements, plus customer authored rules and 24/7 claim autocorrection. Its own answer to how it uses AI is that ML and AI automate the feedback loop between claim insights and systemic fixes, deployed through the rules engine.

That is a learned component by its stated function, and the product the customer buys is the rules engine and its payer data. Candid describing its core as data engineering and rules, rather than calling the whole thing AI, is accurate and useful to a buyer.

The oversight structure is described with parts missing. Candid defines touchless claim rate as the percentage of claims submitted and finalized without human intervention, and places control at the rule layer: customers create custom business logic and enforce their own workflows, with training offered. The platform also validates and autocorrects billing data around the clock, and its services team steps in when a human touch is required.

What is not published is what autocorrection may and may not change, or what routes a claim to a person rather than through the engine. The home page line 'Never make the same mistake twice' is recorded as a claim, not a control.

The approach is named without version or update discipline. Candid describes a rules engine carrying reverse engineered payer rules, continuously refined against payer requirements; customer authored rules on top; 24/7 validation and autocorrection of billing data; and ML and AI that automate the loop from claim insights to rule changes. No model, technique behind the ML, accuracy figure or change log for rule updates is published on the product pages.

The architecture is described and no model provider is named. Candid describes a rules engine with reverse engineered payer rules and ML that turns claim insights into rule changes; nothing published says whether that ML is built in house or uses a third party model, where it runs, or which subprocessors receive claims data, and no subprocessor list is available. Claims carry diagnosis and procedure codes and patient identity.

Customers are named with headline figures, and no method stands behind them. Candid's home page names Talkiatry (manual work down 40 percent while claim volume grew), Bridge (net denials cut by more than half), Tia, Brave Health and Nourish, each with a case study. The home page defines payer net collection rate and touchless claim rate without publishing a baseline, period or denominator beside them.

General assurances that do not answer the questions AI raises. Candid's privacy policy says protected health information is governed by the business associate agreements it signs with customers, not by the policy, and that Candid may share aggregated or de identified data derived from personal information to improve its services. Candid announces a SOC 2 Type 2 examination.

Claims carry diagnosis and procedure codes, patient identity and payment detail, and the rules engine's value compounds across customers, yet nothing published states whether one customer's claims data trains the models or rules used for others, or how long claims data is retained. Safety engineering is not published.

Business associate status is stated and the only privacy document behind it covers the website. Candid's privacy policy (updated 5 March 2025) says Candid signs service agreements and business associate agreements with its customers, that those agreements govern the claims and other information entered into the platform, and that the policy itself does not govern protected health information. No agreement template, subcontractor terms or service level privacy notice is published.

A recognized certification is named without the artifact. Candid announced completing a SOC 2 Type 2 examination with an unqualified opinion, after an earlier Type 1. No trust portal publishes the report or its period, and the site's footer links only terms, privacy and cookie settings.

Claims processing and billing automation sit outside software as a medical device, and Candid scopes the product accordingly, with no device claim. What governs is payer compliance and claims accuracy, and liability for an incorrect claim sits with the submitting organization, under the False Claims Act where a government payer is involved.

Two published features bear on that exposure: customers can inspect and author their own payer rules, and the platform validates claims before submission rather than reworking denials afterward. Touchless by definition means no person verifies the individual claim, and nothing published says what governs rule changes or what proportion of claims a person reviews.

Nothing published on how model or rule behavior is governed or tested. The home page, the Automation and AI page, the integrations page and the privacy policy carry no governance framework, monitoring commitment or evaluation. The risk here is coding drift rather than demographic bias: a rules engine tuned for touchless submission and net collections optimizes for claims that get paid, which is not the same as claims that are correct, and the machine learning that proposes rule changes learns from what paid. Incorrect claims to Medicare create False Claims Act exposure for the billing organization, not the software vendor.

Mechanisms exist and nothing stands behind them. Customers can author, inspect and manage their own rules, the platform flags and fixes issues it detects, and Data and Reporting gives visibility into claim outcomes, so a billing team can see and challenge what the automation did.

Candid defines its headline metric, touchless claim rate, as the percentage of claims submitted and finalized without human intervention, which makes it auditable, but no figure with a denominator and period sits beside the definition, and the company publishes no accuracy commitment, remedy or indemnity. Autocorrection changes billing data on its own, and the site does not say what it may change, how changes are logged, or what Candid commits to when an autocorrection is wrong.

Named systems and standards support, with the direction of each connection unstated. Candid's integrations page names EHR partners Elation, Canvas, Healthie, Medplum, Welkin, Avon, Develo and MassLight, alongside eligibility, credentialing, lockbox and patient collections partners, and states native HL7v2 support and public API documentation at docs.joincandidhealth.com. The named EHRs serve digital health and ambulatory groups; no hospital record system is listed. Which fields each prebuilt integration reads and writes back is not shown on the integrations page.

A single hosted service with the location implied. Candid's privacy policy says information collected through its website and services is processed and stored in the United States, but the same policy excludes protected health information, which is governed by customer business associate agreements. No cloud provider, region, tenancy model, subprocessor list or data return terms are published for the platform. The connection layer is documented well: public API documentation, HL7v2 support and prebuilt integrations.

No price is published and the posture is discoverable. Candid sells a billing platform to medical groups, managed service organizations and digital health companies, plus billing services for when a human touch is required, and publishes API documentation so a buyer can gauge integration effort. Nothing states whether pricing is a percentage of collections, per claim or a platform fee, and the three produce very different costs as a practice grows.

Coverage is named with validation behind part of it: medical groups, managed service organizations, digital health organizations and billing services. Named customers support the digital health and outpatient setting, including psychiatry (Talkiatry), behavioral health (Brave Health), women's health (Tia) and nutrition care (Nourish). No hospital or facility billing customer is named, and the EHR partners listed serve ambulatory and virtual care.