Research program, in development

BillingBell Sentinel™

Verify AI-generated healthcare billing actions before submission.

Medical billing, revenue cycle management, AI verification

The problem

AI tools are entering healthcare billing quickly. They suggest codes, draft appeal letters, fill in authorization requests and summarize payer rules. They can save time, but they can also be confidently wrong.

A wrong code or a missing authorization number sent to a payer causes a denial at best. At worst it creates a compliance problem for the practice. Providers already doubt that AI understands payer-specific rules, and only about 14% use AI for denials today (Experian Health, 2025).

The idea

Sentinel is a proposed verification layer that sits between an AI tool and the payer. Before an AI-generated billing action is submitted, Sentinel would check it against the clinical documentation, the patient's coverage and the payer's rules, and explain any risk it finds.

A trained person then reviews each flag and decides. Sentinel does not make clinical decisions, diagnose, or submit anything on its own.

Proposed verification workflow

Concept only. Not a deployed product.

AI suggests an actionA code, an appeal, an authorization request

Sentinel checks itAgainst the note, the patient's coverage and payer rules

Risk is explainedEach flag states what is wrong and why, in plain language

A person decidesThe biller accepts, corrects or rejects the action

SubmissionOnly reviewed actions go to the payer, with an audit trail

Research questions

How often do AI-generated billing actions contain errors that would cause a denial or compliance risk, and which kinds of errors are most common?
Can automated pre-submission verification detect those errors reliably, with a false-alarm rate low enough that billers trust it?
Does human review supported by explained flags reduce denials and rework compared with unassisted review?
How does the approach affect administrative time and burnout for small-practice billing staff?
What does patient benefit look like: fewer surprise bills, faster authorizations, fewer delays in care?

Research phases

Planned sequence. Timing depends on partners and funding.

Discovery
Interviews with physicians, practice managers and billers to map where AI errors enter the workflow.
Error study
De-identified review of AI-generated billing actions to classify error types and frequency.
Prototype
Rule and model-based checks tested offline against expert-reviewed examples.
Pilot
Supervised use in partner practices, measuring denials, rework time and user trust.

Safeguards

Human review required for every flagged action
HIPAA-aligned handling; de-identified data for research analysis
Data minimization and access only to what a check needs
Explanations for every flag, and a full audit trail
Independent ethics review before any study with patient data
Results published with methods, including what did not work

Why BillingBell

Sentinel grows out of real billing work. Our revenue cycle team sees daily where claims fail and why, which gives the research grounded problems and real workflows to test against.

Read our responsible AI principles.