Serious AI with Chris Duff
October 9, 2026, updated October 12, 2026
In MGMA Stat’s January 6, 2026 poll, 48% of respondents named denials and appeals as the biggest revenue cycle leak and 23% named the front end, and their comments described the denial losses as overwhelmingly payer-driven, from medical necessity and non-covered services to bundling edits, prior authorization friction, eligibility problems and timely filing. In our view, AI helps most on the work around a denial, checking coverage before the visit, sorting denials by cause, pulling the records an appeal needs and comparing what a payer paid against the contract, and it helps least when it is pointed at a leak nobody has measured.
That answer comes from Serious AI Partners, an enterprise AI implementation firm run by Chris Duff, M.D., who trained as a physician at the University of Illinois College of Medicine and spent 2017 through 2019 in healthcare venture capital, and who built his judgment in a business where being wrong cost him money. The work below is the operating side of a practice: billing, coverage and payer contracts.
The Medical Group Management Association (MGMA), the oldest and largest membership organization representing group practice administration, asked its MGMA Stat panel of healthcare leaders on January 6, 2026 where their biggest revenue cycle leak sits, and received 288 applicable responses. 48% named denials and appeals. 23% named front end issues, such as eligibility and coverage accuracy, then 14% billing and collections, 13% coding and 2% charge posting. MGMA Stat is a weekly polling program: healthcare leaders sign up for the panel, and each Tuesday it receives one polling question by text message. We read the results as what those leaders report about their own operations, rather than as a random sample of practices.
MGMA published a summary of respondents’ comments with the poll. Respondents described denial leakage as overwhelmingly payer-driven and preventable, centering on medical necessity and non-covered services, bundling and global package edits, prior authorization and records-request friction, eligibility and coordination of benefits problems, timely filing and credentialing. Many also pointed to payer errors and policy-change misimplementation, downcoding included. Front-end leakage largely comes from eligibility and coverage accuracy problems, including incorrect insurance entry, outdated demographics, retroactive terminations and delays adding newborns, which then cascade into bad estimates and downstream billing and collections issues. Prior authorization and referral requirements were the other dominant front-end issues.
Eligibility and prior authorization show up on both lists. In our read, that overlap is the part of the leak a practice controls most directly, because coverage and authorization can be checked before the patient is seen. A practice that staffs up its appeals team without looking upstream is paying twice for the same mistake, once at registration and again in rework.
This receipt comes from a different industry, which is the point: the method is the same wherever the money leaks. In July 2026 we ran a diagnostic on a real estate education company’s membership funnel. Booked calls closed at around 30%, and cold conversations closed at under 2%. That is a 19x gap between two paths into the same product, and nobody inside the business had measured it. The diagnostic came back as seven findings, each ranked by its impact and the effort to fix it. It sits on the Proof page.
The parallel to a practice is close. In our view, a denial rate is an average across payers, service lines, locations and front-desk shifts, and an average can hide the one path where much of the money goes. Until someone splits it, buying an AI tool for the revenue cycle is a guess about which leak to plug.
Before the visit. Eligibility and benefits checks are repeatable lookups: whether coverage is active on the date of service, and whether the service needs a prior authorization or a referral. We would set software to run them for every scheduled patient a few days ahead and hand your staff only the exceptions, while there is still time to reach the patient. We would expect that to go straight at the front-end leak and head off some eligibility denials before they exist. The authorization side of that work, and which steps stay with a person, is covered in Can AI Handle Prior Authorization for a Medical Practice?
After a denial. MGMA’s write-up notes that a team can appeal, but appeals take time, and time is the resource nobody has. We see sorting denials by reason code and payer, and pulling the chart notes and authorization records an appeal needs, as clerical work that software handles well, and the same goes for keeping appeal deadlines from slipping. The appeal argument and the conversation with the payer stay with your staff.
Paid against contracted. Our read is that a claim can be paid without a denial and still be paid wrong, for example when a payer pays a lower-level code than the one billed, and that nothing flags it, because nothing was denied. Comparing each remittance against the contract, code by code, is work we would hand to software once the contract rates are loaded, because by hand it is slow and repetitive. This blind spot is covered in the October 5, 2026 issue of Serious AI News, Chris’s weekly newsletter for operators: The AI Slide Said 40%. The Data Said 15%.
Coding. MGMA’s own write-up of the January poll says that for practices that have not explored AI coding tools, earning clinician and coder trust and vetting vendors will be a top priority for 2026, which means demanding proof a tool performs on their specialties and payer mix, not just generic demos. We agree with that bar, since a generic demo tells you little about how a tool handles your own claims.
AI does not find a leak on its own. A practice that buys a denial management tool when most of its losses sit in patient collections has automated the wrong leak, and the subscription renews either way. It also does not replace the judgment calls: the clinical documentation that supports a code is the clinician’s, and the decision to fight or write off a payer’s pattern of underpayment belongs to whoever carries the financial consequence.
The risk runs the other way too. If software pushes more claims out faster without fixing why they were denied, the practice gets its denials sooner and in greater volume. Speed on a broken step makes the step more expensive.
If the next question is how to fund that first project, Where Does a Medical Practice Find the Money for AI in 2027? walks through the four places the money already goes.
AI can produce the analysis. It cannot be wrong, because being wrong requires having something at stake, and that stays with a person. Serious means underwritten.
=====
If your practice knows its denial rate but not which leak is behind it, or already has AI working on the billing side without a number that shows it moving, that is the place to start. The engagement opens with diagnosing the constraint gating the practice, then building with verification sitting where a person still owns the outcome. You leave the first call knowing what your constraint is whether or not you ever hire us.
Fill out a short form, and we read every answer ourselves. If there is a fit we will reach out and find a time that works for you, and if there is not, we will say so and point you toward someone better suited. Either way, the moat question is worth answering now, in a market AI has already reset.