What Should a Medical Practice Check First With AI Heading Into 2027?

Check five things on your own data before you plan any new AI for 2027: whether payers are paying visits at a lower level than you billed, whether you are paid what your contracts say, whether the AI tools you already pay for were vetted, whether anyone measures what those tools produce, and how many of your denials start at the front desk. Each one has a test your staff can run this quarter on numbers the practice already holds, and the results tell you where AI belongs before a vendor tells you.

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, spent 2017 through 2019 in healthcare venture capital, and built his judgment in a business where being wrong cost him money. Everything below sits on the operating side of a practice (billing, payer contracts, front desk and the AI tools around them), and none of it is clinical advice.

What changes for a practice in 2027?

Three things are worth having on the calendar.

Electronic prior authorization. Under the CMS Interoperability and Prior Authorization Final Rule (CMS-0057-F), Medicare Advantage organizations, state Medicaid and CHIP fee-for-service programs, Medicaid managed care plans, CHIP managed care entities and Qualified Health Plan issuers on the federal exchanges must implement a Prior Authorization API, with compliance dates generally beginning January 1, 2027 and varying by type of payer. That API has to be populated with the payer’s list of covered items and services, identify the documentation required for approval, and send back an approval, a denial with a specific reason, or a request for more information. On the practice side, CMS added an Electronic Prior Authorization measure for MIPS eligible clinicians, starting with the 2027 performance period: a yes or no attestation that the clinician requested a prior authorization electronically through a Prior Authorization API, using data from certified EHR technology, for at least one medical item or service (drugs excluded), or an exclusion. Ask your EHR vendor now how it plans to connect to those APIs, rather than in December.

Payer policies that lower the level paid. The Peterson Health Technology Institute (PHTI), which provides independent evaluations of healthcare technologies, says in its insight Administrative AI: Current Use and Potential Impact (last updated October 15, 2026) that health plans are beginning to respond to AI-driven increases in billing intensity with across-the-board downcoding and other reimbursement reductions, and that the impact of those cuts is not yet known. PHTI also notes that several health plans have paused, delayed or rescinded policies following provider opposition and state regulatory scrutiny, so read the current E/M reimbursement policy of each payer you bill rather than last year’s.

AI tools past their first year. In an August 4, 2026 MGMA Stat poll, 83% of medical groups said AI plays some role in patient visits, against 71% a year earlier. MGMA Stat is a weekly text-message poll of healthcare leaders who sign up for the panel, so read it as what those leaders report rather than a random sample. In our read, the renewal of a tool bought in 2025 or 2026 is the cheapest moment to ask whether it is working.

Why check before you buy?

The receipt here comes from Serious Land Capital, Chris’s own land investment firm. Underwriting (working out what a parcel is actually worth and whether the deal clears at that price) is the highest-value hour in a land business and the one where being wrong is most expensive, and it is the work the firm’s Deal Engine software now runs. By hand it took 45 to 60 minutes per parcel. The first fully measured automated run, on August 18, 2026, took 3.5 hours and burned 10 to 15% of a week’s model allowance: correct answer, unsustainable economics. It now runs in under 30 minutes unattended at about 1% of the weekly allowance, and when it disagreed with the manual read by roughly $19K on a test deal, the cause surfaced the same day. The re-run landed at $87,992 against a hand-built $88,000.

My read is that none of those numbers would mean anything without the hand-measured baseline sitting next to them. A practice choosing its first or next AI project is in the same spot. MGMA’s February 10, 2026 poll, with 177 applicable responses, asked where practices are turning to AI and automation to improve front-office access, and the answers split across scheduling (31%), calls (27%), registration and eligibility (23%), prior authorization (16%) and other (5%). Pick the constraint the numbers point to, rather than the loudest complaint in the office.

What are the five checks?

Payments that come back smaller than billed. Pull 90 days of remittances and compare the code billed with the code paid for 99214 and 99215 (and 99204 and 99205 if you see new patients), split by payer. My bet is that a visit paid one level down with no denial reads as paid on the remittance, and slips by unless someone runs this comparison. The October 2, 2026 episode of Chris’s podcast, Your payer can underpay you without a single denial, walks through that blind spot.

Contract rates that go unchecked. Take your top payers by volume, load the contracted rate for your most-billed codes, and compare it with the allowed amounts on the same 90 days of remittances. If a scheduled increase never got loaded, this is where you will see it. The wider question of where a practice loses revenue, and where AI helps with it, is in Where Does a Medical Practice Actually Lose Revenue, and Where Can AI Help?

AI tools bought before anyone vetted them. List every AI tool the practice pays for, who approved it, and whether the contract says what happens to your data and how you hear about model updates. In a January 20, 2026 MGMA Stat poll of medical group leaders, with 328 applicable responses, 56% said their organization neither has AI governance or a formal AI policy nor is developing one. The questions to put to a vendor are in How Should a Medical Practice Evaluate an AI Vendor Before It Buys?

AI output that goes unmeasured. In a May 12, 2026 MGMA Stat poll with 257 responses, 46% said new AI tools had made their providers more productive in the past two years, 27% reported no productivity gains and 14% were unsure, and MGMA’s write-up says uneven use, paired with limited tracking and feedback, makes it hard to establish a productivity baseline, let alone measure change. The test: each month, have a coder pull a small sample of AI-drafted notes per provider and record how many support the level billed. We think that error rate is the number to bring to a renewal, and where a payer downcodes, I’d treat a scribe that nudges coding levels up without the documentation to match as a downcode waiting to happen. Chris’s October 5, 2026 newsletter issue, The AI Slide Said 40%. The Data Said 15%., makes the same case for naming one number per tool.

Denials that start at the front desk. Export a quarter of denials and count the share whose reason codes point to eligibility, coordination of benefits or registration errors, weighted by dollars. In MGMA’s January 6, 2026 poll, with 288 applicable responses, 23% of respondents named the front end as their biggest revenue cycle leak, and their comments tied it to incorrect insurance entry, outdated demographics and retroactive terminations. MGMA’s own suggestion, in its write-up of the February 10 poll, is to verify coverage at scheduling and again close to the visit, with automation pushing exceptions into prioritized work queues.

What should a practice do with the results?

Put a dollar figure or an hour count beside each of the five, then pick the largest. Write down today’s value and the one person who owns it, and only then decide whether the fix is a process change, software your team builds, or a vendor. If the number lives in remittance or chart data and an outside tool would see it, settle the agreements with your counsel and compliance lead before the pilot. The list is a sampling of the areas we keep finding worth a look, the same one on the Serious AI healthcare page, and the thing slowing your practice down may well be somewhere else. The receipts behind the method are on the Proof page.

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 is heading into 2027 with a guess about where it loses money, or with AI already working on a problem but no number that shows it paying off, that is the place to start. The engagement opens with diagnosing the constraint gating the practice and where AI is falling short on it, 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.