Sift Healthcare (@MdSift) header
Sift Healthcare (@MdSift) profile picture

Payer-specific intelligence that equips hospitals to prevent adverse payment outcomes (denials, underpayments, takebacks, downgrades) and protect net revenue.

Milwaukee, WI
Joined August 2018
Photo by Sift Healthcare (@MdSift)Photo by Sift Healthcare (@MdSift)Photo by Sift Healthcare (@MdSift)Photo by Sift Healthcare (@MdSift)Photo by Sift Healthcare (@MdSift)Photo by Sift Healthcare (@MdSift)Photo by Sift Healthcare (@MdSift)Photo by Sift Healthcare (@MdSift)Photo by Sift Healthcare (@MdSift)Photo by Sift Healthcare (@MdSift)
A payer deploying AI to deny more claims faster is, structurally, good news for most RCM vendors. Their contracts are built on the problem existing. RCM vendors have gotten very good at tracking denial rates, documenting appeal outcomes, and reporting KPIs. They're systems of record that organize and report on data. But the root causes of revenue loss sit upstream, totally apart from post-bill denial management. The SaaS industry went through this. Companies built on recording and reporting got pushed by public markets to move beyond organizing work to eliminating problems. The ones that couldn't make that shift got repriced. AI is making it possible to move the intervention point from post-bill denial management to pre-bill prevention, flagging documentation gaps before a physician closes an encounter and catching auth mismatches before pre-claim. When that shift happens at scale, the percentage-of-collections model will be much harder to defend. bit.ly/4d7Lodr
1
7
A hospital's MA denial rate can stay flat while MA reimbursement drops. MA plans are paying approved inpatient stays at observation-level rates, and the remit codes it as a contractual adjustment. No denial record, no workqueue. Here's how to find those underpayments: bit.ly/4AJxrf2
3
Ohio just cut the payer clawback window from 24 to 12 months. More takebacks become appealable. But only if you catch them. Most rev cycle teams still check recoupment windows by hand, a few claims at a time. bit.ly/4eM7pPM
2
11
Also, the window your team should check against differs by plan type within the same payer. Sift does that mapping (state laws don't).
4
Most workqueues treat hospital denials populations like they're flat, working the oldest accounts, or the largest balances, and assuming recovery will follow proportionally. This is not supported by any data. When you score a denied A/R population by actual *propensity to overturn*, roughly 75% of recoverable dollars concentrate in the top 15–20% of accounts. The bottom half of the worklist produces almost nothing, regardless of how much time is spent on it. That distribution should matter to revenue cycle leaders, in terms of how teams prioritize, how outsourced vendors get directed, and how much (and how fast) revenue is recovered. bit.ly/4iPOUN9
6
$48B in revenue leakage in 2025, +25% from the year before. Revenue cycle teams are catching problems at the most expensive point in the claims lifecycle. And it's the same breakdowns, at the same predictable points, in system after system, like: -Authorization gaps caught post-discharge -Medical necessity documentation that doesn't meet payer adjudication logic -DRG downgrades processed as contractual adjustments and never flagged -Denial queues worked in a standard order without prioritization Here's a detailed look at 7 common claims lifecycle gaps that drive revenue loss, covering where they live, why they're hard to see, why they're expensive, and what it takes to catch them before they become write-offs. bit.ly/4C1g0Ik
1
10
The Peterson Health Technology Institute released a report last month, following BCBS's coding intensity analysis, with similar messages about AI and healthcare costs. Plans say they see coding intensity jump the day a hospital turns on a tool, and some are already flagging providers with their own algorithms (but how those work isn’t well documented). For revenue cycle leaders, this is a tough position. Your payers can see things about your billing that you can’t. You’re being compared to peers in a model you can’t see, based on documentation that may be perfectly good. Knowing where you’re exposed before a downgrade or takeback letter is increasingly important to protect revenue. bit.ly/4iRzhTQ
1
43
Prior auth volume is down 11%. Rev cycle leaders are now asking about payer takebacks, DRG downgrades & underpayments that post as paid. None of it shows up in a denial report. We mapped where the losses land.
1
14,211
Provider AI use in claims fell from 14% to 9% this year. But payers kept automating. And a denial rate only counts what got denied, not downgrades, short pays & payer takebacks. The gap is bigger than a denial dashboard or standard #Epic report shows. bit.ly/3WEnoJx
1
13
1) Most of what health systsems and hospitals call denial management is really denial maintenance; 2) denial management and EHR tools were built for a payer environment that no longer exists, and the vendors selling them are priced to keep the problem manageable rather than fully fix it; 3) for health systems, #AI changes the question from "how do we work denials faster" to "how do we stop generating them" bit.ly/4yYcPie
15
That payer-reported payer report (eye roll) on how hospital AI "coding intensity" cost Blue plans $942M over two years is also a map of upcoming DRG downgrades.
53
Most hospital revenue cycle teams already have *some* AI tools. Copilot likely came bundled with your Microsoft 365 agreement, and for many teams, it's been sitting there, mostly idle. We wrote a prompt guide to get your team started. 20 prompts you can run in Copilot (Excel, Word, PPT and chat) against the denial exports and AR files your team already pulls: denial concentration by payer, appeal letters, CFO briefings, payer negotiation prep. The guide also helps identify where general AI tools stop reading your denial data, and where deeper Payments Intelligence (predictions, scoring, prevention, actionability) begins. As payers lean harder on AI to deny, downgrade, and delay, this Payments Intelligence layer is turning from nice-to-have into necessary. But, start with the prompt guide first. bit.ly/4rG2t47
34
It's absolutely crazy that payer policy changes show up as lost revenue before you know about them. Or that new downgrade patterns emerge after months of missed reimbursement. Or takebacks come 1+ years after reimbursement. Health systems are losing $$ month after month without real payments intelligence. The data is available, but it does take real work (joining clinical inputs to financial outcomes) and real modeling to get an actionable view. bit.ly/4hw7w2p
8
Plenty of hospital revenue cycle teams feel they've got denials analytics handled. The dashboards are there, sorted by payer and reason code, updated daily. But... Knowing a claim denied isn't knowing *why* it denied. And knowing why isn't *fixing* it in time. Reports don't recover revenue. The work that actually improves net revenue is getting to root cause, turning it into a specific action, and executing before a claim is submitted. bit.ly/47ivKsd
12
Payers are using AI to downgrade claims at scale. Hospital teams are using best-practice PDFs to stop them... When you stress-test AI models across 700+ DRGs and ask it for denial-prevention recommendations, sometimes the results get pretty imaginative, think hallucinated ICD-10 descriptions and invented length-of-stay numbers. That's the problem with generic denial intelligence. It treats every DRG like the average DRG. Payers don't do that. They downgrade specific codes in specific ways. We did the painstaking work of building DRG-specific playbooks that cover the real denial pathways, clinical indicators, documentation flags, and recommendation criteria. We have 320+ MS-DRG playbooks covering 770+ MS-DRGs inside RevProtect's Actionability Engine. Payers already built this depth into the AI that flags and downgrades at scale, but providers are mostly working off "best-practice" advice. These playbooks are part of how Sift closes that gap to drive real denials prevention for health systems. bit.ly/4rxpzdf
14
Payer takebacks hit well after claims adjudicate, months later. The median is ~170 days out, with a tail past 500. This is after revenue cycle teams have reported the period as collected. A recoupment that surfaces 200+ days after service changes a net collection rate nd can re-age a resolved account back into AR>90. Takebacks don't show up like traditional denials. These claims were paid, then partially unpaid. Sift's Denials Insights Team put together a Deep Dive on post-payment takebacks, covering how the lag distorts metrics and how to surface, recover, and forecast them (including the out-of-window recoupments). bit.ly/4yfnVzh
9
"Not having an AI strategy is still a form of AI strategy." In the #revenuecycle this sounds like, "it's on the EHR's roadmap." Most revenue cycle leaders acknowledge AI belongs in their workflows. The harder question is what counts, and what actually delivers results. Because the easy, EHR has it covered version is everywhere. It feels like the answer or progress, but a plan to have a plan isn't adoption, and the gap it leaves is growing. While providers are planning, payers are running AI at full speed, and denials (+ downgrades, takebacks and underpayments) keep accelerating. Claims are coming back in patterns that shift faster than manual processes or even rules-based systems can manage. Every quarter spent on AI planning (or roadmap waiting), instead of real capability, is $$$$ lost, not saved. Doing nothing, or doing the performative version of something, is the genuinely expensive path. Health systems making progress with AI have clarity and urgency. They can tell you exactly which Adverse Payment Outcomes they're targeting, which payers, and which claim types. Then they back it with real AI solutions that hold up against payers who are getting more sophisticated all the time. The gap doesn't close by having a plan to make a plan. "On the EHR roadmap" doesn't recover a dollar. Knowing what you're solving, and putting something real against it, does. (Cartoon and opening quote by the always on-point Tom Fishburne the @marketoonist)
1
1
20