Forecasting is cheap. Every one of these is anchored to research somebody published with a methodology, and each comes with something concrete to do before December.
Revenue recovery trends in H2 2026 are being shaped by five pressures: AI adoption is no longer a differentiator, contact-channel compliance risk is rising, denials are getting harder to overturn, HIPAA/security requirements are tightening vendor viability, and vendor strategy is shifting toward outsourcing and consolidation. The practical takeaway is to replace capability claims with evidence, test revocation and compliance workflows, and focus on vendors who can handle the operational and regulatory load.
Three of the five are uncomfortable, and those are the useful ones.
1. The AI adoption number stops meaning anything, and the failure rate starts to
TransUnion’s seventh annual Debt Collection Industry Report found 93% of debt collection companies now use or plan to deploy AI or machine learning, up from 73% a year earlier. Only 7% report no plans.
Gartner predicts more than 40% of agentic AI projects will be canceled by the end of 2027, naming escalating costs, unclear business value, and inadequate risk controls. MIT’s NANDA report, published August 2025, found roughly 95% of corporate generative AI pilots delivered no measurable P&L return. That MIT figure has been contested on methodology, so treat it as directional and read Gartner’s as the sturdier of the two.
Put those together and the capability stops being a reason to buy. Nearly everyone has the model. Most won’t operationalize it.
The uncomfortable part is what that does to your vendor evaluation. If 93% of your panel can say “we use AI,” the answer carries no information, and any RFP question phrased as “do you use AI?” is generating noise.
Replace the capability question with an evidence question. Ask a vendor to reproduce a decision from six months ago: which model version, which inputs, which policy version, which human touched it, and how long the reconstruction takes. Ask for the model inventory, the independent validation record, and the override log. A vendor who answers with architecture and benchmark scores has answered a different question than the one you asked.
2. Your cheapest contact channel is your largest uncapped liability
Most operators haven’t repriced the asymmetry between the two statutes that govern their outreach.
FDCPA class action statutory damages are capped at the lesser of $500,000 or 1% of the collector’s net worth, plus up to $1,000 for each named plaintiff, under 15 U.S.C. § 1692k. That’s a bounded number you can reserve against.
TCPA damages are $500 per violation, trebled to $1,500 for willful or knowing conduct, with no aggregate cap at all, and the count runs per call and per text.
In February 2024 the FCC ruled that AI-generated voices in calls are “artificial” voices under the TCPA. Under that ruling, an AI-generated voice call placed without the required prior express consent falls inside the TCPA’s artificial-voice restrictions. The FCC’s revoke-all provision, which makes a consumer’s revocation on one message type apply to all future robocalls and robotexts from that caller, now carries a compliance deadline of 31 January 2027 under FCC Order DA 26-12.
The filing trend is moving with it. WebRecon counted FDCPA federal filings up 7.8% and FCRA filings up 37.4% in 2025, with CFPB complaints up 89.1%. TCPAWorld, a trade tracker, reported 330 TCPA cases filed in April 2026 with 255 of them class actions, up 40% year over year.
So the voice agent that shaves two dollars off cost-per-contact is taking on telecom-statute risk, which is uncapped, per-event, and compounding faster than any other category of consumer litigation.
Run a revocation propagation test before you expand any automated channel. Have someone revoke on one channel and time how long it takes for that revocation to reach every other channel, every vendor on the panel, and every dialer. If the answer is measured in days or depends on a nightly file, you have a 31 January 2027 project and you should start it this quarter.
3. Denials get worse before they get better, and the constraint is people
Kodiak Solutions’ March 2026 benchmarking across roughly 2,300 hospitals found initial denial rates rose to 11.6% in 2025 while the rate at which providers overturn initial denials fell to 42.1%. Net revenue lost to final denials and bad debt hit $48.4 billion, up 25% in a single year.
Two things push this further in the second half. Experian Health’s 2025 survey found 90% of denials require at least some human rework, and 68% of providers say submitting clean claims is harder than it was a year earlier. And Holland & Knight counted six states enacting laws in 2026 that bar an insurer from relying on AI as the sole basis for denying care, with effective dates running through 2027. Alabama, Indiana, Maryland, Utah, Georgia, and Washington are all on that list.
Those state laws are good policy and they will slow things down. Payers adding required human review to utilization decisions changes cycle times in both directions, and providers who staffed their appeal function on last year’s volume will feel it.
The evidence says the denied volume is worth fighting for. Premier’s survey of 280 hospitals found roughly 70% of initially denied claims eventually get paid. A JAMA study of roughly 51,000 New York external appeals, published April 2026, found overturn rates rose from 38% in 2019 to nearly 53% in 2025. KFF found fewer than 1% of denied ACA marketplace claims are ever appealed. Both the Premier and JAMA populations are self-selected, and both point the same direction: a large volume of denied claims was payable.
Your appeal abandonment rate for the last four quarters is the number to pull first. Then put your cost per denial into the denial appeal break-even calculator to see which abandoned claims were worth an appeal. Put a dollar value on it using Premier’s $57.23 administrative expense per claim as the cost side and your own average claim value as the recovery side. The result usually justifies the headcount conversation on its own.
4. Half your vendor panel can’t afford the compliance bill that’s coming
HHS Office for Civil Rights proposed a HIPAA Security Rule overhaul that eliminates the “addressable” category, so every implementation specification becomes mandatory. It requires encryption of all ePHI at rest and in transit, multifactor authentication across all systems, continuous monitoring, annual risk analysis across every system touching ePHI, vulnerability scanning, penetration testing, and business associate notification to covered entities within 24 hours.
OCR priced its own proposal at $9 billion in first-year compliance cost across covered entities and business associates.
A forty-person agency holding a health system’s PHI has no share of $9 billion. That’s a barrier to entry stated by a regulator in dollars, and it has nothing to do with model quality or service levels.
The oversight machinery pointed at those vendors already stopped working. EY’s 2025 global third-party risk survey found 43% of companies sending an average of 55 questionnaires per assessment, while RiskRecon found that 4% of organizations have high confidence those questionnaires reflect real-world risk. EY also found that organizations with third-party risk programs less than three years old actively manage a median of 275 third parties, while programs more than ten years old manage 80. Either maturity means consolidation, or maturity means giving up the pretense of monitoring the long tail.
Meanwhile Verizon’s 2026 DBIR put third-party involvement in confirmed breaches at 48%, up 60% year over year, and the Financial Stability Oversight Council named third-party service provider risk among its systemic concerns in its December 2025 annual report.
Ask every vendor on your panel what their plan is for the proposed Security Rule requirements, specifically the 24-hour notification obligation and continuous monitoring. Get it in writing. The answers will sort your panel faster than any questionnaire, and the vendors who can’t answer are the ones you’ll be explaining to a regulator later.
5. Vendor strategy is moving, and the direction correlates with performance
McKinsey surveyed 215 US healthcare leaders in September 2025. 60% planned to change their revenue cycle vendor strategy within two years, and three-quarters of those were expanding outsourcing rather than pulling work back in.
Preference splits by performance. Among organizations exceeding their revenue cycle goals, 59% preferred expanding with existing vendors against 41% adding new ones. Among organizations underperforming against their goals, 68% preferred new vendors against 32% expanding with existing ones.
Organizations hitting their revenue cycle goals expand with the vendors they have. Organizations missing them add new ones. The correlation runs both directions, both groups are in the market right now, and it’s worth knowing which one you’re doing and why.
McKinsey found one more thing. Among leaders considering bringing revenue cycle functions back in-house, 75% cited vendor performance management as a key need. Three-quarters of would-be insourcers are saying the problem is their ability to manage the vendors they already have.
Switching carries a cost that rarely makes it into the business case. Every switch resets the learning curve, re-runs the integration, restarts the diligence lifecycle, and adds a fresh set of handoffs. That cost is worth paying when the vendor is the problem, and worth naming when it isn’t.
Before the next RFP, run a diagnostic on your own side of the relationship. Do you give vendors matched placement files? Do you have champion-challenger data you could actually defend? Do you review performance on a cadence, with a defined escalation path? If the honest answer is no, a new vendor will underperform for the same reasons the last one did.
What ties these together
Four of these five are governance problems wearing operational costumes. What fails in each case is the missing control layer around the model and around the vendor, and that layer is something you can build and buy on purpose.
We’ve argued the longer version in three papers: The model is the commodity on what you’re actually buying when you buy AI in regulated recovery, The rate is the smallest number on the page on total cost of recovery, and The seam is the failure point on what’s breaking in multi-vendor models.
Sources: TransUnion, seventh annual Debt Collection Industry Report; Gartner (25 June 2025); MIT NANDA (Aug 2025); 15 U.S.C. § 1692k; 47 U.S.C. § 227; FCC Declaratory Ruling 24-17; FCC Order DA 26-12; WebRecon via Troutman Pepper Locke; TCPAWorld; Kodiak Solutions (March 2026); Experian Health State of Claims 2025; Holland & Knight; Premier Inc.; JAMA (April 2026); KFF (March 2026); HHS Office for Civil Rights HIPAA Security Rule NPRM, cost estimate via Alston & Bird; EY 2025 Global Third-Party Risk Management Survey; RiskRecon; Verizon 2026 DBIR; FSOC 2025 Annual Report; McKinsey & Company (Sept 2025).
Revenue recovery trends in H2 2026 are being shaped by five pressures: AI adoption is no longer a differentiator, contact-channel compliance risk is rising, denials are getting harder to overturn, HIPAA/security requirements are tightening vendor viability, and vendor strategy is shifting toward outsourcing and consolidation. The practical takeaway is to replace capability claims with evidence, test revocation and compliance workflows, and focus on vendors who can handle the operational and regulatory load.
- AI use is widespread, so buyers should ask vendors to prove operational control, validation, and auditability instead of claiming AI adoption.
- Cheapest outreach channels can create the biggest uncapped legal risk, especially where revocation handling and artificial-voice rules apply.
- Denials, vendor consolidation, and stricter security expectations are pushing revenue recovery teams toward more human review, stronger controls, and better vendor management.
Frequently Asked Questions
What are the main revenue recovery trends in H2 2026?
The post identifies five: AI claims are losing value as a buying signal, contact-channel compliance risk is rising, denial volumes remain hard to reverse, HIPAA/security requirements are increasing vendor burden, and vendor strategy is shifting toward outsourcing and consolidation. Together, these trends make evidence, control, and operational resilience more important than feature claims.
Why is AI no longer a strong differentiator in revenue recovery?
Because AI adoption is already widespread. The post cites TransUnion’s finding that 93% of debt collection companies now use or plan to deploy AI or machine learning. With capability becoming common, the better question is not whether a vendor uses AI, but whether it can reproduce decisions, show model validation, and provide override logs.
How should buyers evaluate a revenue recovery vendor now?
Replace generic capability questions with evidence-based checks. The post recommends asking a vendor to reconstruct a decision from six months ago, including model version, inputs, policy version, human involvement, and reconstruction time. It also says to request the model inventory, independent validation record, and override log.
What compliance risk is growing fastest in contact channels?
The post says TCPA risk can be uncapped, per-event, and compounding, especially for AI-generated voice calls and robotexts. It recommends running a revocation propagation test before expanding automation: revoke on one channel and measure how quickly that revocation reaches every other channel, vendor, and dialer.
How are denials and vendor strategy affecting revenue recovery?
Denials are getting harder to manage because more require human rework and clean claims are harder to submit, while many denied claims still get paid when appealed. At the same time, many healthcare leaders plan to change vendor strategy, with leaders more likely to expand with existing vendors and underperformers more likely to add new ones.