Hospitals bought AI the wrong way — and now they're paying for it twice. McKinsey's August 2026 report delivered the diagnosis bluntly: most health systems "looked at AI with a bolt-on mindset rather than a transformation mindset," creating a proliferation of point solutions and pilots without the integration or scale to create enterprise value. The consequence? HFMA reports that hospitals are now actively trimming vendors from their AI rosters as ROI clarity exposes which tools actually work and which just add complexity.
This isn't a minor correction. It's a structural shift in how hospital CFOs, CIOs, and revenue cycle leaders evaluate AI investments. The era of buying five point solutions for five workflow problems is ending. What's replacing it is the question that determines whether your revenue cycle improves or stalls.
McKinsey's Verdict: Bolt-On AI Is Failing Hospitals
The McKinsey report — published via HFMA under the title "The health system CEO imperative: Turning AI's promise into performance" — identifies a specific failure pattern. Hospitals adopted AI tools to automate individual tasks: a prior authorization bot here, a coding assistant there, a denial management dashboard elsewhere. Each tool solved a narrow problem. None of them rethought the process.
"The productivity improvement, the productivity unlock comes from not just changing a piece of a process but actually rethinking the whole process from beginning to end." — Jessica Lamb, McKinsey Partner
This is the core distinction between bolt-on AI and transformational AI. A bolt-on prior auth tool automates form submission. Transformational AI asks why the prior auth was needed in the first place, whether the eligibility check upstream could have predicted the requirement, and whether the clinical documentation already contains the medical necessity evidence the payer needs — before anyone submits anything.
The "proliferation of point solutions and pilots" McKinsey describes creates a compound problem: each tool requires its own integration, its own data pipeline, its own maintenance overhead, and its own vendor relationship. Multiply that by five, eight, or ten AI vendors, and the management overhead begins to consume the productivity gains each tool was supposed to deliver.
HFMA Confirms: Hospitals Are Trimming AI Vendor Rosters
The HFMA reporting validates McKinsey's diagnosis with ground-level evidence. Kevin Coloton, CEO of HURC, stated directly that "lack of ROI on AI implementations is driving much of the retraction of vendors at health systems." Hospitals aren't cutting AI — they're cutting AI vendors that don't deliver measurable returns.
The trim isn't random. Two filters are driving vendor selection:
Filter 1: The Epic-First Evaluation
Organizations increasingly state they are "Epic-first" — if Epic's native functionality can handle a workflow, third-party vendors get eliminated from consideration. This raises the bar for every AI vendor: you must either do something Epic can't do, or do it measurably better than Epic's built-in tools.
But Epic-first has limits. TailorMed CEO Srulik Dvorsky noted: "There are going to be many, many different workflows and tasks that Epic is not at the moment or maybe even in the future going to go into." The vendors that survive the Epic-first filter are those covering workflows Epic won't — complex multi-payer denial management, payer-specific prior authorization logic, and cross-system revenue cycle orchestration.
Filter 2: Integration Depth Over Feature Count
The second filter is integration quality. Vendors that don't demonstrate tight Epic integration — or tight integration with whatever EHR the hospital runs — are the first to go. A standalone denial management platform that requires manual data export from Epic, analysis in a separate interface, and manual action back in Epic creates more work than it saves. The integration tax kills the ROI.
The Data: Why Point Solutions Can't Solve the #1 RCM Problem
The HFMA 2026 Revenue Cycle Benchmark Report — a survey of 102 revenue cycle leaders — quantifies why the point-solution approach fails. The findings:
- Denials and appeals remain the #1 RCM challenge
- 55% of providers say claim errors are increasing (up from 44% in 2022)
- $20 billion per year in denied claims across US hospitals
- Growing reliance on AI-enabled workflows, automation, and outsourcing partnerships
- Staffing shortages continue to disrupt operations and innovation
Here's what matters about these numbers: denials aren't a single-point problem. A denial originates at eligibility verification, gets compounded by documentation gaps at the point of care, survives claim scrubbing because the scrubber doesn't have clinical context, and finally lands as a rejected claim that a separate denial management tool has to process. No single AI point solution touches all four failure points. An integrated platform does.
The Mid-Revenue Cycle Black Hole
The HFMA Mid-Revenue Cycle Roundtable — sponsored by Waystar — exposed a bottleneck that point solutions structurally cannot fix: the mid-revenue cycle, the bridge between clinical care and financial processes.
Diana O'Connor, Waystar VP, stated it plainly: "Accuracy of clinical documentation is vital to achieving a high-performing revenue cycle." But the roundtable revealed that accuracy fails at specific handoff points that exist between — not within — point solutions:
| Mid-Cycle Failure | What Happens | Downstream Impact |
|---|---|---|
| Siloed Operations | Physician compliance vs. coding documentation disconnect | Claims submitted with incomplete clinical justification |
| Charge Capture Gaps | New clinical services launch without revenue cycle notification | Unbilled services, missing charges, revenue leakage |
| Billing Data Fragmentation | Multiple facilities create separate billing records for one patient | Patient receives multiple bills, satisfaction drops, collections fail |
| Manual Process Persistence | "Spreadsheets, emails, and faxing information" in 2026 | Errors compound at every handoff between manual and automated steps |
Nikki Harper from Mayo Clinic highlighted that patient experience degrades when billing data silos create multiple bills from multiple facilities. Heather Wilson put it bluntly: "It's 2025, and we're still operating with spreadsheets and emails and faxing information." Desmond Jackson added employee satisfaction as a metric — when staff are frustrated by broken handoffs, error rates climb in those exact areas.
Point solutions can't bridge these gaps because they're the gaps between point solutions. Only an integrated approach that maintains a single patient context across the entire revenue cycle eliminates the handoff failures where denials are born.
What Integrated AI Actually Looks Like
If bolt-on AI automates a step and integrated AI rethinks the process, what does the rethought process look like? Here's the architecture that matches McKinsey's transformation thesis:
Stage 1: Eligibility + Authorization as One Decision
Instead of checking eligibility in one system and submitting prior authorization in another, integrated AI treats them as a single decision. The eligibility check returns benefit details, the AI immediately evaluates whether the planned service requires authorization, and if so, assembles the authorization request using clinical documentation already in the system — before a human touches it.
Stage 2: Clinical-to-Coding Continuity
The coding engine doesn't wait for a coder to review the chart. As the clinician documents, the AI validates that the documentation supports the anticipated codes, flags gaps in real time, and ensures modifier usage meets payer-specific requirements. When the note is closed, the claim is already coded — accurately, completely, and with the clinical evidence attached.
Stage 3: Pre-Submission Validation Against Payer Intelligence
Before the claim leaves the building, it's validated against payer-specific rules, historical denial patterns for this payer-procedure combination, and current contract terms. The AI doesn't just scrub for formatting errors — it predicts whether this specific claim will be denied by this specific payer and fixes the root cause before submission.
Stage 4: Denial Prevention, Not Denial Management
The critical distinction. Point-solution denial management processes denials after they arrive. Integrated AI prevents denials before they happen by catching the upstream failures — eligibility gaps, documentation deficiencies, coding errors, authorization mismatches — that cause them. The denials that still get through are genuine payer errors, not preventable billing mistakes.
Stage 5: Payment Posting With Closed-Loop Learning
When payments arrive, the AI compares them against expected reimbursement from contract terms, identifies underpayments and partial denials, and feeds that data back into the pre-submission validation engine. The system gets smarter with every payment — not just for the denial management tool, but for every upstream stage that contributed to the claim.
The Vendor Consolidation Decision Framework
For hospital CFOs and revenue cycle directors evaluating whether to consolidate AI vendors, here's the framework that aligns with the McKinsey and HFMA findings:
- Map your current AI vendors to revenue cycle stages. How many tools touch a single claim from scheduling to payment? Each handoff between vendors is a potential failure point.
- Measure integration tax. Calculate the staff time spent exporting data between systems, reconciling outputs from different tools, and managing vendor relationships. This is the hidden cost McKinsey's report identifies.
- Apply the Epic-first filter honestly. Which workflows does Epic handle well enough? Which genuinely require specialized AI? Be ruthless — every vendor that survives this filter must deliver measurable ROI above what Epic provides natively.
- Evaluate end-to-end context sharing. Can your AI tools share patient context across stages? If the denial management tool doesn't know what the eligibility check found, you're running point solutions with extra steps.
- Prioritize platforms over best-of-breed. McKinsey's finding is unambiguous: "rethinking the whole process from beginning to end" requires a platform that sees the whole process, not individual tools that each see one step.
"Lack of ROI on AI implementations is driving much of the retraction of vendors at health systems." — Kevin Coloton, CEO, HURC
The Consolidation Math
Consider a mid-size hospital system running six AI point solutions across eligibility, prior auth, coding, claim scrubbing, denial management, and payment posting. Each vendor costs $50,000-$200,000 annually. Each requires integration maintenance, staff training, and vendor management overhead. Total cost: $500,000-$1.2M per year in direct vendor spend, plus significant FTE allocation for coordination.
Now consider that 55% of providers in that system report increasing claim errors despite deploying these tools. The point solutions are running, the integrations are maintained, the staff are trained — and errors are still going up. That's the bolt-on mindset failing in real time.
An integrated platform that replaces four to six vendors with one coordinated pipeline typically costs less in total vendor spend while eliminating the integration tax entirely. More importantly, it addresses the root cause the HFMA benchmark identified: errors compound across handoffs, and the only way to eliminate handoff errors is to eliminate handoffs.
The hospitals that moved first on consolidation aren't managing fewer vendors as an administrative convenience. They're preventing denials that point solutions — by architectural design — could never prevent. With $20 billion in annual denied claims and claim errors still accelerating, the question isn't whether to consolidate. It's how fast you can get there.
BAM AI's integrated platform connects eligibility, authorization, coding, submission, denial prevention, and payment posting into one pipeline — with shared patient context at every stage and zero handoff gaps. The vendor consolidation wave is here. The question is whether you're consolidating toward a platform or still managing the sprawl.