Turning Artificial Intelligence into Business Intelligence
What happens when artificial intelligence (AI) is wrong? It happens more often than you might expect, especially with the influx of new technology vendors who lack a deep understanding of the complete clinical data lifecycle. Many AI tools are built in silos without a true understanding of the critical downstream impact of offered suggestions, such as how code modifiers can either boost reimbursements or trigger claim denials when misapplied.
Make no mistake, AI is ready for deployment. But like any tool, it must be managed. That’s where HI professionals are essential. They are the ones who must ensure accuracy while optimizing how the data is used for clinical care support and revenue cycle management (RCM), balancing the needs of all stakeholders from document creation through collections and compliance.
AI generates and routes clinical data far faster than humans. Administrators not using AI risk falling behind on key performance indicators (KPIs) like documentation turnaround time, discharged, not final billed (DNFB), and denial rates—all of which directly impact cash flow. Despite impressive demos and return on investment (ROI) projections, real-world outcomes can vary greatly, especially when applied across different specialties.
When adopting AI, HI professionals should examine how the new technology will add value to the data generated in their clinical documentation, coding, clinical documentation improvement (CDI), and billing processes. Although a complete list of non-clinical applications of AI would read like an encyclopedia, the following list highlights a notable scope of offerings, the undeniable processing speed, and consistency advantages they bring to the workflow, along with KPI, vendor, and health information management (HIM) team member considerations healthcare organizations should evaluate when making a purchasing decision and implementing their selected AI products.
Clinical Documentation
Advantages
Ambient Speech AI captures conversations between patients and providers, transcribes them using speech recognition, and generates drafts based on the history of similar reports in their large language models (LLMs). These drafts reduce the providers documentation burdens and associated burnout while improving coding suggestions, particularly for E&M levels.
KPI Considerations
Speed is impressive, but quality is critical. AI tends to “hallucinate” details—inferring content based on typical patterns in its LLM training data. These fabricated points can lead to inaccurate or even fraudulent coding if providers don’t review documents carefully.
Cost
Priced monthly per provider. Generally, more cost-effective than traditional transcription or Live Scribe services.
Quality
Performance varies by vendor and specialty. Hallucinations are common, especially in complex cases. Engines perform better when built on large language model LLM with a huge volume of reports in their training datasets.
Turnaround Time
Typically, one to two minutes for standard notes; up to 10 minutes for complex visits (e.g., orthopedic or psychiatric).
Vendor Considerations
Hallucination levels vary significantly, often driven by the different size of their core LLM datasets, which can vary from 20,000 reports to millions. Accordingly, it’s important to verify specialty-specific performance before making a purchasing commitment. Contract terms can be restrictive, and some leading brand names offer lesser quality based on head-to-head comparative studies by nationally recognized healthcare institutions.
Team Considerations
High-speed delivery of reports that read well can lull providers into complacency. HIM-led audits are crucial to catch errors. As AI continues to evolve rapidly, it may be wise to avoid exclusive long-term contracts with any single vendor.
Coding
Advantages
AI accelerates coding for routine, repetitive cases (e.g., radiology, ED). It boosts productivity and allows coders to focus on complex cases. While not a replacement for human coders, it excels as a speed-enhancing productivity assistant.
KPI Considerations
AI will soon dominate fee for service (FFS) coding. Coders for these cases will increasingly serve only as auditors and editors, while manual coding will continue to be used for complex cases due to the exponentially higher volume of potential variables.
Cost
AI-assisted coding with human auditing is cheaper than full manual coding. However, misapplied AI in complex cases can cause costly underbilling or overly aggressive severity levels or Hierarchical Condition Classification (HCC) errors.
Quality
AI must be audited by credentialed coders. Providers should not have to shoulder this burden as it has been proven by study after study to contribute to burnout and inconsistency.
Turnaround Time
Instantaneous recommendations with improved workflow speed when auditing is managed properly.
Vendor Considerations
Claims of accuracy vary. Sample sets submitted for evaluation during an RFP almost always include human oversight. References from provider organizations offering similar specialties and case mixes are crucial. Flexibility in integrating with existing workflows is also a key differentiator.
Team Considerations
Coders must evolve into AI editors. Demonstrating precision in coding complex cases with accuracy for modifiers and HCC identification will help justify their personal value. Providers should never be expected to audit AI coding output.
Clinical Documentation Improvement (CDI)
Advantages
AI identifies documentation gaps through point-of-care and retrospective audits. These tools enhance coding accuracy and reimbursement by highlighting missed or inaccurate data.
KPI Considerations
AI’s flexibility can either highlight too many issues (inflating ROI projections) or too few (to try and avoid alert fatigue). Market payer mix calibration is essential to strike the right balance.
Cost
Greatly reduces costs for large-scale audits, especially for HCC retrospective reviews.
Quality
Performance should be judged by actual denial reductions and audit results from comparable provider specialty references, not vendor claims.
Turnaround Time
Recommendations are instant but still require skilled human auditing. Net throughput is improved when implemented properly.
Vendor Considerations
Training data volumes and natural language processing (NLP) engine quality vary more here than in any other AI area. Ask prospective vendors about the size of their LLM and data sources.
Team Considerations
CDI specialists should retool themselves as AI-enhanced auditors, tailoring AI outputs to match organizational and payer-specific expectations.
Billing
Advantages
AI helps improve billing accuracy by automating tasks, flagging errors pre-submission, and leveraging payer-specific rules to predict outcomes through pre-adjudication. It enhances financial performance by reducing denials and submission delays.
KPI Considerations
AI can preempt denials due to missing pre-authorizations, unsupported documentation, or flagging of potential fraud. It optimizes revenue cycle accuracy before submission.
Cost
Costs vary widely between vendors based on the scope of their offerings but generally offer strong ROI. It’s important to compare vendors by claim-type performance and rejection rates.
Quality
Billing accuracy depends on upstream coding quality and team expertise. Overly uniform claims (e.g., copy-paste submissions) can trigger suspected cloning rejections unless balanced by strong differentiators like supporting SDOH documentation.
Turnaround Time
Edits are instant, but resolution time is necessary. Teams must review and validate edits thoroughly to avoid downstream problems.
Vendor Considerations
Feature sets, pricing, and workflow flexibility differ widely. Look for alignment with your existing processes and seek references from similar payer mixes.
Team Considerations
Billing AI applications reduce the manual work but doesn’t eliminate the need for audits. Giving billers time to resolve conflicts—even at the cost of slight DNFB increases—will enhance overall revenue performance.
Overall Considerations
AI products are being developed and launched faster than ever—often before they’re truly fully field-tested. Demos reflect ideal scenarios while real-world results are often far messier. Specialty-based pilots are essential before scaling and the only guarantee you can count on is ‘your results may vary.’
At the same time, AI is undeniably transformative. It improves speed and consistency, while offering structured data insights at a scale the industry has never seen. But these benefits only translate into true business intelligence when orchestrated by HI professionals who understand the nuances of clinical data beyond individual discipline silos and how that data is initiated and consumed by the different stakeholders while simultaneously justified against overall compliance demands.
Yet, with all the industry hype and pressure to adopt AI, it’s important to recognize it is not a magic wand. Like the earlier move to certified electronic health records (EHRs), it’s a productivity improvement tool that presents more structured data—not a plug-and-play human replacement.
History has shown us reducing staff too fast, post-technology implementation inevitably leads to an explosion of errors passed downstream. However, AI is effectively shifting the primary role of HI professionals from data production to data analytics. HI team members are still needed to convert the increased data into true business intelligence by ensuring it’s accurate and properly optimized for the complete workflow.
And speaking of the complete workflow, any vendor who suggests you need to adjust yours to take full advantage of their offerings don’t respect how the US healthcare industry operates and more probably than not will end up like the 97 percent of certified stage one EHR vendors who are no longer in business.
Ultimately, only HI professionals can translate raw AI output into meaningful and actionable business intelligence that serves all the clinical data stakeholders in your organization, demonstrating their irreplaceable value in a world where faster, more consistent data alone is not enough.
Dale Kivi, MBA, is founder of Mercenary Marketing and has represented health information service and technology firms for over 30 years. A frequent HI author and speaker, he currently serves on the AHIMA Non-Clinical AI Community of Practice Work Group.