Council Post: Security, Structure And Trust: The New Rules Of Healthcare AI Adoption
By Jeffrey Sullivan, Chief Technology Officer for eFax.gettyAI continues to take the world by storm, but nowhere is its impact more pronounced than in healthcare. Our industry has adopted AI solutions at an acc...
By Jeffrey Sullivan, Chief Technology Officer for eFax.
getty
AI continues to take the world by storm, but nowhere is its impact more pronounced than in healthcare. Our industry has adopted AI solutions at an accelerated pace compared to others, and the applications are vast: from clinical decision support and ambient documentation to automated denial appeals and back-office coding assistance. Half of healthcare leaders report already implementing a generative AI solution, while 57% of C-suite healthcare executives rank AI as their top technological priority, surpassing even EMRs.
Unfortunately, despite the fervor of healthcare’s AI adoption, many organizations struggle to see the ROI of these new tools. One MIT report found that 95% of respondents saw no measurable ROI from their AI investments. Meanwhile, increased utilization has led to heightened scrutiny, where oversight has unearthed risks related to unstructured data, opaque models and unsafe vendor practices that healthcare leaders can't afford to ignore.
Simply put, healthcare’s AI honeymoon is over. Following the initial investment boom, intelligent, automated tools have become table stakes across the industry, ushering in the new phase of healthcare’s AI journey: normalization.
Now, leaders must understand that adoption is only the first step on the road to operational success, and this next phase requires a strategic mindset shift from utilization to outcomes. Organizations that succeed in the next era of AI will more precisely measure impact, scrutinize and operationalize data security measures, prioritize traceability and trust and take strategic steps to limit tech debt.
Moving Past Adoption To Outcomes
When many solutions are first onboarded, KPIs might include adoption rates, saturation metrics or even the number of transactions on the system. But even with strong adoption, many AI pilot programs are quickly abandoned. Too many health organizations aren't seeing strong ROI because they're focused on adoption, not outcomes. As AI enters its normalization phase, it's not enough to say "this feels like it's working" or "people like it."
Healthcare leaders must quantify direct gains from a tool, and if it's not delivering, figure out why and fix it. Instead, far too many organizations are simply replacing what appears as a lackluster investment with a new AI tool—one that will likely provide the same returns if the root challenges aren't corrected.
Confident Mistakes: The Danger Of Bad Data
Many healthcare organizations struggle with AI implementation primarily because healthcare has a prohibitively complex data ecosystem. Around 80% of healthcare data is unstructured and, therefore, challenging for many AI tools to read, parse and integrate into a patient record or workflow. Items like imaging and scans, handwritten referrals or PDFs can easily be skipped by AI algorithms not trained to handle unstructured resources.
Simultaneously, AI is incredibly successful at sounding confident. The result is an "incomplete data, absolute truth" illusion, where AI receives fragmented inputs, hallucinates to fill in the blanks and confidently presents a next step, potentially leading to wrong—even dangerously wrong—outputs.
Ensuring your tools are working with the strongest set of structured data possible is key. Not only does this allow for traceability, but it aids user trust by allowing them to quickly see where an insight came from and if it's accurate. And it starts with the data.
Security, Scrutiny And Why Governance Matters
AI’s adoption in healthcare quickly outpaced compliance frameworks. As a result, many tools or add-on AI solutions were implemented prior to newly adopted AI governance committees and rules. Though healthcare saw a boom of innovative tools from startups in the past two years, many didn’t fully understand how to develop for healthcare’s highly regulated industry.
Unfortunately for healthcare organizations, it means they may be considering, adopting or actively working with technology solutions that aren’t aligned with security and data protection standards. These organizations may be using healthcare and patient data to train their products without realizing the danger. This type of innovation-first, security-second approach cannot continue. Healthcare organizations should be carefully selecting tools with HIPAA, HITRUST and other certifications, as well as signing standard BAA agreements to ensure sensitive information is properly protected.
Where Rip And Replace Actually Makes Sense
Healthcare organizations are gun-shy of the rip-and-replace process, mostly because of the notoriously hefty lift that goes into replacing an EMR, and because many healthcare tools have specific features and functionality not found elsewhere. This, alongside M&A activity and the proliferation of digital health tools, means that healthcare organizations have immense technological debt.
Often, when healthcare leaders bring a new solution in, it may not cover every use case, so the previous solution is kept around to accommodate a few fringe cases. Instead of solving a problem, this scenario simply complicates it by adding another solution on the existing tech stack.
New intelligence and automation tools should fully take the place of legacy manual processes. For instance, many healthcare organizations historically had on-premises fax machines. Though it’s hard to let go of well-known legacy processes, leading digital partners are willing and able to configure solutions to cover organizational needs to ensure they can truly move into the next level of digital sophistication, not just adding another tool to the already unstable tech stack.
Determining Whether You're Ready For The Next Phase Of AI
Healthcare’s AI journey has moved faster than anticipated, but the overarching lack of ROI isn’t a warning sign—it’s a maturity signal. To ensure your organization is built for this next phase, focus on these four actionable strategies:
• Measure outcomes, not adoption. Focus strictly on quantifiable efficiency gains—like faster referral responses or reduced costs—rather than tracking transaction volumes.
• Prioritize data quality. Ensure AI tools work with highly structured data to eliminate dangerous "hallucinations" and build trust with clinical staff.
• Enforce strict security standards. Conduct comprehensive vendor assessments to ensure full compliance with evolving security and data protection regulations.
• Eliminate tech debt. Fully "rip and replace" manual legacy processes (like paper faxing) rather than stacking new AI tools on top of outdated workflows.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?