Think of a huge dollar amount, then quadruple it. That’s how it feels trying to grasp the global scale of AI’s potential in healthcare. It’ll hit a head-spinning half-a-trillion dollars by 2032, according to the World Economic Forum. In the US alone, McKinsey reckons that AI-driven automation and analytics could slash hundreds of billions from annual healthcare spending.

Despite this, AI adoption in healthcare remains stubbornly slow – held back by a range of barriers, from concerns over accuracy and data security, to AI’s integration with legacy systems and the challenges of scaling. But with rising pressure to improve patient outcomes, cut costs, and ease the crushing administrative burden on healthcare professionals, these concerns must be tackled head-on.

AI in healthcare needs a new playbook – one built around starting small, curating data effectively, deploying securely, and leveraging strategic partnerships to accelerate adoption.

Step 1: Start small, build confidence

Focus on low-risk wins that can deliver rapid value while building trust in the technology.

Quick wins are there for the taking – especially in automating the endless paperwork that pulls clinicians away from patients. Tools that use generative AI to draft discharge summaries, for example, or generate medical coding based on clinical notes. Last year, clinical documentation tools reportedly secured four of healthcare’s five largest generative AI investment deals. It's no surprise – these low-risk efficiencies offer rapid ROI while reducing administrative burden, adding up to a clear win for both clinicians and organizations.

Unlike the longer timelines needed to validate AI systems for diagnostics or treatment decision-making, administrative AI offers a fast path to value—reducing workload pressures while streamlining routine processes.

Accelerating AI adoption is generally not about leaping straight into clinical applications. Hospitals store vast amounts of operational information, such as policy documents detailing clinical protocols, administrative procedures, and safety standards. Applying AI to this type of data – often stored in document management systems such as SharePoint – offers a lower-risk, faster route to value. For example, AI can help clinicians quickly search and summarize information from extensive policy databases, reducing the time spent manually combing through documents and ensuring access to the most up-to-date guidelines.

Starting small builds confidence, sharpens workflows, and reduces risk. Once the technology is proven, organizations can expand to higher-risk clinical data, ensuring that lessons learned early on translate into smoother, safer integration down the line.

Step 2: Boost accuracy with smarter data strategies

In healthcare AI, speed is worthless without accuracy. Even minor AI-generated errors can potentially jeopardize patient safety or expose providers to liability. Fast-tracking AI adoption starts with getting accuracy right from the get-go.

General-purpose AI models might achieve, say, 85% accuracy out of the box, but when lives are at stake, it must be closer to 100%. That requires customization with data specific to a given healthcare provider, such as their patients’ electronic health records (EHRs), discharge notes, and lab results. But there’s no need to start from scratch: open-source datasets such as MIMIC-III, which contain de-identified real-world clinical data, offer a fast track to better model accuracy. Such resources help AI systems understand the clinical world faster, reducing the time needed for customization.

Step 3: Streamline data access for faster AI wins

Even the smartest generative AI model is only as good as its training and the additional information it can access. This is where retrieval-augmented generation (RAG) comes in—enhancing AI by pulling relevant, up-to-date information from trusted hospital-data sources to produce accurate, context-aware outputs. Whether it's an infection control policy or key elements of a patient’s medication history, AI must retrieve the right information at the right time. Robust retrieval isn’t just a backend process; it's a critical accelerator for effective, safe AI deployment.

Building retrieval systems for RAG models from scratch is a complex, time-consuming process—often taking experienced AI engineers several months. Legacy systems add to the challenge, with critical and potentially sensitive data siloed across outdated platforms and stored in varying, multimodal formats, from text and charts to images and PDFs. Integrating AI with these fragmented systems demands significant expertise to ensure accurate and efficient retrieval.

That hospitals are competing with big tech for AI talent only adds to the resource constraints. The faster route? Leveraging pre-built, managed solutions that securely process and index your healthcare data. These systems ensure that information stays accurate, up to date, and instantly retrievable. For example, our AI workspace North, is designed to meet the unique security and customization demands of highly regulated and complex industries like healthcare and provides pre-installed connectors with RAG architecture.

But achieving accuracy isn’t just about data processing—it’s about people. In healthcare, perhaps more than any other industry, AI systems work best with a human-in-the-loop approach. This is more than a safeguard; it’s another accelerator that allows AI to handle the heavy lifting while clinicians focus on refining and validating outputs.

Virtual AI assistants offer another fast-track opportunity for AI deployment. By leveraging RAG, these systems can help to address patient queries, answer routine questions with hospital-sanctioned guidance between inpatient visits, or flag a patient’s concerning symptoms to their doctor.

Step 4: Boost security and compliance with private AI deployments

Sensitive patient data is subject to strict regulatory requirements, particularly under the Health Insurance Portability and Accountability Act of 1996 (HIPAA), which mandates the protection of health information. Any AI system handling such data must comply with these standards from day one.

Private deployment—hosting AI models within a hospital's secure environment, whether in a virtual private cloud or dedicated on-premises servers—offers the most robust solution. This approach keeps sensitive data ring-fenced and under organizational control, reducing the risk of data breaches. With cyberattacks on healthcare systems at an all-time high, according to McKinsey, minimizing systems exposure is more important than ever—especially in a sector in which underinvestment has left many providers vulnerable.

Step 5: Accelerate progress through strategic partnerships

Implementing AI in healthcare takes a lot more than technology—it takes a leadership mandate and the resources to execute at scale. Hospitals face a significant resource gap when it comes to AI engineering talent. This is why strategic partnerships are essential. Working with experienced AI providers allows healthcare organizations to access expertise they may lack in-house, avoiding costly missteps and accelerating deployment.

Strategic partners who understand healthcare data, regulatory compliance, and clinical workflows can provide vital guidance—collaborating with hospital IT teams to navigate complex requirements while ensuring AI systems are secure, effective, and scalable.

These partnerships also reduce the financial burden of building AI systems from scratch. Managed solutions minimize upfront investment and allow healthcare organizations to focus on delivering better patient outcomes, not maintaining infrastructure.

Step 6: Focus on workforce training and change management

Even the best AI system will stall without buy-in from the people it’s designed to support. That’s why training and onboarding the workforce are critical to AI adoption. Clinicians and operational staff need to understand not just how to use AI tools, but why they’re being introduced—what problems they solve, how they’ll improve workflows, and how they’ll support better patient outcomes.

New AI learning programs for workers are popping up quickly, like Section, which aims to make everyone an AI expert.

It’s about creating a culture where AI is shown to be an empowering assistant, not a potential replacement.

A new playbook for AI in healthcare

AI is transforming healthcare—but accelerating adoption requires more than overcoming technical challenges. It demands smarter strategies: starting with low-risk data, leveraging strategic partnerships, and ensuring secure, scalable deployments.

This is the smart playbook for healthcare AI: minimize risk, accelerate progress, and maximize impact. At Cohere, we help healthcare organizations move faster and smarter, delivering adaptable, secure AI solutions that free healthcare providers to focus on what matters most: patient care.

Unlock the potential of secure AI.

Get Started!

The AI shift in healthcare: A playbook for technical leaders