Innovation with GenAI: Lessons from Life Sciences and Beyond

Dan Schulman

For a child with autism spectrum disorder (ASD), effective care coordination across the required medical, physical rehabilitation, counseling, occupational, and speech providers is challenging, especially when the aim is to provide individualized care—at scale across many patients.

Genial Care, a healthcare company focusing on personalized interventions for children with ASD, provides a successful example of how generative artificial intelligence (GenAI) can solve this problem. GenAI monitors sessions with children and their families, synthesizes session records, improves the records’ quality, tracks each child’s progress, and shares this information with providers and the child’s family via an app-based interface. Plus, the company reduced the manual transcription time from approximately two hours per day to three minutes per session, allowing care providers more time to dedicate to their patients.

However, this is one of only a few examples of a life sciences or healthcare company breaking out of GenAI pilot purgatory. How did they do it? They focused on a practical, human-centered use case that easily integrated into daily workflows and could be efficiently scaled.

Breaking free from GenAI “pilotitis”

Why is the life sciences industry stuck in GenAI “pilotitis"? Several reasons come to mind:

1. Regulatory constraints around patient data
2. Risk aversion, stemming from a smaller margin of error related to life-altering decisions, which can be exacerbated by mistrust due to AI hallucinations and viewing AI as a “black box”
3. Lack of an enterprise-wide AI strategy and organizational readiness, often resulting in siloed efforts, wasted resources, and declining innovation efforts
4. Limited input from stakeholders, resulting in solutions that don’t solve real problems or fit into existing workflows
5. Choice of complex use cases with long implementation times and limited impact, resulting in slowed momentum

The light bulb moment: GenAI's potential to illuminate healthcare

The incandescent light bulb’s journey to widespread adoption perfectly illustrates the hockey stick-shaped trajectory of transformative technologies like GenAI. Initial development, which spanned decades from 1835 to Edison's commercialization in 1880, was followed by decades of infrastructure development (power plants, transmission lines) necessary for widespread use. This gradual ascent reflects the “entropy effect”—the disruption to established norms that slows adoption.

Similarly, AI's transformative potential in life sciences, while offering increased speed and accuracy in tasks like synthesizing unstructured data, continues to face resistance, leaving it in the slow ascent phase. To accelerate AI adoption, we can learn from other sectors already leveraging AI at scale.

From auto racing to healthcare: The transformative power of GenAI

Like healthcare, other industries suffer from an overwhelming amount of data. For example, in the world of automotive racing, although a race lasts hours, fans often want a quick summary of the highlights and outcomes. In this setting, a GenAI application was developed to produce a two-minute summary podcast in any language from two hours of race commentary.¹ LLM enrichment using driver data and future race events further set the system up for the next race.

In healthcare, this same approach could be used to:

• Summarize patients’ entire health histories
• Synthesize healthcare professional (HCP) feedback
• Suggest next best steps for HCP engagement
• Predict future health outcomes and treatment-related behaviors
• Streamline benefit verification by payers

We also often want to find specific information from within messy notes or audio or video files. Globant’s Advance Video Search does just that.² It uses information from a text- or image-based search to prompt GenAI to search vast amounts of video. It can find specific clips, images, and frames within the video files.

Imagine if that same precise search functionality could be used to find:

• Personalized, compliant content from a company’s website or patient support site based on a patient’s natural language search terms
• Specific patient insights from unstructured clinical notes

For example, we’ve implemented conversational AI solutions in production for multiple life sciences clients where they are able to find insights and answers from large libraries of analyses and notes. Tasks that previously took many hours of painstaking manual research can now happen in minutes. Subtle, yet valuable insights that may have been overlooked before are being brought to the surface.

Real-world applications in life sciences

The Genial Care example mentioned above demonstrates that we are also seeing examples of scalable GenAI use emerging in life sciences. It could also be applied to more than ASD, such as complex rare diseases or patients with comorbid chronic diseases who require care from many different providers.

Vishal Singal

In addition, Predictive Oncology’s PEDAL platform harnesses AI to make predictions about drug responses in oncology.³ Combining AI, data about drug responses, and a biobank of >150,000 tumor samples from 137 tumor types, the platform uses machine learning algorithms to make predictions about the effect of different drugs on tumors, with an accuracy of 92%. Although the company is using these for drug discovery and repurposing, this type of functionality has implications for clinical care, where the right drug could be identified for the right patient, at the right time.

Another example from a Beghou client involved predictive analytics leveraging 20+ variables into five easily discernable attributes to help drive field alerts. The alerts and suggestions were highly relevant, with the field team taking action on over 75% of them, much higher than industry benchmarks. One alert identified a rural HCP with a potential rare disease patient, which caused their team to prioritize that HCP and ultimately lead to a new patient start.

The AI-generated “digital brain” from Orby is changing neurorehabilitation and pain management for patients who have lost their motor skills due to spinal injury, stroke, or other conditions.⁴ By mapping neurostimulation to affected brain areas, the company’s able to tap into specific disordered functions. Applications for mental health issues come to mind, where targeted treatment could be delivered.

The power of human-AI collaboration

Within this realm of exciting possibilities, some limitations of GenAI also need to be acknowledged, or what it is not:

• Completely autonomous
• Infallible
• Unbiased
• Direct replacement for humans

These misconceptions feed into the entropy effect, mistrust, and reluctance for adoption. They also highlight the important roles of humans to ensure relevant, accurate, and trusted results, by guiding its outputs; providing oversight to mitigate bias in the data inputs; reviewing the results for accuracy; and giving feedback to improve machine learning processes.

What does this look like in practice? Recommendations could include the following:

1. Identify unmet needs with the organizational readiness to solve them: Find issues that are not currently being optimally met by existing systems and processes where the transition to AI would make a meaningful impact to end users (e.g., commentary summaries, synthesis of unstructured notes, or live audio).
2. Prioritize low-complexity, high-impact use cases with a focus on scalability: Review your high-impact use cases for the level of complexity to implement. Tasks that already have proven AI solutions (i.e., are low complexity), such as tagging and summarizing unstructured data to better understand customer sentiment, result in quick wins that maintain momentum.
3. Build the foundation: Establish a powerful, compliant infrastructure of data, technology, and governance to ensure long-term scalability.
4. Center people in the development and implementation: Field an end user panel to provide input and feedback throughout design, development, and review, and leverage your team’s expertise to drive context, accuracy, and judgment.
5. Conduct pilot testing for early exposure and ongoing refinement: Allow users to provide feedback during the build to generate excitement, foster trust, and lay the foundation for enterprise-wide adoption and expansion.
6. Incorporate change management processes: Make sure your entire organization is aligned and on board regarding the uses and benefits being pursued with AI, so that, when you go live, your organization speaks with a unified voice and makes a splash.

We believe that stalled or failed GenAI pilots can become a thing of the past in life sciences. Although data-related factors such as volume, structure, and compliance have proven to be problematic for extracting valuable insights and personalizing care, we’ve found that they can be effectively addressed using a human-centered approach and enterprise-wide strategies that drive scalable AI solutions.

About the Authors

Dan Schulman and Vishal Singal are both partners at Beghou Consulting.

References

1. Google and Formula E Break World Record for Most Participants in Generative AI Hackathon. FIA. August 2, 2024. https://www.fiaformulae.com/en/news/504402

2. Globant Releases Innovative Advanced Video Search Tool Leveraging Google Cloud’s Gemini Models. Globant. October 1, 2024. https://www.globant.com/news/globant-releases-advanced-video-search-tool-avs

3. PEDAL Proof of Concept. Predictive Oncology. https://predictive-oncology.com/poc-report-page/

4. Ortech. Orby. https://orby-company.com/ortech/