Artificial intelligence versus real insight – does pharmaceutical industry consulting face a dilemma or a progression?
August 29, 2024 | AI, Pharma, RAG, data infrastructure, retrieval-augmented generation
By Frank Fisher and Zoë van Helmond
Since AMICULUM’s inception, we have been supporting clients using our consulting expertise to enable medical and commercial strategic decision-making across the drug development pathway, from early-stage landscape analysis and insights interpretation through to positioning, launch planning and change management. Of late, AI has started appearing in conversations and pilot projects around all these topics, and we’re finding that our strategic consulting clients (across the pharmaceutical, biotechnology and medical device sectors) are mostly divided into two camps: one group sees AI as the instant solution to all of their woes; the other views AI as all hype and no substance, with significant risk attached to its use. We see things differently: AI can produce expensive gibberish, but it can do some things very well if the conditions are right and the tool is readily suited to the data it’s being fed.
Let’s look at some real-world examples that might echo some of your challenges with AI use and see how these can be turned into success stories. Both examples relate to the extraction and provision of medical insights. The second example addresses long-term strategic perspectives and decisions, but the first is focused on real-time, tactical insights – we’ll begin there.
AI provision of ‘Next Best Action’ nudges
Our client was aiming to provide their medical field force with pre-call prompts for the ‘next best scientific action’ based on insights generated by an analytical AI monitoring system operating on their Veeva/Salesforce Customer Relationship Management (CRM) system. Establishing the system had been expensive and time-consuming, not least because exposing CRM data to a third-party tool meant an extensive digital governance process. In practice, the AI was failing – few insights were being provided and these were low grade, generic, and had no personal value to healthcare professionals (HCPs) or medical science liaisons (MSLs).
What was the problem? This wasn’t an AI problem; this was a data issue. It was a case of NINO – Nothing In, Nothing Out. MSLs had not been populating the CRM with regular and detailed call reports. Most communications with HCPs were not omnichannel and MSLs had not been capturing useful data on HCP interactions or writing it back to the CRM. Furthermore, the omnichannel tools had not been optimized to generate ‘active’ omnichannel analytics (ie those monitoring progress towards specific strategic goals or HCP movement along adoption ladders). In short, the AI tool had nothing to go on and without good data, it couldn’t draw any conclusions.
Our advice to this client was to empower MSLs to be the data gatekeepers, to bring all communications into an omnichannel strategy, and to make use of their business’s data analytics team to better manage digital touchpoints to create and capture more valuable data relating directly to HCP actions, perspectives and needs. In other words, to generate better data!
Preparing data infrastructure for AI enhancement
In the second example, our clients needed our help to understand their existing data availability and veracity prior to handing it off to a third-party AI provider. The goal was to harmonize the collection and use of data across the enterprise to help provide guidance for lifecycle management decisions for future years or even decades. Such a broad topic meant the search for insights went well beyond traditional medical affairs niches. Business intelligence was tapped, and a wide net was cast to gain insights from market access, HCPs via CRM, congress and key opinion leaders (KOLs). However, until we became involved there had been no coordinated attempt to manage this. Immediately we saw that provenance was vital to establishing insight veracity. Where did the insight come from? Furthermore, the roles and responsibilities were unclear. Who owned the data? What legal or regulatory constraints were there?
We also needed to understand business processes: what was the cadence of data generation and where were the bottlenecks? Our consulting teams successfully organized a series of workshops to expose and mitigate all the above issues.
Meanwhile, our technical colleagues at AMICULUM determined the architecture of suitable AI tools needed to maintain the useability and integrity of the data and generated insights. We were not concerned about NINO here, or even GIGO (Garbage In, Garbage Out), because we knew the data were fundamentally good; however, they were largely unstructured, and despite claims to the contrary, AI performs best with structured and explicable data. Not only was it vital to weigh the data for authority but it was also necessary to tag the data to ensure the right stakeholders could be alerted to relevant outputs. We also knew that most AI models would not carry such provenance data through to final outputs, meaning they would lack the ‘explainability’ needed to support insights that would be used to guide multimillion-euro decisions.
By using workshops to investigate stakeholder needs, generate buy-in and guide towards solutions, we were able to create an ideal AI approach to work alongside the managed data flows our consulting teams had planned out. This project resulted in considerable change for our client company and a refocus on AI analysis, which included the creation of a new team to lead the transformation.
AI can be supportive without being transformative
AMICULUM is working with a retrieval augmented generation (RAG) AI approach that is purpose-built for flexibility and modular deployment. This technology can support consulting analysis by exposing patterns and insights in data from almost any source, including clients’ ‘legacy’ data lakes, small-scale data pools such as congress reports or a series of advisory board outcomes, publications databases such as PubMed and even social media.
We can scale this service, meaning our consulting teams can offer AI-assisted services to clients at almost all price points – we’re really excited by the possibilities. We’re a very practical bunch. We don’t think AI is going to solve every problem in the short term, nor do we think it’s going to take our jobs, but we do see genuine opportunities to improve efficiencies, expose hidden patterns, provide more routine and rapid insights generation, and seek valuable business insights almost everywhere that there are data.
So, returning to our original question – does AI bring dilemmas or a road to progress? The simple answer is both. The slightly more nuanced answer is of course it’s both because any question that needs wider consultation to answer is always going to be somewhat hedged. A more productive answer would be to suggest that the road towards AI adoption is likely to be just as informative as the AI outputs if you explore and plan that roadmap carefully.
Tell us what insights you want to unlock with AI, and we’ll show you how we’ll do it. Contact Frank Fisher (frank.fisher@amiculum.biz) or Zoë van Helmond (zoe.vanhelmond@amiculum.biz) to learn more. For more insights, visit the AMICULUM News and Insights page.
This content was provided by Amiculum
Company Details
Latest Content from Amiculum
AMICULUM® collaborates with the University of Dundee to develop a new Master’s course in Science and Health Communication
AMICULUM Limited is proud to announce its collaboration with the University of Dundee in developing a Master’s course in Science and Health Communication. The Master’s course was an idea conceived...