Pfizer’s Charlie is Powerful, But What Can’t it Do?
Pfizer’s in-house Large Language Model is extremely powerful. Dubbed Charlie, the proprietary GenAI can create responses no human could ever dream of. Need a new marketing message? Charlie will sift through all the historical messaging for the brand, look at competitor content and patient sentiment, merge that with the requested direction for the new message and give you the best possible copy. Want to make sure the new message complies with all regulations? Charlie will do that for you as well. When it comes to creating content based on staggering amounts of information, LLM capabilities cannot be beat.
Having said that, there are many things Charlie cannot do. Below is a hypothetical chat with Charlie.
--
RICH: Charlie, can you tell me how I should spend my marketing and sales budgets across all promotions and brands to maximize profitability.
CHARLIE: Promotional attribution and optimization is a necessary part of successful marketing and sales operations. The most accurate approaches involve using machine learning to attribute impact to various promotional activities and to generate corresponding response curves for use in optimization. I have no machine learning experience specific to this question. I would suggest you talk to a data scientist who can create a model and provide that information.
RICH: That is a shame. How about this, can you give me the next best action by HCP for the HCPs called on by sales reps this week?
CHARLIE: What you are asking for is commonly referred to as omnichannel and can drive a significant lift in outcomes over traditional multi-channel siloed marketing. Unfortunately, I cannot give you an answer because I lack the data and machine learning knowledge necessary to generate a response.
RICH: Strike two. Let’s go for something easy. What is the prominent sequence of engagement activities leading up to a new patient start?
CHARLIE: I can try to answer that question assuming I have been trained on the right source data but I would also need the data organized longitudinally to accurately answer the question. I am not confident in my ability to do that and I fear my answer would not accurately address the information you are looking for.
--
Will CHARLIE’s LLM someday be able to answer these questions? Maybe, but not today and not in the near future. Can these answers be provided through an ask and respond paradigm using GenAI? Absolutely. Using AI Ops to merge the code generation capabilities of an LLM with the results generated by machine learning delivers the answers to all of the above. To learn more about how to do what Charlie cannot, set up a time to talk with SENTIER and explore ways to dynamically deliver these insights to your organization.
