"Yesterday, my son asked me how to write the name of the community in English for his application to an overseas school. I told him to write it in Pinyin, because the English name is just a data record on the school's side, and the ultimate use is for the Chinese postal staff to read when you receive the materials." — Lu
This little story is about who you are conveying the information to and how to let the other party receive your information efficiently. Coincidentally, at the 2024 CMAC conference a few days ago, we were invited to the "Medical Information: Accurate and Timely!"forum, which was also about the topic of information communication.
I won't say much about my self-introduction. I am considered an AI enthusiast within the company. For example, the outline of today's PPT was actually made by AI. However, during the process of finalizing the PPT, my colleagues told me that there were many problems with the outline that needed to be changed.
I think this might be the norm for future human-AI collaboration: AI will give everyone a starting point, and at the same time, human value is indispensable — the final implementation still depends on us humans.
01 AI+ Content Management
Under this premise, for the medical information communication link of pharmaceutical companies, where can AIGC help us in business scenarios? After communicating with many medical department friends, we found that many people have an uncertain attitude towards AIGC, not only uncertain about what it can do, what to do first, what to get, but even uncertain about what it is — "AI I know, GC is..."
Based on these questions, the first thing we thought of was to solve the problem of information collection and sorting that consumes a lot of time and energy. The amount of information in the pharmaceutical field is huge and updates very quickly, just like an iceberg: below the water surface, it is the information collection, reading comprehension, and summarization that we spend a lot of time and energy on; and the tip above the water surface is the information finally presented to the outside world.
With AI capabilities, we can complete everything from literature information acquisition to understanding and sorting, to summarization and induction, using large models.
Taking the business perspective of the medical department as an example, everyone may need to write papers, and behind writing papers is finding new articles. "Finding new articles" seems simple, but in reality, it involves far more than just "finding" — you need to search, filter, read, organize, and even everyone might repeat the same action.
We extract this task and find that fundamentally it is the 'understanding and judgment' of content information, 'understanding and judgment' is to find the essence from a huge amount of content, which is also the part that AI is better at. It may not be able to give the most perfect insights, but it can quickly reduce our time and energy costs.
Especially now, everyone is talking about "thousands of people, thousands of faces" and personalization, which means we need a more refined understanding and judgment, not only understanding literature but also understanding the new content we have generated, including audio, video, one-picture-understanding, etc., to provide a basis for personalized content, which comes to the 'refining and processing' link. Many times, the processing of new content is based on the combination of existing content. When the total amount of existing content is not large, traditional manual understanding and combination are no problem. What if the content volume reaches thousands or tens of thousands? I'm afraid it's not something a skilled worker can complete, but AI can.
In this scenario, MeDomino Content Hub is an enterprise-level intelligent knowledge base that solves the collection and sorting, summarization, summarization, and screening search of information, and even cross-content intelligent Q&A, helping everyone complete these routine actions efficiently, leaving more time to do high-value work.
02 AI+ Customer Insight
On the other hand, it's more important.
We all know that we need to understand customer portraits, or doctor/expert portraits, so is it based on the information provided by front-line colleagues, or based on data insights? How deep and accurate is the understanding? Is it subjective judgment or objective analysis? These factors will affect the quality of our entire work results.
For example, when we organize meetings, we need to visit experts, and what we need to consider is far from just "noticing". It's about determining which customer is suitable for which meeting, what I should tell the customer, which other customers can influence this customer, etc., this is another set of information — information strongly related to people — a large amount of subjective cognitive information.
In the final analysis, the first step is still 'understanding and judgment', only this time the object of understanding has changed from content to people.
We have an actual cooperation case, the first party is a product in the field of anti-infection, they need to pay attention to including Zhang Wenhong and other top doctor information, and it was found that: in that special period, Dr. Zhang Wenhong was very active, generating dozens of new data every day, the colleague responsible for Zhang Wenhong was surrounded by a large amount of information, unable to see it, and there was no energy to do information understanding and judgment. In the end, we used AI to help them analyze information, summarize key points, and transform a large amount of information into targeted in-depth insights and action suggestions, such as topics to reach.
Similar scenes can now be solved with AI, AI can help us collect and gain insights into customer information, and then combine personalized content to better convey information.
03 The Pareto Principle of AI Value
We jump out of the scene and look at the value at a higher level.
My view is that we must be fast, don't do fine carving work at this stage. You carved a very exquisite castle on the beach, and a wave came and it was all gone, this wave is the iteration of large models. So, I would think that at this stage, don't get entangled in fine carving, but to do things that can quickly reflect business value, quickly let every colleague feel the effect in their daily work.
For example, when we do content generation, we may spend 80% of the time on content understanding and generation, and the last 20% of the time on optimizing the format. Then we should use AI to compress this 80% of the time cost, and don't get entangled in whether AI can solve the remaining 20%. Being able to compress 80% to 20%, that's the value.
In fact, before the large models came out, Vein Insight had its own small models. We spent two or three years, from data labeling to special training, and landed our own small models, building castles on the beach, and as a result, last year the large models came out of the blue...
Perhaps it is precisely because we have paid a high cost for the changes of this era that we can stand in the cracks and see the sunshine, and want to tell everyone some more efficient ways to land, of course, we also hope to help more pharmaceutical companies to realize value with AI.
