November 7, 2019

Why AI Should Be a Pharma Marketer’s Best Friend

Zeroes and ones on a computer screen

Intouch VP of Innovation Abid Rahman recently spoke on artificial intelligence (AI) at two DTC forums on social media and technology, one in Anaheim, California, the other in Jersey City, New Jersey. His presentation, “Making AI Accessible for the Rest of Us,” looked at the growing ubiquity of AI, identified challenges to implementation, and discussed three key ways to implement AI that make smart business sense. Read on to learn more.

Artificial Intelligence in the Real World
There’s never been a better time to put the amazing capabilities of AI to work in pharma marketing, but first it’s worth remembering the difference between the perception of AI technologies and the reality of AI implementation in practice. Obviously, the AI technologies we experience in the real world are different from those we usually see in the movies and popular media. Thankfully, we don’t have sentient robots or computers vying for their overlord status yet. Instead, the proliferation of AI has been more subtle — some of it we interact with knowingly, some of it unknowingly. Every consumer computer and smartphone now comes with AI built into it, for example, but who among us thinks about that? Other examples of AI implementations we already take for granted can be found in platforms like Facebook, Google, Microsoft and Amazon, that we use every day.

  • Facebook: In addition to its AI-powered facial recognition and ad targeting, Facebook uses AI for content moderation and news feeds.
  • Google: AI is used in every Google product to some extent. If you’re a Gmail user, you’ve experienced Google’s suggested replies, which makes responding to email faster and easier. There’s also Google Assistant, the search giant’s version of Amazon’s Alexa, and Duplex, an AI-powered tool that helps users set up restaurant reservations via Google Assistant.
  • Microsoft: The software colossus uses AI to power Skype chatbots, data analysis and its own assistant, Cortana, and is currently working on bringing AI into Office 365.
  • Amazon: You can’t talk about AI and Amazon without mentioning Alexa, but the company also uses AI for dynamic pricing, purchase recommendations, and most recently, checkout-free shopping in brick-and-mortar stores.

Three Ways to Implement AI
There are a number of ways pharma can implement AI, internally and externally; for example, AI assistants can be used to generate and distribute medical information, help field reps be more efficient and successful when meeting with physicians, conduct content analysis and review, complete MLR reviews, and more.

Given the ubiquity of AI, it’s interesting to note that the term itself is somewhat confusing. In fact, what we mean by AI has been evolving over time. It’s worth revisiting more established terms like machine learning (ML), natural language processing (NLP) and robotic process automation (RPA), since these are areas where most of the AI research and funding is heading, and because there are already many use cases that demonstrate AI’s potential.

Machine learning is the ability for machines to learn from data. In traditional software systems, you provide some data to a program, and you get an output. For example, if you provide the input of 2+2 to a calculator, the output is 4. This output is very much expected, and it would be strange to not get that output. In ML, our inputs don’t always look like that. We provide data and an ML algorithm to a machine, and the output is a program. The program can then look at data it has never seen before and predict outputs. The use cases for a conventional program – think PowerPoint or Excel, which are designed to perform a specific kind of task — are very different from the use cases for ML. In areas where conventional programs work well, we wouldn’t use ML. But what about situations like predicting medication adherence or predicting if a person will like a certain movie? Conventional software that is rule-based isn’t designed to be able to come up with those answers. Especially when you have thousands of variables and massive amounts of data; making rules for each of those variables is not practical. That’s where ML comes to the rescue. ML can learn from existing data to predict outcomes for data it has never seen before.

Natural language processing is an implementation of machine learning. NLP is the ability of machines to understand language. In this case, understanding language is really about identifying the context of spoken/written information to extract meaning by parsing the words and phrases in a way that machines can understand. Language is difficult to deal with by its very nature. Language also changes over time. A system that’s trained on NLP has to evolve with the evolution of language. It also has to learn the way population groups, such as a specific patient population, talk about their disease.

Robotic process automation is all about teaching a machine to do tasks that are easily repeatable by machines. Over time, we’ve learned that tasks that machines do well, humans don’t do well and vice versa. Therefore, machines shouldn’t do something that humans do very well already, such as building meaningful relationships with other humans over dinner. RPA is more about augmenting human intelligence and activities, rather than artificial machine intelligence. It’s worthwhile to identify repeated activities that are time-consuming and non-scalable when performed by humans; these activities can often be automated to produce massive efficiencies. The best current AI implementations are seamless for consumers, and they combine ML, NLP and RPA to provide meaningful and actionable experience, such as predicting the expected traffic patterns using Google Assistant.

Obstacles – and Solutions – to AI Implementation
While AI is all around us, not all industries — e.g., pharma companies — have been able to derive major benefits from the technology yet. There are a number of reasons for this:

  • There’s a misconception that AI can only exist in a large, multichannel ecosystem that requires massive investments. This isn’t true. In fact, some of the most quick and effective AI implementations can happen using algorithms that are freely available and can be implemented without needing large infrastructure.
  • Due to the newness of the technology, there is a perception of high risk associated with the return on investment. This can be minimized by implementing smaller-scale AI projects to show the benefit of AI before jumping into bigger investments. You also need education and advocacy from the top down to mitigate the perception of risk. AI is not just another technology; it’s also a mindset about doing work differently. Over time, companies will experience cultural shifts because of AI.
  • There is often misalignment between the needs of marketing/customer experience teams and those of the IT team with respect to technology and platform, as well as the company’s tech and AI roadmap. Which means it’s also important to get alignment and buy-in from all relevant stakeholders by involving them at the beginning of a project.
  • AI is hard to do right. This is because of resource challenges rather than technology challenges. There is a lack of people who understand the AI technologies and also understand the business needs at the same time. Identifying and empowering the right resources is key to successful AI implementations.

It’s clear that not only are we and our consumers ready for AI; now industries beyond the flagship technology companies have to catch up to provide what our consumers have been yearning for. There is a lot to win with AI and nothing to lose except opportunities.

Reach out to your account team to learn how AI implementation can work for your brand.