With special thanks to Hans Kaspersetz from our AI sponsor Arteric who kindly shares this blog post. Hans will be presenting at our forthcoming co-located 2019 BioPharma eMarketing Summit & 2019 MedDev eMarketing Summit on May 13-15 2019 in San Diego. (The only unique, fun & interactive ‘TED-style’ digital marketing events for the BioPharma & Medical Device industry!)
Artificial Intelligence as a Marketing Tool Is Bull. Don’t Waste Your Time.
Almost half (47%) of the 318 marketers surveyed in September 2017 think that artificial intelligence (AI) is overhyped as a marketing tool. When asked what emotions they felt when they saw or read about AI, 40% responded with “skeptical.”1
I understand why so many marketers might feel this way. Vendors, industry publications, and bloggers constantly bombard them with urgent stories about the next great digital savior and how their campaigns will fail if they don’t hop on the bandwagon. It’s only natural for marketers to question whether AI will be the next iPhone or the ever-promised flying car.
AI isn’t bull
The evidence that AI is as transformative as the iPhone is very strong and growing rapidly, and that with the proper guidance, AI is a very effective tool for healthcare marketers. Arteric recently created an AI solution that’s providing novel, unexpected, and actionable insights into what information HCP and patient audiences need and how they search for this information online. We’re making discoveries in 15 minutes that previously took 40 to 80 hours of manual analysis.
But the benefits go way beyond speed. With AI, we’re analyzing search patterns in multiple languages. In one case, we discovered unexpected Spanish-language search volume for a client’s product in an English-only search campaign targeted for audiences in the United States. We shared the insight with the client, and it correlated with and supported data that they had from other sources, data we that we were unaware of. The interesting facet of this insight is that we found it while we were analyzing more than 250,000 searches for a specific topic by using natural-language-processing AI that implemented latent semantic analysis using semantic folding.
For fun, we looked at the least relevant searches to our topic of interest. Surprisingly, this revealed a couple of searches in Spanish. Following this insight, we implemented a different algorithm that leveraged utility-grade SaaS AI that could identify the language and translate the text. Once we tested this algorithm and data, our application passed the brand’s search data through to the new AI and found hundreds of searches in Spanish that should not have been there. Eureka!! The really exciting part was that we could pass up to 100,000 searches through our system per minute and categorize the results almost instantly.
These insights revealed some interesting knowledge about Google AdWords algorithms and Google’s organic algorithms. It also led us to develop demographic-specific content and campaign recommendations for the client’s product. The results aren’t in, but we believe there are strong signals that this targeted approach will bear fruit and revenue.
Our combinatorial use of multiple AI algorithms and training sets afforded our client several actionable benefits:
- Ultrafast analysis of more than 250,000 natural language data points
- Identification of new market opportunities
- Identification of real-world conversational language describing the brand
- Unexpected insights about customer and search engine behavior
- Cost-effective processing of large sets of data
I strongly believe that the magic of AI and machine learning does not arise from the technology or the data. It arises from our human curiosity and desire to look beyond the obvious. While the pursuit requires a high level of technical and analytical expertise, it demands a playfulness, almost a child’s mind, to reimagine the world when we can look past the obvious and identify patterns and insights that historically have remained hidden because they had a very low signal-to-noise ratio. These anomalies were beyond human identification because it is impossible for most people to see a pattern in 250,000 data points, but are clearly obvious to a machine.
I am fortunate to be surrounded by brilliant people at Arteric who demonstrate these characteristics and are willing to take the risks necessary to explore uncharted concepts, ideas, and data.
Our experience isn’t unique. GlaxoSmithKline used AI technologies to learn why some parents resisted vaccinating their children against measles and mumps, while others do not. GlaxoSmithKline transformed these learnings into educational materials that addressed the concerns of reluctant parents.2 And in a June 2017 survey of 3500 global marketing leaders sponsored by Salesforce, 64% of marketers currently using AI report that this technology has greatly or substantially increased their overall marketing efficiency.3 In addition, 57% of AI users found it absolutely or very essential to create 1-to-1 marketing across every touch point.3
For healthcare marketers to reap the value of AI as a tool to solve business problems in the real world, they need to understand what an AI solution is and what it takes for an AI solution to succeed.
AI solutions contain 3 functional components.
Each must be planned in advance and optimized for the specific business problem that the AI solution is intended to solve.
This includes pay-per-click and Search Console data, customer service transcripts, medical information department inquiries, surveys, and email inquiries. Any information relevant to the brand or to the brand’s products contains value. The data is a priceless commodity. The challenge is that it takes time to aggregate enough data for it to be useful. Start collecting data immediately and warehouse it.
The brand, NOT the brand’s vendors, must own, store, and control the data. A brand that isn’t gathering every possible byte of data will fall behind competitors that do.
AI and machine learning solutions must be trained with large sets of data. The selection of the training data will have a significant impact on the usefulness and accuracy of the system. For example, one of our solutions is trained on all of Wikipedia. This is useful for topics and data that originate from common knowledge. However, this approach is less useful for disease education on new mechanisms of action for a specific disease or pathway. Wikipedia will not have a sufficient volume of relevant content for this training. In this case we may need to add a library of medical jargon and a large collection of relevant journal articles to supplement the training set. This requires domain expertise in the semantics of the topic and availability of relevant content and training data. It also requires significant testing to determine if the output matches what is expected across a variety of inputs. When we input “apple,” do we mean the food or the company? This will be determined based on the training data and the context around our use of the word “apple.” If we used the Sunday newspaper circular and coupons to train the machine, the machine would probably interpret “apple” as the fruit. If we used the content of Wired magazine instead, the machine would probably interpret “apple” as the company.
The capabilities of AI and machine learning algorithms vary dramatically. Selecting and implementing AI or machine learning solutions requires a thorough understanding of the strengths and weaknesses of each implementation. For example, there are multiple implementations of natural language processing such as semantic folding versus dense embedding models. Both of these are useful, but they have different strengths and weaknesses. The team must be willing to implement multiple approaches to identify which will work best based on the business problem and available training data.
Solutions need purpose-built user interfaces that make interacting with the underlying technology intuitive and practical.
General purpose AI does not exist for the purposes of marketing.
For AI and machine learning to be successful, the implementers must narrowly define the problem to be solved and determine if it can be solved algorithmically. Broad, loosely defined problems cannot currently be solved with AI and machine learning: the technology and data required are not practical for healthcare marketing.
The team implementing the solution must be cross-functional and include:
- Business owners/champion/strategists
- Data experts
- Creative software developers
- A visionary to connect the dots
The team must recognize that human curiosity, creativity, and oversight are required for AI and machine learning to be successful.
Marketing data is the fuel that powers the engine. Your data holds the value that technology and strategy unlock. Broadly collect and store all the available data, even if you don’t plan on implementing AI in the near term. You will thank me in 3 to 5 years. Marketing data is a brand’s intellectual property, so treat it accordingly.
The models that connect audiences with brand content are undergoing a tectonic shift. We’re moving from content strategy and generalized answers to conversational design and ultrapersonalization. Understanding the role that data and AI play in this revolution, knowing how and when to apply AI, and ensuring that your brand collects and retains relevant data will result in a competitive advantage that will be difficult for slower competitors to overcome.
Recently, I sat down with Chris Conner from Life Science Marketing Radio. We discussed topics that ranged from applying AI and machine learning in your business to optimize your PPC campaigns and content strategy all the way to how captchas work and how the human race has been enslaved by Google to train its image recognition algorithms. If you want to take a deeper dive into what we discuss, I invite you to listen to “How Artificial Intelligence Can Dramatically Improve Your Pay Per Click Campaigns.”
For two decades, Arteric has been demystifying technology to help healthcare brands make wise, appropriate choices to drive their digital marketing campaigns. Contact us at 201.546.9910 to discuss how to apply AI to maximize the impact of your marketing plans.
- Resulticks. Introducing the marketing flab to fab challenge. Resulticks website. https://www.resulticks.com/marketingflabtofab.html. Accessed November 2, 2017.
- Fagella D. Mining online discussions for deeper customer insight. TechEmergence website. https://www.techemergence.com/case-studies/Luminoso/glaxosmithkline/. Accessed November 2, 2017.
- Salesforce. Fourth annual state of marketing. Salesforce website. https://www.salesforce.com/research/. Accessed November 2, 2017.
SIGN UP FOR BIOPHARMA UPDATES