Remember when everyone said AI would revolutionize the content business, saving pharma companies millions while generating exquisitely personalized messages that would improve people’s lives? Yeah, us, too.
Yet so far, the road to Content Nirvana is littered with ambivalent CMOs, underwhelmed strategists and irritated customers whose fervent wish is — as always — that pharma companies would quit spamming their inboxes and news feeds with irrelevant messages. And every story about an AI gaffe hardens the resolve of C-suite skeptics, making them more hesitant to trust a technology that occasionally cranks out misinformation, misdiagnoses and bland malarkey.
But there are also a growing number of sanguine AI travelers. They believe that the more hours poured into AI-powered personalization, the sharper it gets. While their AI dreams may have been attitudinally downgraded from “Wowza!” to “OK, this might take a while,” they remain intent on shaking up every part of the labor-intensive content creation marathon.
“No one has figured out how to use generative AI in the end-to-end process of content creation,” says Ashkan Afkhami, a global leader of the healthcare practice at BCG X, the AI and analytics arm of Boston Consulting Group. “There’s no driverless car yet. But some of the larger healthcare and biopharma companies we work with have reduced the time spent on content creation by 30% to 50%.”
Still, it’s hard to square those impressive gains with the collective sighs of disappointment. Putting it all in perspective requires a look back to November 2022, when ChatGPT’s generative AI became the corporate world’s unrelenting obsession. It’s easy to see why pharma was so smitten: Healthcare companies pump out oceans of content, from catchy TikTok videos to data-dense white papers, and each piece has to thread through tedious approvals.
So the idea that clever machines could personalize each piece of content, writing bespoke messages that address allergies in myriad ways, took off. You want geographically pinpointed pollen counts from Kissimmee to Kalamazoo, or a comparison of the benefits of nasal mists versus tablets? All in seconds? Yeah. Of course that captured pharma’s imagination.
Which explains the current cynicism, as slower-than-expected results have plunged some initial boosters into what Gartner’s famous Hype Cycle calls the Trough of Disillusionment. Jeff Rohwer, head of commercial solutions at Real Chemistry, has a favorite example.
“I do a lot of work in rare diseases, so I made an assistant, asking the bot questions such as, ‘How did you feel when you got that diagnosis? What did you do next?’ After playing with it for five or 10 minutes, I was like, ‘This is going to change the world.’”
Alas, his euphoria was short-lived. “The next day, I played with it for 30 minutes and realized, ‘I’m talking to a bot, not a patient. These answers are mechanical and are not going to change anything. What was I so excited about?’”
Rohwer nonetheless believes the life sciences world has started its climb out of the disappointment ditch, and are now trudging up the Slope of Enlightenment.
Real Chemistry is already deploying AI in countless business cases, but the company is more tentative about content creation.
“We’re not doing it with every client,” Rohwer explains. “We are testing a lot. For example, we ask LLMs [large language models] to react to content as if it were various types of patients, and then we compare those reactions to actual humans to see if the results are similar.”
It’s not perfect, he concedes. In some cases, however, it’s proven reliable.
“It has allowed our writers to be that much more confident that their content is more likely to resonate with a patient or an HCP.”
The first burst of enthusiasm about generative AI focused on content creation. Basically: How well could it write? How accurate are the images it creates? But organizations enjoying the best results are using it in other parts of the content process, Afkhami notes.
AI can help in the first step of the process (e.g., generating briefs and personas) as well as the second (e.g., generating text, images and videos, even if they are error-prone). But it shows its greatest promise in the painful third step, which is shepherding content assets through the proper medical, legal and regulatory channels.
“The goal is to navigate those approvals as quickly as possible while being safe, effective and efficient,” Afkhami says.
Once content is approved, AI can help deploy it across different channels. In the final step of the content dance — measurement against key performance metrics — AI finds quicker ways to close the loop.
The risk of not trying
While no one has nailed the perfect approach to AI content generation yet, there are few excuses not to try. Given the intense speed of AI development, the once-valid tactic of letting other companies figure out the risks and rewards (and then building off their successes) doesn’t make economic sense.
Yes, glitches and mistakes remain a headache. But Rohwer constantly reminds himself that the LLMs in use today are “probably the worst LLMs we will ever use — and they do some things pretty well.”
Some organizations are already promising one-to-one content personalization. CVS, plagued by a shortage of human pharmacists, has revealed that it is using machine-learning personalization to increase prescription refill rates and reduce gaps in treatment. To protect vulnerable customers from extreme heat, the company plans to send personalized warnings to its Aetna members. This will pair advanced environmental data analytics, which provides localized forecasting on air quality, wildfires and weather, with a person’s medical and pharmacy data.
Opella, formerly Sanofi Consumer Healthcare, is also digging in. “We are now using AI to test and learn ways to generate contextual content for consumers we target or retarget,” says chief marketing officer, North America, Claudine Patel. “We look for search information and behavioral practices to make sure we can serve the right content to our consumers.”
By way of example, she points to Opella’s work on constipation treatment Dulcolax. “A consumer may search for why they are experiencing constipation, or they may look for some of the best solutions for the issue. By using AI, we can be sure to provide the right content for the individual,” Patel adds, noting that the company’s content teams are also using AI to “develop consumer personas for innovation workshops.”
Real Chemistry, on the other hand, is experimenting with content for people approaching the point of revaccination. Explains Rohwer: “What should we say, and at what moment? Should messages be family-themed, such as ‘Your family depends on you, you can’t afford to be sick?’ Or is it more effective to target parents differently? We can test how we phrase things to see what resonates more.”
To be clear, he’s not talking about a copywriter asking a machine to draft some ideas. “Our large-language models are literally pulling out our audience insights and letting the AI make suggestions based on that.”
But should we trust it to do so? Rowher responds, “We very much believe in the human in the loop. With generative AI, you need to review what AI creates.”
While safety and accuracy will always be the top concerns, marketers using AI in content creation aren’t neglecting quality or forging the emotional connections that build brands. “We want to create content that resonates, and we still believe that only humans can do that,” Rowher adds.
Building an AI army
Precisely who those humans should be is still a question mark. While every pharma company and marketing shop is scrambling to hire experts, most believe AI can only deliver on its promise when it is entrusted to the rank and file. Read: It can’t just be the geeks.
Rohwer says Real Chemistry has constructed what he calls “a safe playground.” On it, employees are free to experiment with the agency’s proprietary AI tools. But he cautions that “all-access” doesn’t necessarily translate to “all-enthusiasm.”
“Right now, the people playing with the technology the most are the hand-raisers, the people who like trying things out and testing them. We’ve set up guilds so they can share what they learn with others.” Eventually, the thinking goes, the AI busy bees will pollinate the entire company.
At Sanofi, all 22,000 employees have access to AI and 9,000 are using AI to help make decisions, Patel reports.
“By adopting AI to facilitate business transformation and enable data-driven decision-making, we’re also reshaping how projects are managed and governed,” she says. “Our integration of AI is not just about technology. It’s also about driving a cultural transformation within the company.”
As tests and pilots evolve into best practices, there remains precious little clarity about what all of this might mean for staffing content teams. Most observers, however, stress that we are still a long way from eliminating human beings.
Marketers worry about taking their hands off the steering wheel of brand voice. They’re nervous about ceding control, first to an army of bots supervised by just a few humans and then, perhaps, to a content process featuring no people at all.
Which doesn’t mean that companies haven’t already started to wrestle with thorny staffing questions. For instance, one BCG client has gone from creating 20 to 30 assets a week to between 300 and 500 with the same team, Afkhami says.
“But how many assets get through, and how effective are they? Conversations have to be about driving efficiency. If I made you 50% more efficient and productive, what other projects can I assign you? I can’t hire half a human.”
The promise of personalization
Given the number of complexities associated with AI-aided content generation, it’s easy for companies to forget that the point of personalization is not merely creating more content. It’s about creating content that delivers the right message at the right moment. That goal has been the same since long before generative AI began dominating every conversation.
Personalized content must be adaptive and multimodal. It must be context-relevant. It must know when a lunchtime text will resonate more deeply than an evening email.
“When someone has diabetes, sending a message about food choices to their Apple Watch while they are standing online at Panera is useful,” Afkhami says. “But it doesn’t help to send a ‘remember to pack your insulin’ reminder when they’ve already boarded a plane.”
Which is to reiterate: The basics of personalization will remain daunting, even before you factor in AI.
“Serving the right content dynamically requires that you know a person — not just they are a doctor or what kind of doctor, but that they are Dr. Smith. That’s the basic building block,” Rohwer says. “It requires that person to be in a connected channel so that you recognize them compliantly. You need templates to load content and you have to have approved content modules, and then you have to have the business rules to pull it all together.
“That’s a lot to get right, even before you add in the AI.”
From the August 01, 2024 Issue of MM+M - Medical Marketing and Media
