Pharma is running out of names. Can an AI help?

The defining feature of the AI discourse at Reuters Pharma USA this week in Philadelphia was its specificity – while previous years and similar events have featured a lot of panels about the idea of AI writ large, this year was full of concrete examples of AI in action.

And AI is coming to the rescue to solve all kinds of problems. One of the most interesting examples offered up was an AI to support pharmaceutical naming.

Naming a drug is one of the most perilous and complex branding exercises there is, according to Dorothy Linvill-Neal, director of nomenclature development and strategy at Regeneron.

Not only do pharma companies have to undertake the artistic and design challenges common to any product developer – coming up with a name that’s memorable, catchy, and evocative – but they also must find a name that will be approved not only by the trademark offices in every country where they intend to market the drug, but also their regulatory bodies.

“The trademark analysis through the patent and trademark office is distinguishable from the health authority analysis … which centres around potential for inadvertent substitution,” Linvill-Neal explained at a presentation at Reuters. “That's where the patient goes to the pharmacy or is in the hospital, and they are given the wrong drug, thus patient harm or death could occur.”

The FDA and EMA aren’t involved arbitrarily – studies have found that 33% of medication errors are the result of drug name confusion. But they do complicate the process.

“The EMA is particularly challenging because we do have 27 member states that must have a harmonised consensus on the acceptability of a single name to be used in a centralised procedure and sold in all of those different markets,” Linvill-Neal said. “They also could object on the basis of negative associations or connotations, and that could be a sole basis as well as sounding or looking like to another drug.”

All this results in what Joe Daley, president of Addison Whitney, the global branding agency that named Ozempic and Wegovy, describes as “a process of taking in information, creating names, viewing, and then attrition, creation, and attrition.”

“It's a process of compromise,” he continued. “It's a process of dilution to brand strategy. And the ... attrition pressure is increasing. As more and more names get approved, there's less room for names that don't sound like and don't look like another name.”

Addison Whitney is aiming to solve this problem with an AI named Ari, which has been in development for about two years and in beta for the past six months. It references databases of existing names to screen and risk-stratify names as human designers brainstorm and iterate on them, as well as using generative AI capabilities to generate name ideas based on strategic parameters.

“We can load a spectacular amount of strategic information and insight on the front end of Ari, and Ari helps us mine that data to come up with names that are on strategy and at the same time are being stratified for risk, regulatory, and legal risk,” Daley said. “This helps our brand design or verbal design people very rapidly select the high probability names, iterate, and run the process again. It takes literally a quarter of the time that it used to take to accomplish this same cycle I’m describing.”

Linvill-Neal says this is a tool that takes burdens off not only the product naming team, but also the legal and regulatory teams they typically spend a lot of time consulting.

“We know that regulatory is of paramount importance and that that prescreening for patient safety is the North Star,” she said. “So the ability to rapidly fire prescreened names makes a huge difference.”