This Philly-based company is using AI to predict which medical research can lead to profitable drugs and therapies

A Philadelphia company has been analyzing readers and users of academic research since 1960, long before internet metrics fueled today’s focus on page views, engagement times, global audience, and influencers.

That analytics pioneer, now called Clarivate, is adding artificial intelligence applications, predicting where edgy new research could feed its clients’ designs for profitable new therapies and other products.

Clarivate, with its main headquarters in London, is the successor to the Institute for Scientific Information, founded by the late University of Pennsylvania-trained linguist Eugene Eli Garfield. ISI was housed behind a University City facade tiled to look like the punch cards that fed programs to early computers.

Clarivate is now based among the TV studios and data centers in the former Smith Kline & French pill factory on Spring Garden Street. Around 500 of the company’s 12,000 global staff are based in the region.

Publishing giant Thomson Reuters bought ISI in 1992, joined it to other businesses as its science-data arm, and sold the combination to the Onex and Barings private-equity groups in 2016. They took the company public and pocketed the sale proceeds, leaving Clarivate with nearly $5 billion in debt, which it has slowly trimmed.

Clarivate counts “99 of the top 100″ drugmakers, including Smith Kline successor GSK, among its clients. Clarivate’s share price has lately wobbled between $3 and $5, down from its peak of $25 soon after it went public in 2019, after flat sales and operating losses dampened investors’ initial high hopes.

As president for life sciences and healthcare, Henry Levy runs one of the three Clarivate business groups. He took questions from The Inquirer at Clarivate headquarters. The interview was edited for brevity and clarity.

There was a lot of academic and government research generated after World War II. How do we measure it? How do we know if someone is doing good research?

Garfield created a model to measure who is using this research. He created this scoring called the Web of Science to measure [citations] .

The Web of Science became the preeminent depository and in indexer of research. We come up with our list of “citation laureates” on par with Nobel Prize recipients.

Our customers use it to see where research is showing up. Pharma companies can check it to see where to do their next clinical trial.

We have other services, such as ProQuest, which does digital literary management. … This is used in the majority of libraries, all over the world — at the British Library, the Library of Congress, Penn, Yale.

Librarians [are often] conservative; we are moving them forward into AI to make their searches faster and more effective.

We also have our intellectual property business. When I submit a new patent application, I can use this to figure out what’s already out there. And we can manage your patent and trademark renewal process. We are the biggest player in that business.

The life-sciences and healthcare business I run includes our competitive corporate intelligence business, Cortellis. If Pfizer wants to see if they should develop an obesity drug, we can show what drugs are in the market, how each of them is selling, what are the delivery platforms, what is the science — and how does it work in the body?

We can also tell you the patents that are a threat, and those that give you an opportunity. And which professors do the best research in that area.

People say the pharma industry makes drugs. What they really do is collect data to support drugs.

AI will make us better and more effective. But AI cannot make expert decisions. We are teaching AI how to help make those decisions without hallucinating — making errors it thinks are true, taking wrong suggestions seriously.

There’s a big difference. If we say that automated search gives us a return of 60%, that means it answers 60% of our questions right. For the other 40%, it returns no answer; it needs more information.

AI makes some decisions with 60% to 80% accuracy. Sounds similar. But AI returns an answer every time — you don’t know which are wrong, and what part is right.

We have figured out how to ask questions in certain patterns [“prompt engineering”] that can reduce the amount of hallucinations. We have shown we can get to 99% accuracy.

You need experts to do this. Our experts will move from running data to teaching [AI programs] how to ask questions of the data. Science is infinite, so are the questions we can ask. AI will make us more efficient.

Veeva was founded in 2007 as Verticals on Demand by Matt Wallach from Radnor and Peter Gassner from California. Like SAP [the German business software company, with U.S. headquarters in Newtown Square], their software runs a lot of processes for their clients. Harvard Business School now teaches about Veeva as a model start-up.

We make a product or two that competes with them. But we now see a lot more opportunities to partner together.

Our slowdown in growth was self-inflicted. The idea when they set up Clarivate was that bringing these businesses together could get massive efficiencies. They cut $300 million in costs.

But we did not at first focus on innovation. Customers tell us they love our products, but need us to focus on continuing innovation.

I won’t predict what’s next in the Trump administration. New approaches in some policy areas have proven to be a negotiating tactic. Our customers are waiting and watching. Right now we haven’t seen anything that tells us we will have a major issue.

It won’t happen quickly. Samsung in Korea has started new pharma manufacturing facilities. It has taken them seven or eight years. We should work with our regional neighbors Mexico, Colombia, Brazil, Canada.

Clarivate will be there. We link our clients to venture capitalists, bankers, universities. It’s fun to be in the nucleus of information. The best information wins.

This story has been updated to correct the first name of Veeva founder Peter Gassner.