Navigating Forecasting Pitfalls in Pharma and Beyond: Joseph Sterk's Expertise

[00:00:01] Welcome to the Chalk Talk Jim podcast, where we explore insights into healthcare that help

[00:00:07] uncover new opportunities for growth and success. I'm your host, Jim Jordan.

[00:00:18] Welcome to Chalk Talk Jim, where we explore the evolving landscape of business and innovation.

[00:00:24] Today's guest is Joseph Sterk. He's a seasoned expert in pharmaceutical forecasting with a unique

[00:00:30] background that merges scientific research and business strategy. With years of experience

[00:00:35] working in both equity research and consulting for top pharmaceutical companies, Joseph has

[00:00:41] had an insider's perspective on how accurate forecasting can drive success in the healthcare

[00:00:46] industry. Now in this episode, Joseph dives into the complexities of forecasting in the pharmaceutical

[00:00:51] industry, covering everything from the role of AI in data analysis to the pitfalls of market

[00:00:57] bias. He also shares insights into how to evolve business models and products and how they're

[00:01:02] reshaping the future of healthcare. If you're all curious about how the pharmaceutical industry

[00:01:07] stays ahead in a rapidly changing world, this conversation is for you.

[00:01:12] Well, Joseph, tell me a little bit about yourself and your background.

[00:01:15] All right. Like a lot of people, I come from a scientific undergrad background, but I gradually

[00:01:22] fell in love with business and finance and sought to merge them together. I really started in forecasting

[00:01:30] inequity research at Deutsche Bank and RBC Capital Markets covering the biotech sector. That's where I

[00:01:37] learned what drove the stock, the business. From there, I went to consulting and product management with an MBA

[00:01:45] sandwiched in there. For the last six years, I have been a forecaster for Beringer-Ingelheim's

[00:01:52] emerging pipeline across therapeutic areas. I should note that I'm not speaking on behalf of Beringer-Ingelheim

[00:01:59] or any other employer or client past or present. To any resemblance to past work is strictly coincidental.

[00:02:05] So let me pause and just have you express your education because I find it interesting. You're the second

[00:02:10] person in a few months that I've talked to that started out in the science and ended up on the forecast side.

[00:02:16] And what strikes me is disease states come together and probabilities of this happening and that happening

[00:02:21] and understanding disease states. So I guess it's a natural fix, but please share with the audience your background.

[00:02:26] Like a lot of people, I thought I would be pre-med or pre-grad school, but decided against going that route and

[00:02:35] really developed a fascination with the business side of pharmaceuticals. I came of age in the first biotech boom

[00:02:42] when the human genome was first being sequenced. A lot of people thought that you sequence the genome or you're

[00:02:50] going to do something like that and money is going to or out of it. But then when a lot of these companies

[00:02:57] started trying to monetize it, people started asking, oh, so how are you going to do that? And then they had an

[00:03:03] identity crisis. They were forced to look back and say, okay, how are we making the money from it?

[00:03:09] So learning how to monetize scientific advances became fascinating.

[00:03:14] I think my other thought on that is that we thought the genome would unlock certain things. And now I

[00:03:21] think we're recognizing that we don't have enough data on proteins, right?

[00:03:25] Yes. I think that was one part of it. Of course, sequencing the genome is

[00:03:31] relatively straightforward compared to proteins. You have four bases, two strands, and one serves as a

[00:03:38] template for the other. Proteins don't have that. There are 20 of them and so many post-translational

[00:03:44] modifications and confirmations that it gets very tricky. It gets very unrolling. So tell us about

[00:03:51] the function of forecasting within the continuum of healthcare in general, because you've been on

[00:03:58] the analyst side, you've been in the company side. Just share with the audience the implications of that

[00:04:04] function. Forecasting is, as I've heard, called a strategy put into numbers. It is meant to be

[00:04:13] a decision support tool for management. That's why it goes hand in hand with functions like market

[00:04:20] research, finance, operations. It forms a backbone for a lot of decisions.

[00:04:28] So when you work for a pharmaceutical company or a bank, what's the different information that

[00:04:35] they're looking for? How do they use this function differently?

[00:04:38] So at a bank, it is a lot more about how the revenue and profitability of the stock have to

[00:04:46] do with the stock price and the multiples and things like that. I would say in the biotech sector,

[00:04:52] in particular, in smaller companies, it's a lot less about profitability and a lot more about

[00:05:00] will this drug make it to market and when will this drug sell. Whereas in a pharmaceutical company,

[00:05:08] you're interacting with so many different functions. Finance and resource planning,

[00:05:14] like the forecast that I've been working on are going to be used to determine how much is spent on

[00:05:19] a given project. There's a lot more scenario planning and risk analysis that needs to be done.

[00:05:26] In pharmaceuticals, you're also working on a lot fewer products. When I was in equity research,

[00:05:32] the team covered maybe 20 companies, a lot of whom had multiple products. Whereas right now,

[00:05:40] I primarily work in three therapeutic areas that are fairly well related to one another.

[00:05:47] A wonderful line when we were talking before we started this regarding data and forecasting,

[00:05:52] I'd love you to share with the audience. I thought it was pretty awesome.

[00:05:54] It was around the concept of data forecasting, not taking into consideration the realities of people

[00:06:03] in real life. Yes. And so I was prepared a lot less to talk about quantitative techniques for forecasting.

[00:06:10] I think a lot of the math is actually pretty simple and more in terms of the pitfalls of both

[00:06:17] constructing and evaluating forecasts. Because even if you don't have forecasting in your title,

[00:06:24] you are probably consuming them and it really helps to know what the different motivations

[00:06:29] are behind them and how they're going to be used. So what are the pitfalls of forecasting?

[00:06:34] I think a lot of them to start is the math is not difficult, but sometimes people make the math out

[00:06:42] to be difficult. Revenue equals price times quantity. You subtract cost of goods, tax, et cetera,

[00:06:49] and you get the profitability. A lot of people, they don't know if they're going to do it on price or

[00:06:57] on quantity. You need to have one or the other. If you're not getting reimbursed, you don't have the

[00:07:03] price input or you're going to lose money every year. There's no demand or unmet need, or you

[00:07:10] overestimate the market for your product, you will not have revenue. I've heard a lot of people over

[00:07:17] the years try to put their own personal biases behind it, whether it's don't be so negative or

[00:07:24] this is the optimistic forecast. You're not helping your employer or your bank with decision support

[00:07:31] if you don't realize these biases. I remember many years ago being a young product manager.

[00:07:36] The CEO of the company kept saying, this is a $5 billion market. When you're a young product

[00:07:41] manager, how do you fight you CEO of a publicly traded company that it's a $5 billion market?

[00:07:46] What I was finally able to do in a presentation to get his team to question him is I said,

[00:07:50] it's not that there's one data point that says it's a $5 billion market. There are 10 data points

[00:07:55] that say it's going to be a billion dollar market. That caused people to reflect and say like any data

[00:08:00] set, the argument isn't whether there's a data set that represents that. The argument is whether that

[00:08:05] is an overly optimistic, unoptimistic, or too risky data set to use. And so my sense is in some of the

[00:08:13] presentations you were showing me, you were showing me a one drug company that said sales were going to

[00:08:17] be X and it was 10X to that. And another company that said it was going to be a number and it was

[00:08:22] 10X minus that. You want to share some thoughts on your experience in that area?

[00:08:26] Yes, there are a lot of them. All the market research in the world can be twisted or biased

[00:08:32] by its design and that will lead to some very biased results. Just like you can say revenue

[00:08:41] forecast is price times quantity or forecast is structure times inputs and bad structure and bad

[00:08:49] inputs are multiplicative, especially if you're talking about a new market, rare diseases and biotech.

[00:08:56] There's so little known about these diseases. There are constellations of seemingly unrelated symptoms.

[00:09:03] You get prevalence ranges that are literally 10X and your 4% chance it's 10,000 patients.

[00:09:11] The 90% chance it's at least 2,000 patients. There's a lot more judgment that comes into that.

[00:09:18] So do you find that you commented earlier that it's not the data analysis that's the critical thought?

[00:09:25] So if I'm listening to this conversation, it seems to me then the input is the critical thought.

[00:09:29] The input and how the input is connected to the structure, how those mathematics are being performed.

[00:09:37] Someone sent me a really nice graph that said precise number plus garbage equals garbage or precise number

[00:09:44] times garbage equals garbage. Anything that is relatively unknown is a liability that forces you to admit

[00:09:54] your forecast or to use mathematical techniques to Monte Carlo to give a probability spread.

[00:10:01] A lot of our audience is not from the pharmaceutical industry. They may be thinking that's not applicable to them, but looking at your background and the things you post about, you have a broader view.

[00:10:12] You think of diseases like obesity, home care and moving patient flow and different things like that.

[00:10:17] Do you want to share some of your visions with the audience and how you would apply techniques in different spaces so they have a sense of learning from your experience?

[00:10:26] So my expertise is definitely pharmaceuticals, but revenue being price times quantity is universal.

[00:10:34] Whether you're selling a drug, a diagnostic test, an MRI machine, that is always going to be the case.

[00:10:42] So we're coming up near the end of the year and insurance companies will start trying to figure out what their rates are for next year, right?

[00:10:49] So they look at their populations and in the craziest example, imagine if my total population, 10% of it had cancer next year, right?

[00:10:57] My rates would have been awful and I would have lost my shirt.

[00:11:00] When you look at populations, which is what you look at when you're promoting a drug, how do you look at that?

[00:11:06] There's a bit of public health in that looking at general populations and seeing how these diseases move.

[00:11:11] You've got prevalence and incidence and timeframes as we extend some diseases, keeping them open a lot longer with different treatment sets.

[00:11:19] It becomes very complex, doesn't it?

[00:11:20] Extremely complex.

[00:11:22] And I think that a lot of insurers in particular, they fall into the trap of one bucket for drugs.

[00:11:30] One bucket for procedures or hospital stays.

[00:11:34] If a drug costs $1,000 and it prevents or reduces the number of $50,000 hospital stays, that kind of thing can be a screaming bargain.

[00:11:45] In your career thus far, can you share with the audience when you've had to pivot and adapt in terms of which direction you were taking your career

[00:11:54] or technologies that changed your forecasting methods?

[00:11:58] I think of AI now.

[00:11:59] I'm sure people are asking, what does AI mean to this kind of thing?

[00:12:02] Yeah, I think AI and forecasting is still very much in its infancy.

[00:12:07] And it comes less into play in terms of are not particularly difficult.

[00:12:14] But in looking at things like trying to find the right analog, integrating, say, quantitative and qualitative expertise,

[00:12:24] that's where I think AI is really going to shine in this area in terms of identifying patterns that are not readily apparent to the human eye.

[00:12:33] Going back to your about inputs plus garbage equals garbage, it seems to me that AI might be really applicable in the input end

[00:12:41] in terms of making sure that a broader base of representative inputs or maybe some sense of outliers.

[00:12:47] I think one of the interesting use cases I saw was one of those rare disease companies.

[00:12:53] They have a drug for a syndrome, one of those acronym syndromes, which usually means it's extremely rare, disconnected constellation symptoms.

[00:13:03] They have the probability that these individual symptoms, the letter of the acronym, occurred independently or together.

[00:13:12] And that to the number of front physicians who actually recognized it.

[00:13:17] I thought that was an extremely clever tool to get at the prevalence of something that there is very little published literature about.

[00:13:27] How do you keep current on all the rapid changes in this space?

[00:13:31] I consider myself to be something of a nerd for the pharmaceutical industry.

[00:13:35] I read press releases for fun, industry journals and things like that.

[00:13:40] And I think that if you see an interesting use case or a presentation, you just dive into that and think about the application.

[00:13:50] So if you had to frame the state of the union for this industry right now, what would you say?

[00:13:55] I would say with respect to AI, we're still in the very early stages of adoption.

[00:14:01] It will certainly have a disparate impact in some places versus others.

[00:14:07] My bias is towards the pharmaceutical pipeline and the use cases that I thought of.

[00:14:12] But there are also for inline forecasting, looking at patterns, looking at seasonality, and particularly looking at analogs from the past of market shocks, demand shocks that couldn't be predicted,

[00:14:25] and how that would play into the current forecast in order to match supply and demand.

[00:14:31] Are there particular resources you could share with the audience?

[00:14:34] They'd love to follow up on how people research things and use some of the sources they use.

[00:14:39] There are some big consulting companies developing AI-related inputs or models.

[00:14:46] If you look at some of the big healthcare consultancies and the white papers they've put out, that is a great place to learn.

[00:14:54] Do you have some favorites?

[00:14:55] I'm thinking McKinsey, Boston Consulting Group, MIT.

[00:14:58] What do you use?

[00:14:59] Yeah, I tend to prefer the more healthcare-focused consultancies.

[00:15:03] Things like ZS and IQVIA come to mind immediately as having put out some very good work.

[00:15:11] LNK had some positive experiences with as well.

[00:15:14] Are there particular people you follow in the industry?

[00:15:17] A little less so on the precise forecasting methodology.

[00:15:21] I look to see what's out there and keep learning.

[00:15:24] What do you see as the biggest opportunity for growth and the biggest threat in the next several years?

[00:15:30] With respect to forecasting in general?

[00:15:32] In healthcare in general.

[00:15:34] I know I saw a beautiful commentary you wrote on the complexity of obesity and the number of,

[00:15:40] I call it the spindles of disease that come off of that.

[00:15:43] I wonder, given your sense of these clusters of data that have implications in several different areas,

[00:15:50] if you see any that are particularly promising to you.

[00:15:54] I think the industry as a whole is extremely exciting.

[00:15:58] Of course, there's always going to be clinical trials failing.

[00:16:02] You hear a lot about AI-related drug discovery or certain platforms.

[00:16:07] As a pharmaceutical investor, not a platform until you develop three drugs.

[00:16:14] No matter how a molecule is developed, it still has to go through clinical trials.

[00:16:20] The same as any other drug.

[00:16:22] New drugs raise the bar.

[00:16:23] Lately, we've seen some high-profile failures with new immuno-oncology drugs,

[00:16:29] once PD-L1s have just raised the bar so high.

[00:16:33] I can think, with respect to overall healthcare spend, like you mentioned,

[00:16:38] if you think about commentary that the government has said,

[00:16:42] the concerns about, first it was hepatitis C drugs and Alzheimer's drugs,

[00:16:48] and now obesity drugs, possibly bankrupting the system.

[00:16:53] You pull on one thread and there's just so much to come out.

[00:16:56] Obesity is the latest incarnation of it.

[00:16:59] I read something that you wrote regarding the amount of spend on pharmaceutical development

[00:17:05] and the cluster of areas that had major impact.

[00:17:11] Would you be able to share your insights into that area?

[00:17:14] So we just talked about obesity, and I think there are so many obesity-related diseases.

[00:17:19] If you look at the phase three programs for all the GLP-1s and their derivatives, multi-agonists,

[00:17:26] you'll see they're running trials in not just obesity, weight loss, but liver disease,

[00:17:33] obstructive sleep apnea, osteoarthritis.

[00:17:37] We're really seeing something that goes after the root cause of so many different diseases.

[00:17:43] And I think that as better drugs come out for conditions like Alzheimer's,

[00:17:48] we're going to continue to see that.

[00:17:50] We're going to see more of healthcare spend as an integrated whole,

[00:17:55] rather than just keeping these disconnected buckets.

[00:17:59] Yeah, that's really profound.

[00:18:00] Anything else you'd like to share with the audience?

[00:18:02] I'd like to talk a little more about some of the biases that I've seen in forecasting and some of

[00:18:08] the errors.

[00:18:09] If you picture a two-by-two matrix, you have Donald Rumsfeld's and unknown unknowns to oneself and

[00:18:17] others.

[00:18:17] That kind of thing can help look at the motivation behind a forecast.

[00:18:22] For example, in investment banks, they have a very strict wall between investment bankers and

[00:18:29] equity researchers.

[00:18:31] Yet it's pretty well known that a lot of companies will not do banking business with any bank that has

[00:18:37] rating.

[00:18:38] The pressure to issue a better rating doesn't need to be explicitly communicated to know that it's there.

[00:18:44] If you look at things that are known to oneself, but not to others, I have seen companies with

[00:18:52] forecasts that they know are wrong.

[00:18:55] For example, if they're using the wrong analog, they chose the wrong endpoint or their market research was poorly designed or just a senior executive's project that they didn't want to let go.

[00:19:07] Major decisions are being based on them.

[00:19:10] As a forecaster who often presents several steps of the ladder, my favorite phrase is, are you prepared to believe X, Y, Z?

[00:19:18] Okay, you can say this is going to be a $5 billion drug, but you're assuming that all these inputs are the very most optimistic and that's what you're signing on to.

[00:19:28] The harder ones, there is more room for error.

[00:19:31] The ones that we don't know, but someone else knows, like actions of competitors, regulators, and investors, or the unknown unknowns, things like disease states, clinical trial failures.

[00:19:46] And the only way to prepare for those is to have analogs at the ready and sensitivity analyses, just to really be prepared for what is going to come up.

[00:19:57] As a consumer of forecasts, you really have to know what you're looking at.

[00:20:03] Anything else we should explore?

[00:20:04] I think that kind of thing covers it pretty well.

[00:20:07] Revenue equals price times quantity, precision plus or times garbage equals garbage, and know all the biases that come in and the psychology of it.

[00:20:17] So if you have to give one quote, an old boss said, all forecasts are wrong, but some are useful.

[00:20:23] How can the audience find you and read some of the things you write about?

[00:20:27] You can look me up at Joseph Sterk on LinkedIn.

[00:20:31] My personalized URL is Joe Sterk, J-O-E-S-T-E-R-K.

[00:20:37] Perfect.

[00:20:37] Thank you very much.

[00:20:38] Thank you so much for being a guest.

[00:20:40] My pleasure.

[00:20:40] Thanks for tuning into the Chalk Talk Gym podcast.

[00:20:46] For resources, show notes, and ways to get in touch, visit us at chalktalkgym.com.