Automation has become a pivotal force in the advancement of the biotech sector, ushering in a new era of AI-driven solutions.
In this episode, I'm joined by David Williamson, the CIO at Abzena and a trailblazer in biotech innovation. We're peeling back the layers of AI integration, from commercial chatbots that defy the conventional, to AI's grip on scientific research.
All that, plus a dive into the pressing need for data governance and the critical role of innovative IT leadership in propelling the industry forward.
Our exchange on the strategic incorporation of technology, and the necessary collaboration with tech giants, underlines the gravity of pursuing AI initiatives in biotech.
So tune in, get inspired, and let's push the boundaries of what's possible in biotech together.
Specifically, this episode highlights the following themes:
Links from this episode:
Automation has become a pivotal force in the advancement of the biotech sector, ushering in a new era of AI-driven solutions.
In this episode, I'm joined by David Williamson, the CIO at Abzena and a trailblazer in biotech innovation. We're peeling back the layers of AI integration, from commercial chatbots that defy the conventional, to AI's grip on scientific research.
All that, plus a dive into the pressing need for data governance and the critical role of innovative IT leadership in propelling the industry forward.
Our exchange on the strategic incorporation of technology, and the necessary collaboration with tech giants, underlines the gravity of pursuing AI initiatives in biotech.
So tune in, get inspired, and let's push the boundaries of what's possible in biotech together.
Specifically, this episode highlights the following themes:
Links from this episode:
David Williamson [00:00:00]:
Data governance is at a very difficult sell, probably in any organization, and it's a lot of work by going on your share folder. And now I'm going to classify all the different documents. I find it's very labor intensive. The more smartness you can add to the AI model, the better it will perform. So if it can utilize those classifications, the metadata about files, images, and so much, it's going to be much more important.
Steve Swan [00:00:31]:
So welcome to Biotech Bytes podcast, where we sit down with biotech it leaders and learn what's working in our industry. I'm your host, Steve Swan, and today on the podcast, I have the privilege of talking with Dave Williamson from Abzena out of San Diego. Welcome, Dave.
David Williamson [00:00:48]:
No, welcome. Thanks for having me here, Steve.
Steve Swan [00:00:50]:
Yeah, sure thing. First things first, kids, wear your sunblock, okay? Had to get a little something removed from there. So cautionary tale, please listen to your mom's wear your sunblock. So again, Dave, thank you very much for joining us. And I think what I'd like to do is kind of run through real quickly anyway, some of the things that, because what I've done and the way we formatted this is we want to do it with CIOs or heads of it at biotech, but we also gone to other leaders of it and asked them, hey, what are your thoughts? What are you thinking about? I don't think it's any surprise that on the top of everybody's list right now is for sure, and away from asking real specifics, just kind of thoughts and your feelings around AI, where we are, where we're heading, those kinds of things. Where are you on the whole AI spectrum, Dave?
David Williamson [00:01:50]:
Well, I think there are certain use cases where it just provides a great capability with very little lift. If we're talking about documentation, any sort of like proposals, rfps, you just feed it in your history and in a couple of minutes at most, it gives you out a very solid draft, first or second draft. We're experimenting around video and taking a script and having the AI convert that into a video, sort of like a cartoonish type thing for marketing actions and such. The amount of effort you'd have to have done previously for that would probably be a few weeks, and now you can get it done in minutes. So I think there's some use cases like those two that to me, why wouldn't you do this? It saves you so much time and gets you much further in the process quicker. Other things around, some of the business use cases we've looked at, like in supply chain around inventory optimizations sales and operations planning. With the right partner you can get up and running pretty quick. But I think the Chat GPT in some ways is detrimental as it gives the impression, like anyone can do this, there isn't any sort of special skills needed and I think that that's quite false.
David Williamson [00:03:23]:
Outside of the first couple of use cases. You need some specialists and there is some lift and the quality of the answers require training the model, working with the model and such. And that's not simple to do. I think also related to it is privacy, data security. So again, if you use Chat GPT, you use free version, you're opening that up to the world. That might be okay. But if you're talking about doing something, like I said, like proposals, I don't think you want your commercial information potentially being available to others.
Steve Swan [00:04:07]:
Yeah, I've heard a lot of folks talk about that. Right. The security and the governance around the free model. Right. I mean, you start putting things in there and who knows where it's going to go or what's going to happen, right, that opens your organization up.
David Williamson [00:04:24]:
It does. I think it's more from anyone can get to it and use it. If you're taking an enterprise approach, Google, Amazon, OpenAI themselves all offer enterprise versions where it's secure to you and they're not using the learnings more broadly.
Steve Swan [00:04:45]:
Right. Okay, now you mentioned when you were talking about that there that it's not as easy as, I don't know, plug and play where anybody can do it. Do you need an it person to do that or who's actually the specialist?
David Williamson [00:04:58]:
Well, I think there's a few specialists you need. So first off, I need the data that I'm going to use to ingest into the model. So if we're talking about proposals that might be scattered in different sharepoints, shared folders, there might be a bit of work in collecting that. If you're talking like the inventory, then I need to connect to my ERP system. So there may be data engineering, data access work that needs to be done. There obviously could be data science work. How do I want to train the model? What is the best model to use? These large language models or llms are very complicated to build, but then there's dozens if not hundreds publicly available. Which one do I use? I mean, you don't think about those questions when you go to Chat GPT and you just start throwing stuff in.
David Williamson [00:05:55]:
But if you're going to use this for an enterprise purpose and you're looking to gain business value, then those become important. So having someone who understands it and understands why I would want to pick a overseas as an example right now.
Steve Swan [00:06:09]:
Am I supposed to also talking about the security and the governance, what are your feelings around? Am I supposed to be folding this into as I look for somebody that's my security person or my governance kind of person, or auditing security, depending on the size of your organization, you're going to have different, maybe different folks doing that, smaller organization, you're going to have the same person doing all that. Right? So do I need to fold that in? Does Steve Swan need to fold that in? When Dave says, hey, find me somebody that know handle my technology security, do I need to start thinking about that?
David Williamson [00:06:47]:
I think absolutely. I think with any third party that you're working with, whether it's a subscription, as a service or externally hosted, you need to think about those topics. It also with the AI gets into your whole data governance, data classification. So maybe you're sophisticated in that area and so then you can apply security or access based upon how the data is classified, which could be the inputs or the outputs that the AI is working on. So I think there's quite a few considerations to do this at scale. Most endeavors like this I would advocate to doing something, a pilot proof of concept that's pretty focused with a small group of individuals. And then in that you should gain a lot of learnings about what is it going to take for this to be successful. What were the lessons to be learned around engaging this? But there's quite a few software as a service offerings now that are tailored to different particular business functions like marketing versus supply chain.
David Williamson [00:07:58]:
And I think the main contributors like Amazon, Google and such, they have their own technology teams. They're very willing to get involved and help you with these use cases.
Steve Swan [00:08:11]:
So we're still really figuring out what the right platform is or what the right, I guess, end company is that's going to supply our industry for this, aren't we?
David Williamson [00:08:24]:
I think so, because again, there's layers to this, layers to how do I get started and then what model am I going to use? Do I need to build an industry specific or a specific use case model? I've seen some very detailed presentations from folks generating value from this and I was surprised at the level of effort. Talking to a mergers and acquisitions financial firm that said they had 30 offshore developers, that this is all they were working on, was really continuing to tune their model for doing due diligence, examining the potential prospects financials and why did sales decline in February of this year? Those sorts of questions and I thought like maybe three or four specialists, but having a whole sort of army of developers took me by surprise.
Steve Swan [00:09:28]:
That's data science, that's analytics. They're doing that kind of work with that. Right. A lot of the use cases that I've encountered or seen or have heard about are more along the lines of automation. Right. And things like that, automating processes and automating, like you talked about maybe doing some of those proposals and things, but I hadn't heard much yet anyway, about the analytics and the data science piece of it, which only makes sense. Right. And finance firms are typically at the front end of this whole thing, so that's interesting.
David Williamson [00:10:00]:
Yeah, the automation is another more simplest. There's several vendors around automated AI enabled chat bots, either for internal or external. So internal, how do I find my vacation day balance? Where's my paycheck located? Being an ask, especially if you're growing, you get new employees in. You can deal with a host of these HR employee related things very easily with these bots from an external focus. Like we all see this on websites you visit, where do you want to chat and having it, instead of just giving you two answers, having it be more interactive and learning from those interactions. So there are quite a few solutions available, commercially available, yes.
Steve Swan [00:10:52]:
Well, when I look at these things or when I'm thinking about this. Right. So I guess I'll just ask it how I'm thinking about it. Would a small company get involved with the chat bots? Is it economically feasible for a small company to start doing all those? Because you've been at both, right, big and small, and the problems get solved, I think different ways. Right. The bigger companies, they have a little more money to spend on things. The smaller companies, you got to worry a little bit more about what you're spending. But I think with the automation, I guess what you're saying to me is it doesn't really matter, big or small.
Steve Swan [00:11:27]:
I mean, the cost, is the cost.
David Williamson [00:11:29]:
Okay, no, and I think the cost scales pretty appropriately to the size of how many people am I talking about interacting with it? But also where's the information that I want it to find its answers from? So in our case, I can point you to a couple of individuals who will know where everything is, and you get the vendor to set the access up to the bot and it runs. We're very interested in pursuing some of those, especially for internal use cases. Our whole revenue model is client projects. So I set up a new sharepoint site for every client. So being able to finding. I'm going to do a new project. Have we done something similar to this before? The answer is probably yes. Finding the information is another challenge.
David Williamson [00:12:25]:
So if I can automate that versus somebody just spending time, I'll check this site, I'll check these folders. I think we all experience that in our personal lives, too. I know I have that document somewhere. And you know how Microsoft searches it goes to 100,000 records found, right? So having something that actually can go through the documents and get that down to like ten, because then I'll spend the time to look through ten. You tell me 100,000, I'm like, oh.
Steve Swan [00:12:54]:
Well, I'll just recreate imagine on the research side, right. Scientists. Now, I'm Steve Swan, biotech. Right. And I'm probably not sharing my research with Dave Williamson, Biotech. Right. But at least for my internal scientists, if I have reams and reams of data, they can do those kinds of searches. Has anybody that's been here, worked here currently or whatever done this experiment or something similar? And can I find that and can I not reproduce that to waste my time and the company's money?
David Williamson [00:13:30]:
Yeah, I think there's a couple of business use cases for that application. If I'm an instrument vendor, medical device vendor, finding people doing research in the area where my instrument would provide value or would be used in such a study, because then I can reach out to you and say, hey, Steve, I saw you put a proposal out for a paper around genomic sequencing. I sell genomic sequencing instruments. Have you bought the instruments? Are you interested in using our instruments or trying one? I think also around other applications where I provide a service or a product that could have value to that study, or say, I'm doing a clinical trial and I'm looking for people interested in a particular therapeutic area. So getting connected to the right researchers, because you know how it is, once you're a well known quantity, then everybody wants to work with you. If you're more newer, then this is a way to get discovered. Is somebody using AI to troll through all the scientific proposals? So I think that I have seen from instrument vendors where they're exploring that and trying to use that as a prospecting tool.
Steve Swan [00:14:52]:
Now, the next thing that comes to mind as I think about all this, right, is that the AI is, right, the shiny object that everybody's building. But the inputs, the data, right. The data is the most important piece of any of this stuff. Junk in, junk out, right? So if your data is not good. So now are we talking that we've got to spend more time and emphasis on having data organizations inside. What are your thoughts about that?
David Williamson [00:15:19]:
Yeah, I think data governance is a very difficult sell, probably in any organization, and it's a lot of work. I go in on your share folder and now I'm going to classify all the different documents. I find it's very labor intensive, but then that adds a lot of the more smartness you can add to the AI model, the better it will perform. So if it can utilize those classifications, the metadata about files, images and such, yeah, it's going to be much more important. So I think you're correct in that statement that data governance becomes much more critical and organizations are going to invest in that because it will pay for itself.
Steve Swan [00:16:10]:
Yeah, I think they're going to have to, right? Yeah, they're going to have to, because all these tools, what you hear about and what you read about is you've got to give it exactly what it needs to use to calculate or do its thing. It's not making inferences, it's taking the data you give it and the specific instructions you give it and executing on those instructions based on your data. Right.
David Williamson [00:16:31]:
Well, I think degenerative AI does make inferences, but the quality of those often needs to be human guided. So you use Chap GPT, you give it data, you ask a question, then you often ask another question and another question. But I've seen models where instead of you asking the next question, it asks you, and then it's learning and the quality of the questions and such you can train. So I think the value of it goes up quite a bit. The more you work and train the model and having the subject expert working with it along with the data scientist. Yes. So back to the data governance, the inputs, they're all critical to the quality of what does it give you?
Steve Swan [00:17:33]:
Right. And so again, my brain just keeps going to, okay, now we've got this new shiny thing, we've got AI, we've got this data thing, right? But we're still only probably, maybe I'm wrong in this and I haven't spoken to a lot of folks about this, but we're still only going to get a fixed amount of budget for our whole budget. Right? There's still the low single digits. Right. So given that, does that mean that it's a zero sum gain, there's not unlimited funds? So are we talking erps are going to, I mean, what are your thoughts and feelings around CRM? Who's going to suffer? Something's got to get taken away somewhere. Right.
David Williamson [00:18:11]:
Well, any good it project has a strong business sponsor. So I need a problem that I want to solve. So when you hear the anecdotes, which are very true, I've experienced this. Where's our AI project? That's not going to go very far. And yes, you could spend a lot of money. Like with most things, you could spend a lot of money and not get much for it. So I think it's having a clear business problem that fits with the technology and that you have a sponsor who is willing to spend time and also just again, a proof of concept, a pilot where I try to solve one facet of the problem. Success on that would then build the momentum and the funding to solve other problems.
David Williamson [00:19:02]:
I think often what I have found around the automation is we have very expensive resources that are spending time doing very routine tasks and freeing them up is like creating free headcount. I've been in situations where we have PhD level experts examining high resolution photos, looking for defects and how many photos can I show you? Then your error rate is going to start going up and up. And now using the AI to here's the five things on the photo and highlighting them that I want you to look at. Not only can you, and with AI I can process 100% of the photos, not just the sampling that you would do. And I'm directing you to here's the things to look at. So I think that often you are generating value and very quickly that pays for the investment.
Steve Swan [00:20:05]:
Sure, yeah, no, that makes sense. The ROI is there. And so then what you're saying is the other corporate functions won't suffer. Like I said, the commercial side, the supply chain side, the manufacturing side. From an IT perspective, you don't think they'll suffer much?
David Williamson [00:20:22]:
Well, I think the innovative companies with the innovative IT leadership have some sort of funding on a yearly basis for innovation, exploring new technologies. I haven't found it difficult to fund at a small amount. Again, the challenge to me is more getting the business involved, that I should spend time on this and I will get the value. Often you have this impression of it, you just want to go buy a new toy and play with it, right?
Steve Swan [00:21:01]:
Do we have to be more reactive to the bit? I guess you do, because everybody's asking, I'm sure everybody's asking it about AI right now.
David Williamson [00:21:10]:
In that instance, yes. When you get into now we want to automate this or we want to apply the technology in an innovative still, you still have the often answer of Steve's really busy, he can give you an hour a week, sort of. I think, you know, like many things, having the right level of commitment. And again, Chat GPT often created a lot of problems because people's perception is, well, you just set it up and it works.
Steve Swan [00:21:49]:
Right.
David Williamson [00:21:50]:
But when you're dealing with complex business problems, especially problems specific to your company, your industry, there's work involved to get it to that point.
Steve Swan [00:22:00]:
Right? Yeah. And there's a lot of work to get it to that point. Right. There's a lot of inputs, too. Like we already hit on the data and things like that. Yes. Okay. Now, if you go and try and find somebody that's going to do, I don't know, your AI or your Chat GPT work for you or internally from an it perspective, what are your thoughts and feelings around that? How do you do that? Or is that not something that you have to get a specialist for?
David Williamson [00:22:29]:
I do think this is a specialty. It seems pretty easy to find a dozen or so different firms that state an expertise in this area. I've been doing it primarily through my network and reaching out to people who have done some things. But, yeah, that's difficult, asking for references and specifics around what use cases did they provide a solution for. How was the solution architected? How did it work? How are we going to maintain it?
Steve Swan [00:23:11]:
Right. That's a big one, I'm sure. And again, just to go back to coming back to the whole data side of things, smaller organizations don't necessarily have the bandwidth or the dollars. Right. To have a whole separate data group. Right. Bigger companies are going to do that. Right.
Steve Swan [00:23:31]:
But you've got a smaller company, probably has to get vendors right with their data and trust that it's going to be right. If it's not right, that's where your AI goes awry on you.
David Williamson [00:23:41]:
To me, being in biotech, folks like Google and Amazon, they're interested in working with light science. They've done, say, health initiatives in the past. So again, I don't think the cost is as much a hurdle as having a clear idea of what would be an exciting solution for you and your company.
Steve Swan [00:24:05]:
Right. Yeah, I've seen a lot of different ones. Right. And I've seen, again, a lot of automation through AI. That's where a lot of this has gone. All right, well, cool. Well, so it's been AI, it's been security, it's been the governance around it and then the funding. I think we hit on most of that.
David Williamson [00:24:22]:
We covered a lot.
Steve Swan [00:24:24]:
Yeah, we did. We covered most of that. Now, from technology, biotech perspective, anything that you'd like to cover about your roles, what you've done, maybe what you'd like to see others chat on. And again, we can edit this stuff, so that's not a big deal. Any thoughts on anything that you'd want to hear more about as I continue through my journey talking to leaders like yourself, Dave?
David Williamson [00:24:51]:
Well, to build upon our previous conversations, to me some of the more being a chemist by training a long time since I've been in a laboratory, is how do we apply these technologies more on the research end? How do we solve very complicated. You talk about biology and the complexity of the therapies being developed. How do these therapies interact within the body and gaining insights that move those forward? There's got to be thousands of unmet medical needs, and how can I get help to patients sooner, and how can we use these technologies? In a way, in some ways a bit disappointed out of COVID with the rapid approval of the vaccines for Covid, it didn't change the underlying way we approve new medicines in the world that hasn't been repeated. So how can I use data and these algorithms, models to help speed approval and help, again focus the effort of the experts on something that has a high probability of working versus our more sort of random. I have an interest in this. I'll try it out and see how it goes.
Steve Swan [00:26:22]:
Right, yeah. And to dovetail off that, I know a lot of companies have been trying to, whenever you hear a CEO or somebody getting skewered in DC about why does this cost so much? Well, it's heart length. The research and development side is so long, rightfully so. It takes a long time to get through this stuff. But what we're talking about here is if we can use technology to lop off, I don't know, six months, a year, a year and a half, whatever, that's huge for everybody. To your point, it gets things to patients, folks that need them quicker, right?
David Williamson [00:27:00]:
Yeah. So it's something like what, 6% of programs actually become an approved drug if I can make that twelve.
Steve Swan [00:27:08]:
Right?
David Williamson [00:27:09]:
Because now I've just doubled the number of winners, so I don't need to make as much money off the winners.
Steve Swan [00:27:15]:
And all the way along the way, you're using this, right? We just talked about a use case where you can pull some of that research. I don't have to do that again. Steve doesn't have to do that again because somebody did that six years ago here, and it either worked or this part of it worked and this part didn't. That kind of mean.
David Williamson [00:27:31]:
But often the reaction I get is, I've been doing this kind of science my whole career. I don't know how you're going to program something that's as smart as me. And my experience is, yes, actually we can. Or it's smart enough to direct you to. Like, my previous example to these four things versus the hundred I would have given you.
Steve Swan [00:27:56]:
I think my reaction to that would be baseball has been around a very long time, too, and they have analytics in the dugout. Right. Oakland athletics, Billy Bean. Right. Billy Ball. Does that ring a bell with anybody?
David Williamson [00:28:15]:
Well, often the data and the analytics give you an answer you didn't expect.
Steve Swan [00:28:22]:
Right.
David Williamson [00:28:23]:
And I've seen that in my work career, where the model gives you an answer and it does not make sense. It's like, that can't be. That runs across accepted truths. And then when you dig into it.
Steve Swan [00:28:39]:
Yeah. I was going to ask you what happens when you prove that out. Yeah. Okay.
David Williamson [00:28:43]:
Then people get really excited, and then you have way more demand for machine learning and automation than you can deal with, which is a great place to be.
Steve Swan [00:28:52]:
Right. Yeah. Be careful what you wish for. Right.
David Williamson [00:28:56]:
It's all good in the end. It's all good. And it gets very high level attention very quickly. Well, I've seen it in both ways. Prove sort of what I call urban legends, where the salespeople believe certain there's good instruments and bad instruments, and I only want to sell the good ones. And then when you prove that that's actually true, but it's not necessarily statistically significant. So I proved both, like proved urban legends and then proved disprove common belief to where now people understand that, oh, there's a particular behavior I wasn't taking into consideration.
Steve Swan [00:29:41]:
Yeah, no, that's good. Yeah. To your point, when it comes up with that surprise answer, that does get proven out and people are like, wow, we're looking in the wrong spot. We're looking in the wrong corner of the room for the gold. It's over. Know that kind of so. Well, you know, I'd like to thank you for coming on with us, Dave. I want to ask one question, though, and it's kind of a fun question, but I want to ask it anyway.
Steve Swan [00:30:13]:
Okay. Do you like music? Yes. Okay. Who was your favorite band? You ever saw live?
David Williamson [00:30:23]:
Oh, that I ever saw live. I'd have to say the Ramones.
Steve Swan [00:30:27]:
Really?
David Williamson [00:30:31]:
It wasn't in the. They were a little older, but it was so much fun. Like they were having a great time. You could tell they were smaller venue they were having a great time on stage. We're all having a great time out on the just. I still remember that one.
Steve Swan [00:30:48]:
Those guys are pioneers and they're revered by a lot of different bands. I recently saw a band, Pearl jam as a matter of fact and they love those guys. They talk about them all the know. I think the Ramon's probably played down in City gardens too, in Trenton but I can't remember if I read that or not properly.
David Williamson [00:31:04]:
Yes, I've seen a few in City gardens and then you're from New York area so cbgbs back in the day and another venue that launched so many brand names.
Steve Swan [00:31:18]:
It did, yeah, it was awesome. Good stuff. Well, thank you very much Dave.
David Williamson [00:31:23]:
Welcome.
Steve Swan [00:31:23]:
And anybody that wants to learn more about Dave, his link will be here on the podcast. Mine will be as well. You can get in touch with either one of us at any point.