From Reactive Data to Predictive Intelligence: Inside Convoke's Platform | Alex Telford (4/4)

Intro - 00:00:06: Welcome to The Biotech Startups Podcast by Excedr. Join us as we speak with first-time founders, serial entrepreneurs, and experienced investors about the challenges and triumphs of running a biotech startup from pre-seed to IPO with your host, Jon Chee.

In our last episode, Alex shared the coordination tags burdening pharma, why tacit knowledge is the industry's most undervalued and most endangered asset, the electrification analogy for why AI's efficiency gains require full organizational redesign, and why disrupting pharma demands either breakthrough science or a new commercialization model. If you missed it, check out part three.

In part four, Alex unpacks the full founding arc of Convoke: the cold turkey exit from consulting, the writing and prototyping period where he discovered the problem was deeper than any point application could solve, and the co-founder connection that came through a podcast that came through a piece of writing. He also breaks down Convoke's platform architecture and customer approach—building validated structured datasets from trial registries, regulatory filings, and patents; stitching them together with workflow tools for competitive intelligence and document generation; and embedding closely with customers to encode their specific processes—before sharing his roadmap from reactive data architecture to proactive predictive intelligence that doesn't just store what your organization knows, but tells you what to do with it.

Jon Chee - 00:01:56: Absolutely. Be proud, but, like, don't get paralysis to the point where you don't do anything.

Alex Telford - 00:02:01: Yeah. And I think now, you know, pre-LLMs, I was much more you should just put out a bunch of stuff, and some of it's gonna take off and whatever. It's just a volume game. Now I think I actually feel like you want to really put out stuff that's very high quality.

Jon Chee - 00:02:17: Oh, for sure.

Alex Telford - 00:02:18: Because nothing else is gonna get traction. Like, you actually do want to spend six months writing the best thing that's ever been written about a given topic. And I actually think there's a lot of room to write the best thing about some topic. Like, I think even very well covered topics are generally not covered very well.

Jon Chee - 00:02:35: Yeah. Yeah. Yeah. Like, right, everyone's just using AI to augment their writing, and everything is just mid and smooth, I think, is what you—

Alex Telford - 00:02:41: Yeah. Yeah.

Jon Chee - 00:02:42: Described it. It's very smooth. But, like, I mean, look at stuff like—oh, this is not writing, but just, like, look at the Acquired podcast. Like, let me just, like, do six months of research on a given company.

Alex Telford - 00:02:54: Just like I'd give a piece of advice to someone in university who wants to write a blog. My advice now would be spend six months writing the best piece that's ever been written about something and go really, really deep and go into the details because LLMs are not good at details generally. Or maybe they can be, but someone just naively prompting them is not gonna get them to be good at going into details. And, like, go interview people and collect new data and present a differentiated perspective and, like, really try to write the best piece. Like, good stuff tends to get noticed, I think, on the Internet. Like, the Internet is very efficient at sorting information. If you have something that's, like, pretty good, then it's hard to get traction without an audience. Like, pretty good stuff can get a lot of traction if the distributor has an existing audience. Even if you have no audience, I think truly excellent stuff will get traction.

Jon Chee - 00:03:39: For sure. Absolutely. I feel the same way. I mean, like, I just go super deep on Substacks, and I'm like, oh, shit. Like, this is really good. And, you know, people just end up sharing it. Like, this kind of, like—this ends up what hap—like, someone throws it into a group chat.

Alex Telford - 00:03:55: Yeah. I mean, algorithms are very efficient. Like, the post I wrote, the most viral one I wrote, I didn't have much of an audience then—the one about, like, the history of tech development. But that very quickly got sorted up to some fairly high-profile people. And I just think the Internet is just so good at it—it's just so good at, like, the sorting mechanism. Absolutely.

Jon Chee - 00:04:08: Okay. So, like, you now, from your writing, got the ingredients of, like, this early company formation. Just talk about you, your co-founders, the company that you are now about to stand up. What's your guys' mission and vision?

Alex Telford - 00:04:28: Well, I mean, the mission, it's kind of broad. So the idea is we want to try and build autonomous or enable autonomous drug development companies or in commercialization. So, like, building the infrastructure to support this transition from stuff being done this one-off manual way to autonomous, like, information factory-like processes where you can encode your knowledge and your workflows and then have the organization almost run itself. That's the kind of ultimate vision is that you could have an academic with an idea for a drug, and you can hand it off to a system. And the system could just do all the knowledge work required, like the administrative work to go and take this drug through the development of regulatory process, and then get on to market and be—we're very long way away from that. But that's sort of the future we want to try and work towards is a radically more efficient industry because so much of this manual work has been automated. And I think the way I think about the industry today, it's like the space industry pre-SpaceX, where every launch is its own project and it's a bunch of throwaway work and you don't learn from that work in a compounding way. And I think now you can build a system that can—when you're encoding knowledge and encoding workflows, you can really like engineer organizations in a way that was only available to maybe software organizations before LLMs. So that's like the ultimate vision.

And then, practically speaking, we are working with biotech and pharma companies to help them pick different tasks they may do quite manually today, often in areas of, like, intelligence or insights adjacent. So competitive intelligence, medical affairs, some document, like, document writing—like, these areas that are very required to analyze large amounts of information, then helping them build systems to ingest that information, structure it, and then process it to help them make decisions more effectively or to produce some document or deliverable that it would then take to, you know, external audience, internal audience, or regulators. So, yeah, like happy to go in in more detail on any particular use case, but generally, like, we think of ourselves as like having this pretty broad ambition and wanting to serve a lot of different use cases, but, like, fundamentally, we're a data architecture infrastructure company.

Jon Chee - 00:06:32: Cool. And let's just, like, talk about the use cases for various partners or customers of yours. Maybe let's, like, dissect it into large pharma and then maybe someone that's, like, a biotech, like, a growing biotech. Like, for you, this broad application, like, how does this application differ?

Alex Telford - 00:06:49: Yeah. It's actually quite different between the two classes of customers. So the way we're trying to architect our platform is to be this, like, fairly broadly applicable modularized system. So you can imagine that there's some datasets that we produce. So we'll go out and we will structure a bunch of information to—you know, it could be, like, trial registry data, publications, regulatory filings, patents, and make our own datasets that are easier to use and analyze and incorporate into alongside other datasets. So stack standardization is part of what we do, and our customers get access to those datasets. So you know, if you wanna do a query, like, what are all the patents against some given target, then that's something that we offer. And then we also have another set of functions, which are more like workflows or tools. So things like—you know, there's like an Excel-type functionality. There's our document writer functionality. There's a bunch of custom tools we build so our agents can—you can stitch together the data and the tools to do some workflow. So if I want to write, like, a regulatory document or something, then, you know, we might upload some clinical trial data. We might upload some prior filings. There might be some information on the web, and then we get the agent to go and find that information and stitch it together to write a document. And we would work with them to, like, test and validate the system and then roll it out. So that's, like, a very high level the system architecture.

And with pharmas, maybe I'll start with simpler. Typically, what we do is we'll go in and ask them what's the specific discrete task that you'd like to automate, either because you're spending a lot of money on consultants to do it or it's taking up a bunch of your time or it's just, like, something you would like to do but haven't been able to have the capacity to do. And then we'll go and understand, okay, where does that data come from, what's the process you go through to do that, and how can we encode it on our system. So, like, one example we did recently was for a big pharma firm working with a consulting firm. Every few months, they would go and, like, do this refresh of about 200 different specific oncology or cancer indications. And they would rank them by how many patients are there, how many competitors are there, what's the latest competitor data, what's the standard of care in that indication, what's the bar to beat. Like, what is the current standard of care bar to beat? Survival, OS, like, ORR, what's the emerging bar to beat, and all these other things. And they would use that to help them prioritize development strategies for early assets. And they pay a consulting firm every few months to go in and do and, like, refresh it, keep it up to date, and do it manually—like, manually go and check the registries to see if there's new data.

Jon Chee - 00:09:05: Meh. Like, count the number—

Alex Telford - 00:09:06: —of programs in development, then manually go and, like, look at the latest trial data and the conference data and see if there's any new data. So they're spending a lot of money on it, and it's taking a lot of time. We could build a system that could basically replicate that consultant work. That's like one example use case. Yeah. Yeah.

Then with the smaller companies, it's often a lot more expansive. It's like they actually want to reimagine their organization. Like, they see an opportunity to rebuild their—or not rebuild, but even, like, they're building their organization. So they want to figure out all sorts of use cases. Like, they wanna build a company knowledge base, and they want to—one week, they're gonna be working on regulatory documents. Another week, they might be working on how do we automate our competitive learning. Another week, it's like, how do we do our patent scraping? So those partnerships are, like, often a lot more expansive because there's one person in an organization who's, like, smaller biotechs who's, like, the AI person, typically. Yeah. And they're just, like, trying to figure out what are all the things we could do to make our company more efficient. So instead of maybe hiring 10 people, we can hire one person, have a system do these, like, automatable tasks, and we can be much more efficient as a company.

Jon Chee - 00:10:03: Very cool. And I've heard when you, like, layer in these agents, like, you gotta be, like, pretty hands-on on the thing. Like, are you getting, like, embedded in the company for a little bit and just, like, trying to, like, get that agent, like, basically up to speed? Or is there something where you can do, like, plug-and-play it?

Alex Telford - 00:10:20: I mean, it's not quite plug-and-play. We have to actually work with the customers pretty closely. I mean, we're not necessarily going on-site with every single customer. Some stuff is more plug-and-play, especially if we've done it before, then, you know, we have good confidence. Companies have the same use cases typically or, like, similar use cases. So if we've done it before, we know that we can roll it out to other customers with high confidence and maybe it's, like, minimal customization needed. But with some, you're doing a new use case, and then you usually have to work quite closely. You get a lot of feedback from them. And the success is very much dependent on how much the customer is willing to engage with us on actually automating their workflow. You know, if they expect it to just work out of the box, it's not gonna work out of the box because everyone is particular in their own way. They want something formatted in a certain way, or they want, like, the data sent to their Teams channel or whatever versus their Slack channel, and their own data is weird, or it's like they use a weird system or something or a citation manager. So you always have to do a bit of, like, embedding and understanding their particular workflow because everyone's workflow is a little bit different. It's, like, different in the details even if the shape is very similar.

Jon Chee - 00:11:17: And when it comes to, like, your guys' like product development, is it such that you can just, like, shape your product suite to exactly an end user when it comes to a product road map? Like, do you have to be, like, really deliberate in what you're rolling out? Or is it so malleable that you're just like, no. Tell us what you needed to get done, and we'll get it done.

Alex Telford - 00:11:38: Well, I think the trap with AI is just because you can kind of do anything, and you can use it to write code very fast, the trap is to say, "Okay, we'll just do everything—we'll do an everything app for our customers." I don't think that's correct.

Jon Chee - 00:11:48: So you have to be more opinionated?

Alex Telford - 00:11:50: I think you have to be opinionated on what parts of the stack, like, where you can deliver value. And I think the way we can deliver value is in—like, there's that consultative approach of, like, how do you actually operationalize this workflow as an automated system, which may be—may use parts of our system and may use parts of a different system. And then there's the, like, okay, where can you actually deliver value versus these other people in the ecosystem? Like, where do you add a differentiated value on top of the foundation models and the generic tools like Claude Co-worker or Codex or whatever? So my view is that I think the way we add value is in helping customers create and maintain these high-quality validated datasets with high signal-to-noise. So like sourcing data, cleaning data—that's valuable. Token efficiency, I think, is valuable. So as we're spending more and more compute, can we build a system where if you ask a query in our system, you're gonna spend less compute than if you ask it to a generic system, which may have to spend some time, like, recreating the wheel before it can go off and solve your question. And then things like generic things, like making a dashboard or making slides. My view is that actually, probably, we wanna expose, like, an API, and they could just use Claude Co-worker or whatever to go make some slides using the data in our system. Like, we don't need to do everything for everyone.

Jon Chee - 00:13:00: Yep. Interesting. I tend to agree too. Like, it's kind of like you have this, like, blank canvas kind of, like, trap, but you're just like, "I'm literally gonna do everything." And then what you just get is, like, some, like, middling, just, like, useless—like exactly what you said. You just, like, throw it away.

Alex Telford - 00:13:14: Yeah. You just get, like, something that's kind of bad at everything.

Jon Chee - 00:13:17: Yeah. It's like, god, this is awful. And have you found on this, like, go-to-market motion that it sounds like the smaller folks are, like, really embracing it. But, like, are the larger folks embracing this transformation, or are you having to do a lot of, like, education around this? Or they're like, "We're game. This transformation is necessary"?

Alex Telford - 00:13:36: No. I mean, they're—they're embracing it. I mean, especially, like, the executive level, they're really embracing AI and, like, the individual people within the teams who really want to reimagine their workflows. I wouldn't say, like, the culture of the people at the companies—like, I wouldn't characterize it as, like, resistant to big pharma and early adopters of biotechs, generally. I would characterize it more like your scope of responsibilities at a pharma is generally much more constrained, and your scope of influence is much more constrained. So even if you're working with someone who's very excited about the potential of the technology and wants to build something, they might be responsible for something like—like, "My remit is like medical affairs in Asia Pacific or something." So I can roll out something maybe with that scope of influence, but I have to also work with enterprise IT procurement, and it has to play with our systems. And so they're very constrained on what they can do versus someone at a biotech. Often, what we see is that there's, like—okay, the budget's not as large at a small biotech as in big pharma, but their scope is very large because you might just have, like—the CEO might just say, like, "Arun, we gotta do something in AI. Jon, you like AI. Go figure out things to do," and they can just, like, go through the organization and just, like, do stuff, and no one really stops them. So I think it's just that you have this big company, you have this, like, veto-cracy, where even if, like, the organization as a whole wants to adopt it, it's kind of hard to solve the coordination problem. But in biotech, you can just sort of do stuff.

Jon Chee - 00:14:54: Yeah. Yeah. Yeah. Interesting. And as you're continuing to grow your business and, like, you know, we talk about a little bit of product road map, like, let's look one year, two years out. I know this shit is moving really freaking fast. What's in store for you guys? Like, Convoke's—like they've seen how fast this thing is moving, and you guys are probably having to, like, move to stay ahead of the game. Like, what's in store for you guys?

Alex Telford - 00:15:17: Well, I mean, we've gotten pretty good at building a system to build validated datasets that can then feed many of these downstream use cases, whether it's, like, competitive intelligence monitoring, alerting, whether it's document generation, whether it's these arbitrary ones for, like, mapping different potential development options and ranking them. And then the thing we wanna get to next is, okay, how can we go from dataset creation and, like, being a store of your organization's knowledge to being more predictive? So can we actually operate like an analyst that will make recommendations or predictions on what you should be doing based on the knowledge your organization has? So right now, for instance, we do some work on, let's say it's like site selection or trial timeline estimation. So we've gotten pretty good at doing things like we create a dataset of all the potential sites that you could enroll a trial at, and we can integrate data from the public registries, hospital websites, plus your internal information about prior experience with those sites, and we can structure it. So we create these, like, arbitrarily defined datasets that people are using to make decisions about, maybe where do I enroll. But you can imagine that you have that data encoded and structured. You can use it to train a model that's gonna tell you, "I think you should enroll at these three sites because of x, y, whatever, z learned features," and then getting us to do it autonomously. So I think you're moving from this, like, reactive intelligence to more proactive intelligence where the system is continually ingesting data, monitoring it for you, and telling you what you should do in the world based on what you know. So I think that's the next frontier of what I'm excited about.

Jon Chee - 00:16:47: Sick. Like we're moving out of StarCraft. It's like—

Alex Telford - 00:16:50: Yeah, I know. We're moving out of StarCraft of, you know, you need to be the person in the system making the datasets. And it's like actually quite laborious, I think right now, to really create these datasets, to a system where you've created this like system for gathering and aggregating, cleaning information. And now you can start doing the really cool stuff on top of it. You know, this like autonomous system running parts of your company for you.

Jon Chee - 00:17:09: That's really rad. I think—all of this is just so crazy to me of, like, how quickly, like, we're moving. That's the part that is, like—is, like, mind-bending for me, which is just like, I thought I could, like, keep up, like, with kind of the progress of—like, I'm like, of being being from the Bay Area, like, I'm always, like, surrounded by tech. But I'm like, holy shit. I'm like, this is, like, next level. And I love your description of, like, the electrification of, like, a factory because I think that's, like, really apt. And I think for anyone thinking about what their—you know, large pharma, biotech thinking about this transformation—I think that's a very good metaphor or just example, an Alex example, of, like, how you should be thinking about this and exactly what you said. You either get with it or you don't. And you kind of can see the writing on the wall of kind of, like, what it means. But, Alex, this has been super fun. I'm learning a ton and, like, thanks for entertaining my naive and probably, like, novice questions about AI and agents. I, myself, I am on that journey of, like, learning about this and what it means for our business. So it's, like, really cool to hear how you're thinking about it as you're on the forefront. You know, in traditional closing fashion, I have two questions. So first question: would you like to give any shout outs to anyone who's supported you along the way?

Alex Telford - 00:18:30: Yeah. I mean, I think, you know, anyone—like, my co-founders, investors, early employees—I think anyone who believes in the mission and the vision. Like, I have a lot of gratitude for people who wanna get on the ship with us, and our customers too. You know, I think it takes a lot of bravery to try and bring in an unproven system and to try and change how you work and stake, you know, some part of your career on rearchitecting a process. And I think like those early partners, whether they're working with me at the company or working closely with them at their firms to help them accelerate drug development, then that's what I'm pretty thankful about. Cause it's kind of surreal in some ways to go from, like—I think that's the nice thing about being an entrepreneur is like, you have this like weird experience of going from something on paper, or not even on paper, it looks like an idea in your head, to like a real physical thing in the world. And I guess I'm like always grateful that, like, you step back and you think like, "Oh, wow, this is like a thing that I have in some way imagined into existence." And you don't get that many other places where you can do that.

Jon Chee - 00:19:34: Love that. Love that. And last question: if you can give any advice to your 21-year-old self, what would it be?

Alex Telford - 00:19:40: I think coming back to some of our earlier discussion points, I think a lot of the things I did when I was younger were just like, I was sort of taking the opportunities that were available to me, and I hadn't quite realized how malleable the world is. And you can just, you know, if the opportunities that are available to you do not seem that good, you can often create opportunities, especially if you live in the Western world. I mean, this will be like a privileged position. But I think if you do live in the Western world, you do have a lot of opportunity available to you to just instantiate new opportunities. So, yeah, don't, like, wait on what you have in front of you.

Jon Chee - 00:20:13: I couldn't agree more. Right now is a really fun time to be building. I've been doing this for a really long time, and I feel like it's not my second wind—I have many winds—but, like, I got this, like, crazy amount of wind in my sails from just, like, all the technology that's available to you now. It's crazy. It's, like, absolutely crazy. And if you have that entrepreneurial itch, go get it. Like, really? Like, if you, like, if you see that opportunity, it's really fun.

Alex Telford - 00:20:41: Yeah. It is fun.

Jon Chee - 00:20:41: It's super fun. And we've been talking about games, but, like, even the use of these tools scratches, like, the gaming itch. Like—

Alex Telford - 00:20:48: Oh, yeah. Yeah. Yeah. That's why. I think it's probably dangerous because of that.

Jon Chee - 00:20:51: Yeah. Yeah. Exactly. Exactly. Maybe a little bit too much. I've gamified it a little bit too much, but just, like, it just makes company building just, like, that much more fun. Obviously, keep it in check. Like, don't let it autopilot, like, all the time. Use your brain. But absolutely. Well, Alex, thank you so much for your time. This has been super fun. I'm glad that you're in the Bay Area. Hopefully, we can grab coffee in person. Maybe the robots won't take over everything. But, yeah, thanks again.

Alex Telford - 00:21:15: Yep. Thanks, Jon.

Outro - 00:21:18: Thanks for listening to our four-part series featuring Alex Telford. From expat kid reading science magazines in Switzerland to biochemistry at Bristol and a synthetic biology master's at UCL that taught him more about the gap between promise and practice than the field itself delivered, through nearly seven years of biopharma strategy consulting where he learned that clients need opinions, not just analysis, and a cold turkey exit from a stable career, a period of writing and building in public, and the founding of Convoke on the conviction that drug development's biggest bottleneck has never been the science, Alex's story shows what happens when someone spends years as a student of an industry's deepest problems and then refuses to wait for someone else to fix them. If you enjoy the show, please subscribe, leave a review, or share it with a friend.

Join us for our next series featuring Matti Gill, CEO of ION Labs, a first-of-its-kind AI venture studio built on the Israeli innovation ecosystem and backed by global pharma and technology leaders with a mission to build and grow groundbreaking AI companies in biopharma, bringing together brilliant minds, pharma expertise, and cutting-edge technology to shape the future of drug discovery and development. Before founding ION Labs, Matti built a career at the intersection of healthcare strategy, innovation, and policy, developing a reputation as a coalition builder and leader in AI and biotech health strategy and life sciences innovation across global markets.

At ION Labs, Matti leads a unique model of company creation that combines pre-seed funding, access to proprietary pharma data, and direct validation against industry standards, removing the risk that kills most early-stage AI biotech companies before they ever get traction. With a portfolio of AI ventures built and seeded from the ground up, Matti's journey from healthcare strategist and innovation leader to venture studio founder shows what it looks like when you stop waiting for the right companies to appear and decide to build them yourself, making this a conversation you won't want to miss.

The Biotech Startups Podcast is produced by Excedr. Don't want to miss an episode? Search for The Biotech Startups Podcast wherever you get your podcasts, and click subscribe. Excedr provides research labs with equipment leases on founder-friendly terms to support paths to exceptional outcomes. To learn more, visit our website, www.excedr.com. On behalf of the team here at Excedr, thanks for listening.

The Biotech Startups Podcast provides general insights into the life science sector through the experiences of its guests. The use of information on this podcast or materials linked from the podcast is at the user's own risk. The views expressed by the participants are their own and are not the views of Excedr or sponsors. No reference to any product, service, or company in the podcast is an endorsement by Excedr or its guests.