Designing an agentic, future‑ready tech roadmap for emerging pharma (Sponsored)

Introduction to Zayden

0:00In global life sciences, insight isn't the problem. Execution is. Disconnected dashboards, silo teams, and fragmented workflows make it harder to act consistently, compliantly, and at scale. Zayden by ZS closes that gap. Zayden is a life sciences intelligence platform that embeds intelligence into the flow of work based on domain expertise and agentic AI across commercial, medical, patient, and content work.

0:30Trusted by more than 150 life sciences organizations across 100 countries, Zayden is built for global scale, regulatory compliance, and trusted AI. So teams act together across brands, markets, and channels. Zayden, move beyond reporting to intelligent action. Learn more at Zayden.ai.

0:53You're listening to a sponsored episode of The Top Line.

Guest Introduction

1:00Welcome everyone. You're listening to The Top Line brought to you by Fierce Biotech. I'm your host, Kelly Hogan. Today, I'm joined by J.R. Raylin, the Executive Director of Analytics and Insight at Synox Therapeutics, and Krishna GS, Business Technology Solutions Manager at ZS.

1:40It's great to meet you both. Would you like to share for the audience a bit more of your background? Sure. Thanks again, Kelly, for the invitation. My name is J.R. I live just outside of Boston, and I've been in the biopharma industry for almost 20 years. Half of my time as a consultant, half in-house leading technology operations and launch initiatives. Thank you, J.R. Thanks for the invitation, Kelly. I'm Krishna GS, a manager from the ZS Chicago office, and I lead our digital data and AI-driven transformation programs for licenses-based clients.

2:17I've been in the industry for close to 11-plus years now and worked across large-size, mid-size, and emerging pharma companies and biotechs. And I primarily work with senior business and technology leaders like J.R. to essentially design and develop scalable cloud and AI solution.

Emerging Pharma Challenges

2:34Emerging pharma teams face increasing challenges to deliver rapid growth while understanding rising expectations for AI and overcoming resource constraints. Today, we'll explore how emerging pharma organizations can better design a future-ready technology roadmap, one that delivers value today while scaling intelligently for tomorrow. So, J.R., what common mistakes do emerging pharma organizations make when designing their technology roadmap?

3:05Great question. I've played around with the precursors of AI and machine learning for 15-plus years now, and I typically see the same three common pitfalls repeat over and over. The first is chasing trends, chasing AI trends without data readiness. I've seen teams jump into AI without clear understanding of their data ecosystem and without the right foundational elements in place.

3:36So, although AI can be a powerful ally, it really depends on human judgment to set the right guardrails in the context. Similar analogy that we've all heard, we build a house with a weak or flawed foundation leads to instability. Same if you implement AI without strong data fundamentals, causes promising initiatives to fail. Second pitfall is something that I see often throughout the biopharmaceutical industry and is technological complicity in a fragmented siloed ecosystem.

4:11Often in our industry, the pace of business outstrips the speed of IT and speed of enterprise processes. The result often proliferates shadow IT and point systems that make it difficult to scale successes and or maintain proper guidance. Finally, building on both of those, underinvestment in integration and interoperability. When we have these siloed and point solutions that proliferate that shadow IT across the ecosystem, it becomes increasingly important for them to communicate in a thoughtful, structured, intentional way.

4:51Without that connectivity, you end up with these, what I call, islands of automation that compound problems rather than solving it. And I'll almost add like a fourth pitfall that I specifically see in a lot of the first launch companies, which is what I call as the pre-commercial paralysis. Many emerging teams wait almost until close to the approval stage to design their commercial data foundation. But usually the highest performing launchers that we have partnered with through ZS is basically the ones that started designing their data architecture or their MDM strategy and the agentic workflow blueprints pretty close to phase three.

5:26Because by the time they receive approval, they'll be in a much more better position to scale as opposed to building stage. And one of our ZS research actually confirms that most underleveraged strategic window is the phase two to phase three transition, where you get clinical signal clarity that is improving, the competitive position is sharpening. And for a company with one product and one shot at a peak revenue, which is basically that six to 12 month window of Head Start will make a difference between a strong launch trajectory versus a permanently compressed one.

Technology Roadmap Design

5:57Emerging pharma teams face very different constraints than large pharma when making technology decisions, especially as AI expectations rise while resources remain limited. Krishna, from your experience, how should these organizations approach early platform decisions and what criteria matter most when shaping the technology roadmap? Great question. When I work with first launch or pre-commercial companies on platform technology or roadmap decisions, we use basically something like a three horizons agentic maturity framework, right?

6:34The horizon one is what I call as the launch readiness. Basically, think of pre-commercial to commercial to month six, right? That is a window. That's a window where you basically build a very robust, unified commercial data foundation, BTO data warehouse, MDM, audio governed data layer, right? This is what I call as the nervous system of every agency workflow that kind of follows after that. Now, at this stage, agents are primarily your consumption agents, surfacing insights and aspect actions that are feeding directly into CRM and field workflows, right?

7:06Then comes your horizon two, which is the in-market acceleration. And this is the window between month six to 18, where actually the layer execution agents that automate targeting, field alignment, incentive compensation, and omni-channel orchestration comes in. This is what I call as the sense, decide, act, and learn loop that becomes operational here, right? And this is where the agentic AI continuously surfaces, head CP-level barriers, and links signals directly to the engagement decisions.

7:37Another ZS data shows that this approach reduces the decision cycle time by close to 40-50% and delivers at least 5% to 10% lift in the brand sales uptake and the overall brand performance. And then the last horizon is the portfolio intelligence, maturing from a one product to a portfolio-level thinking, where the single asset launch becomes a bit more of a portfolio-learning intelligence engine, right? Where the configuration and the monitoring agents capture signals, update the assumptions, and inform those decisions for the next indication or the next level of geography expansion and other stuff.

8:12So, essentially, the first launch pays dividends into the second, where the data models and the workflows are reusable and doesn't have to be rebuilt. So, in a way, I would say design these horizons in reverse. Start with the horizon three in your mind so that the horizon one infrastructure is already portfolio-ready. We have our own, which is the licensed purpose-built platform called Zayden, which are pretty much architected for this exact modular extensibility. And if I tack on to that, certainly cash-burning biotechs are under different pressures than large pharma companies.

8:51What I've found in my experience is when you're at a large pharma company, you can typically get best of breed and stitch things together. You have a lot more resources to choose and optimize a solution. When you're a smaller company, that biotech space, you typically don't have the time, the expertise, or the energy slash resources to do that. So, getting singular platforms that can do many things well is a lot more efficient and effective.

9:23AI brings powerful capabilities, but also new risks.

AI and Data Readiness

9:27J.R., how should emerging pharma teams balance AI and advanced analytics ambition with data readiness, compliance, and organizational maturity when designing a trusted, AI-infused ecosystem? Honestly, I see no difference between an AI initiative and rolling out a new system or a new change management process.

9:53Always, when you're going through an effort like that, you want to have your end result as the outcome. So, in this situation, AI is not the starting point. It's the outcome you're trying to target. To get there, you typically begin by designing governance, compliance, things like explainability and audibility into the system. And as you build out, that continues to build upon itself. I'd recommend that teams start small. You begin with those narrow, high-confidence use cases that allow you to be successful with AI, allow you to learn quickly or fail fast, and then optimize as you go.

10:33Additionally, I would recommend that you match AI sophistication to organizational maturity, not just the market type. And then, finally, I'd recommend that teams focus on trusted, usable intelligence that fits real workflows. If I bring that all together, for an example, just from my own experience, When we deal with complex processes in the biopharma space, like master data management, MDM, something like that has wide-ranging reach and includes everything from identifying, managing, ingesting source data, to business rules and survivorship, to data stewardship, and so on and so forth.

11:19So it touches a lot of different teams and a lot of different complexities across the organization. So my previous company, we leveraged a platform, in this case, Zinan, working with Krishna, where we adopted an agentic approach to automate our data stewardship. Obviously, we still had human oversight. However, that AI agentic approach dramatically improved our throughput, our accuracy, and our reliability. It allowed us to catch up to all of our data change requests and get ahead of it so that we could then shift our time and detection to more thoughtful and strategic work.

11:57I think JR hit the nail on the head there, and it's easy to get caught in a generative AI and agentic AI worldwide these days. I think what is important to understand is we all work in a very regulated admiral environment, where agentic AI is not just about autonomous systems running unchecked, but actually about reliable decision to execution with full traceability and human accountability, right? So that's where, for a lot of the first launch companies specifically, I would say responsible AI is almost, is not just a compliance checkbox, but it's actually a trust-building strategy with your regulators, your payers, your HCPs of the world, right?

12:36So we basically apply five principles here, like first one being human in the loop by design. So the agents handle execution-heavy activities like D-Ingestion, KPI generation, the scenario exploration, the targeting, but humans want the judgment, the context, and the key decision-making abilities. So we can codify explicitly what the agent can execute autonomously versus what requires a proper human decision approval process or a workflow, and when the system must actually escalate.

13:07That is a very important orchestration model as part of the first principle. Then the second principle would be a governed guardrail before deploying any agent. In an agentic workflow, compliance guardrail, the auditability and the traceability should be baked in from day one and should not be retrofitted, especially for emerging pharma, where patient data privacy, the HIPAA alignment, promotional content review when it comes to medical and the content that gets used by some of the reps, and the syndicated data licensing compliance. There are a lot of standards like that which require proper auditing and proper traceability.

13:42The third one is explainability as a feature. AI that no one trusts will never get adopted. So first launch companies with lean teams, the team must be able to explain to the CXOs or the VPs or the directors or the executive directors of the world, and the compliance team, and the field rep around why a particular agent made a recommendation that it did. It shouldn't be a black box. The moment it becomes one, then it'll be a trust killer in the environment. Explainability should be a must here. The fourth one is what I call as the model monitoring and drift detection, especially in a first launch environment where the market dynamics shift rapidly, the formulary changes, the competitive entries happen, the air pushbacks happen.

14:22AI models can easily go stale. And that's where the monitoring agents that can surface these anomalies, detect those drifts, and flag issues before they escalate are super essential. The last one is the incremental trust building. Start with those narrow, high-confidence use cases. Could be your HCP targeting recommendation before expanding into a very higher-stake decisions. Let the team build the confidence in AI outputs through some easy wins before automating some mission-critical workflows. You've given us a great use case and a lot of factors to consider, but technology only creates impact if teams adopt it.

14:57Krishna, what does effective implementation and onboarding look like, especially in the first 60 to 90 days? The 60 to 90-day window is not just a technical onboarding milestone or a chip box. It is the moment where the organization decides whether the AI becomes a trusted partner or an expensive shellware subscription. And we've seen both. I think the day 1 to 30 is basically the place where you wire the foundation, not the features. It's very hard, but we have to resist the temptation to turn on every capability at once, especially with lean teams in a first-launch environment.

15:32The cognitive overload can easily kill the adoption faster than any technical issue can. Pick one or two high-volume, high-visibility workflows. Could be the pre-call planning, could be the HCP targeting, and get those working flawlessly before expanding. So the goal in the first 30 days is more to deploy it and make sure it's at least trusted by those key people. Then the day 30 to 60 is what I'll say as connect the data and prove the signal. This is where the data foundation pays off or could expose the gaps. So by day 60, your field team could actually be able to look at an AI-generated targeting recommendation and actually say,

16:05yes, this actually matches what I know about this territory or about a doctor. Not where did this even come from? So that trust moment is very key in this world. And if you're still reconciling, let's say, MDM issue or waiting on data feeds, then you have a data readiness problem, not an AI problem. So it's important to stand up your core data products and governance by that day 60. Then comes your day 60 to 90, the scale that works. So by day 90, you should have clear signals on two things. Which workflow is the team actually using without being told to and the ones that they're just simply exporting to Excel instead.

16:40Scale the former one very aggressively. Diagnose the latter because it's almost always a trust or explainability gap and not a feature gap. This is also the window to introduce your first execution agents that could be the likes of automating like a routine production work. So your analyst or your data scientist can shift from data wrangling to more of a scenario planning and strategic problem solving. Net-net, the first launch companies often do not have a dedicated analytics ops team. JR knows this firsthand. JR is almost a one-man army when it comes to his themes.

17:11So the same person configuring the platform could also be the person reporting or presenting to the VP of commercial. So that means the platform must be intuitive enough to self-serve pretty much from day one. And the implementation partner must pretty much act as an embedded team member and not like a service ticket queue. That's a place where when it comes to even within ZS, we invest a lot of time in some of these connected workflows. And one of our proprietary platforms, Zayden, we actually consciously make sure they are ready to deploy packages for emerging pharma companies that are designed exactly for this kind of a lean team reality.

17:46And that's where once you're done with this 30 to 90-day window, pretty much from day 90, the question should be like, are insights from this platform appearing in the leadership meetings without even being prompted? That is the real adoption signal. Not the login rates, not the dashboard views. Are these AI-generated insights truly changing decisions? If the answer is yes, then you know that the roadmap is working. From my perspective, smaller companies, you tend to have wide-reaching, wide-ranging roles. You have to be a master of everything. So to Krishna's point, start small, start where you can win, but start in a way that you can build upon that foundation and deliver value practically and pragmatically.

18:26And from your experience, what does success look like for a future-ready technology roadmap?

Future-Ready Roadmap Best Practices

18:33What best practices would you include in a practical playbook? Like we talked about before, AI today is touted as being the sell for everything, but we can't really believe the hype, right? It's not a one-size-fits-all solution. Instead of chasing those buzzwords, we need to take a step back and think of how to build more flexible, scalable, and adaptable solutions. What I recommend is, first of all, aiming to build a modular, interoperable architecture.

19:04We want an ecosystem that allows us to drop in AI services where they make sense and then replace them, right? Replace those agents or plug other ones in as they evolve over time. Today, folks are using Claude here. They're using ChatGPT there. Tomorrow, a new provider might come in. So you want that interoperability so you can get best of breed where it makes sense. Secondly, I still recommend that clear ownership and governance. So you want those clear guidelines, those racy matrices to establish those principles and that structure to help provide transparency across the organization.

19:41Third, I would say you want to adopt outcome-driven technology. So instead of just plugging in AI anywhere, you want to focus that and all your technology on specific business decisions and use cases that allow you to really solve key problems, right? AI should be a meaningful accelerant to getting insight. It shouldn't be a replacement for the way you do your work. Fourth, I'd recommend building out unified workflows that include strong data foundations.

20:16When you tie together data, business rules, and analytics, you're much more ready and able to meet the ever-changing needs of your commercial, medical, and patient teams and reduce that fragmentation. So when we bring this all together, today, we all have workflows, right? Some people will call it a workflow. Some people will call it a process of their proprietary approach. Even this podcast has a workflow, a set number of steps to it.

20:47So if we look at an example where, let's say we're building up a data warehouse, we can start thinking of different areas and ways to bring in AI to help augment that existing process rather than reinventing it. So if we're kicking off a project, perhaps we have a few meetings on business requirements. Normally, we would then have to break apart, iterate on a business requirements document, which could take a couple of weeks before we get to something final. Let's say that during those meetings, you drop in a document agent, you can now near instantaneously have your business requirements document.

21:24Much quicker, much more effective, doesn't change your workflow. Then, once you move on and you approve those requirements, you could either have a DBA and a team of infrastructure engineers work to set up those pipelines, or alternatively, you could bring your configuration agent in, pass it some credentials, like through an S3 bucket, some data sources, and it can onboard your data again almost instantaneously. It doesn't remove the human intervention that you need to ensure quality, but it does speed things up.

21:59And then, finally, you could have a data quality agent that massages that data, makes it work much more tuned, so you can push it through from start to finish to a report. That example, much like with anything in a roadmap, follows our well-established process that we exercise today. The only difference is that we're leveraging AI to reduce turnaround time, reduce potential for human error, and honestly, free up resources to do much more value-add activities.

22:30We still need those highly trained and valuable team members to review, to do some configuration, but we don't need them to do some of that quote-unquote grunt work that is more repeatable over time. So, in the end, I think a future-ready roadmap provides an architecture that grows with you. It supports you for what you need today to build from a small biotech to launch, to then build for maybe a second indication, and then can iterate all the way through to scaling through multiple geographies over time, all while doing nothing different than you've done today, just doing it more effectively, efficiently, and thoughtfully with AI.

23:14And I couldn't agree more. I think I would say the first dimension is speed-to-insight driven decisions, right? Questions like, are commercial medical and free teams making decisions in hours and not days and weeks? I think for a first launch, this is super critical, like early HCP adoption signal, those formulary wins that you get in months one to six, confound over the entire product lifecycle. Then the dimension two could be whether the agentic flywheel is turning or not, the sense, decide, act, learn loop that I mentioned before, whether that loop is actually operational and compounding, whether the signals are flowing in from the field, is the models are updating, are the targeting decisions that are getting refreshed and the outcomes that are feeding back into the system.

23:53If your AI is producing the same recommendation that it did 90 days ago, despite the new market signals, then the flywheel is not working. So each cycle has to make the system smarter. Only then, the flywheel essentially will convert a first launch into a portfolio intelligence engine. Then the last one is the connected workflows, and this is connected workflows across commercial, medical, patient services. These roadmaps could succeed only when the data is not just clean, but it's when it's connected, because the HCP engagement signals will inform a patient support program design. A formulary when feed could essentially into the targeting models, a medical affairs insights flow into the promotional content strategy.

24:29So if your field team is still exporting AI recommendations back to an Excel to do their own analysis, then agentic roadmap has broken down at the adoption layer, not at the technology layer, right? So that is a change management and the trust problem, and it must be addressed before scaling. So pretty much all the three dimensions that we just talked about needs to be tracked simultaneously for us to measure the success of this. Can you give us an example of what a connected agentic stack could look like specific to emerging pharma? Sure, Kili. I think that's a very good question, especially when it comes to emerging pharma.

25:02An agentic roadmap is not just a collection of AI tools. It's actually a connected system of decisions. So the value compound happens when workflows are wired together, not when individual agents are just smart in isolation. When it comes to a fully connected agentic stack, I think it could look something like this for a first launch company. The HCP master data agent will cleanse and govern your targeting universe, speeding into field targeting or digital channel or postation simultaneously. Then there is your field execution agent, which surfaces your pre-called planning intelligence and next best actions directly go into the rep's CRM workflow.

25:32No separate tools, no context switching. Then there is a market access agent, which monitors your formulary status in real time. And when a payer decision changes, it automatically updates, targeting priorities and field messaging guidance. And then there is your patient support agent that could identify patients who have been prescribed but not yet started on therapy and could also trigger an intervention without human manually reviewing a spreadsheet of any kind. Then there is a monitoring agent that could detect anomalies in launch performance. There could be a region underperforming versus forecast. Some could be even performing better than the forecast. And all of this can be surfaced to the leadership even before the monthly or quarterly reviews.

26:06So there has to be an orchestrative agent, which is actually monitoring all of these anomalies, the changes, the drifts, and kind of informing each other between the workflows. So every one of these workflows is individually valuable. But when they're all wired together on a single governed data foundation, sharing those contexts, feeding into each other's signals, updating in real time, that's when you have an agentic flywheel. This is what one of our proprietary license platform called Zayden. Architecture also enables not just smarter tools, but an orchestrated commercial intelligence system.

26:37I've enjoyed learning more about the trajectory of AI and pharma with you today. And you've given us a lot to think about when it comes to designing a future-ready technology roadmap. Any final thoughts to share before we wrap? I'd say the teams that succeed are the ones that sequence correctly. Instead of starting with AI, start with those foundations. Data first, then analytics, then AI. Focus on decisions, real decisions, real users, real impact. And I will add to that, right?

27:08For emerging pharma specifically, the agentic roadmap is your most important strategic asset before your second product. Every decision you make about data architecture, workflow connectivity, and the AI governance in your first launch, either is going to convert or compound into a portfolio value, or is going to create a technical debt. So you will spend your next funding round paying off for that. So build it right, build it modular, build it governed, and let the flywheel do the work. The companies that will lead the next decade of bioformal commercialization are the ones that are building this agentic, intelligent systems today.

27:41Not because the technology is impressive and cool, but because it converts this launch learning into something durable and gives you that competitive advantage. Now that is the real measure of a future-ready roadmap. Thank you, JR and Krishna, for your time and your insights. And thank you to our audience for joining us on this episode of The Top Line. Once again, I'm your host, Kelly Hogan, and that's The Bottom Line from The Top Line. I'm your host, Kelly Hogan, and I'm your host, Kelly Hogan, and I'm your host, Kelly Hogan, and I'm your host, Kelly Hogan, and I'm your host, Kelly Hogan, and I'm your host, Kelly Hogan, and I'm your host, Kelly Hogan, and I'm your host, Kelly Hogan, and I'm your host, Kelly Hogan, and I'm your host, Kelly Hogan, and I'm your host, Kelly Hogan, and I'm your host, Kelly Hogan, and I'm your host, Kelly Hogan, and I'm your host, Kelly Hogan, and I'm your host, Kelly Hogan, and I'm your host, Kelly Hogan, and I'm your host, Kelly Hogan, and I'm your host, Kelly Hogan, and I'm your host, Kelly Hogan, and I'm your host, Kelly Hogan, and I'm your host, Kelly Hogan, and I'm your host, Kelly Hogan, and I