Transforming pharma CRM with agentic AI and Salesforce

The pharmaceutical industry is at a turning point. It’s time to get the most out of your CRM supported with new ways of working and enabled by technology. In this webinar, a panel of experts from Salesforce, AstraZeneca, Pfizer and ZS explores how Salesforce’s Life Sciences Cloud—combined with the next generation of agentic AI—can transform customer relationship management across commercial operations.
 

You’ll walk away with:

• An understanding of how Salesforce’s Life Sciences Cloud and Agentforce are reshaping engagement models
• Strategic insights into successful Salesforce Life Sciences Cloud implementation and change management
• A glimpse of what’s next in pharma CRM, AI and the evolving customer journey

Key takeaways:

• Pharma CRM innovation with agentic AI is reshaping how customer engagement and effectiveness are defined
• Unified data is becoming the foundation for scaling CRM with confidence
• Aligning AI with compliance and productivity sets the stage for stronger customer outcomes

In a Fierce Pharma webinar, ZS Principal Srihari Sarangan spoke with Salesforce Life Sciences Cloud expert Tara Helm and other industry leaders, including Pfizer’s Dr. Suman Giri and AstraZeneca’s Richard Mendoza, on how agentic AI is transforming customer relationship management (CRM) across commercial operations. Here are the highlights from their discussion:

Srihari Sarangan (SS): Suman, how are you thinking about the agentic world? What capabilities are you trying to unlock? We all know customer engagement is top of mind for the industry right now, both across healthcare providers (HCPs) and patients. How do you see agentic AI transform that engagement?

Suman Giri (SG): The way I think about it, an agent-based ecosystem comes into play when two things are true. First, you have a set of actions that you need to take that have some stochastic element to them. By stochastic, I mean that it’s not fully deterministic, so you can’t put a business rule against it and there’s some sense of judgment required.

The second is typically that there are multiple systems, tools and resources that need to be consulted. These are the places where agentic workflows shine. If I think about the major areas where we are evaluating the use of agentic architectures, they primarily boil down to three categories: insight synthesis, productivity and effectiveness.

Within insight synthesis, there are implications in terms of quality of engagement and consistency of messaging. Classical use cases include precall planning for our reps—the right messaging, whether that be field or omnichannel. Productivity is where constructs like routing come into play. Agents are really good at things like that. Effectiveness includes coaching, compliance and making sure activities are within guidelines. Targeting, from an omnichannel perspective, is also something we are evaluating.

SS: Suman, that’s a fantastic way of describing the different layers of agentic AI. Because you’re right—there’s a lot of confusion in the industry around what’s inquiry, what’s insight, what’s productivity and what’s effectiveness. Richard, I’d love to know more from you as well.

Richard Mendoza (RM): We’re already living in a world of multiple agents. We’ve been rolling out different types of agents for a few years now. This is an evolution of that. Demands and expectations are growing exponentially. People are getting more used to ChatGPT, Perplexity and Gemini. Because of that, their expectations are higher.

So for now, I think pharma-owned assets have value. You would come to pharma organizations to get that information more accurately and have less chance of hallucination. But as we move forward, the agent ecosystem is growing the agents’ interoperability. And that’s going to be really important. I want to see more partners, more kinds of ecosystems developing and growing. You think of agents as APIs connecting with each other, so the next evolution is to unlock that capability.

SS: Agentic architectures—agent-to-agent and multiagent systems—are fascinating. Some researchers suggest agents should be performance-monitored, like individuals with ratings and reviews. In a multiagent system operating largely in an automated way, workflows trigger on their own. Suman, how is your organization thinking about CRM, and how does the AI ecosystem factor into those decisions?

SG: CRM is one of the major ways to activate AI workflows, but there are other martech and stacks where activation can drive value, especially on the commercial side. Normally, as an AI leader, I’m the one saying we should use AI. But now, because of the democratization of AI technology, everyone has a point of view on what AI should do for the workforce. Sometimes we confuse the job to be done with the tool we use.

CRM shouldn’t conflate its core value proposition with AI hype. With changes in the industry, we’ve publicly decided to go with Salesforce as our CRM of choice. This includes a unified platform across apps, data, agents and Life Sciences Cloud, which includes Salesforce Data Cloud and Salesforce Agentforce. Tableau also comes as part of that offering.

The AI ecosystem is important but not the be-all-end-all of CRM. All CRM offerings will have some AI capabilities. You can also build intelligence layers independent of the CRM. It’s early to have a point of view on what intelligence should be native versus built internally for strategic advantage.

SS: Eventually we’ll arrive at an ecosystem approach to CRM. Even within activation, workflows are complex. With the nascent maturity of agent architectures, there will be a learning process before CRM becomes an ecosystem across federations. Tara, I’d love to know how Salesforce is thinking about it.

Tara Helm (TH): We’re in such a unique position building out Life Sciences Cloud, in particular, because it’s a brand-new cloud coming to market. And we’ve been able to build it with that agentic-first mindset and on unified data. Agents are only as good as the data they’re able to access. So ensuring they have the CRM data, but with much broader access to data that empowers them, is critical.

People are expecting the consumer experience they get with ChatGPT in their everyday life and workplace. So we’re designing a platform and a solution where you’ll have those experiences in your CRM. You’ll be able to ask the agent different questions, just like you’d ask ChatGPT. In fact, we’re thinking about what to put directly in the product as opposed to what customers are going to build uniquely for their business. We’re trying to work through the agents we want to build internally or empower customers to build on their side.

Precall planning today often requires working across multiple systems, making it time-consuming and fragmented. Agentic AI can bring this together in one place. Our solution can answer “What does my day look like?” by showing who you’re seeing, providing recommendations and next best actions, and generating smart summaries. It draws on both CRM data and other data sources, all with guardrails in place. Precall, we’re putting agents at the forefront and partnering with customers and partners like ZS to decide what’s embedded versus customer-specific.

SS: Pharma has always been an industry that has significant compliance and regulatory requirements. Richard, I’m curious to know how you build governance, trust and transparency into this ecosystem?

RM: We still go through standard due diligence—cybersecurity, privacy, vendor assessments and architecture. But processes and policies need to catch up with the pace of AI. Internally, we apply AI governance on top of that, so there’s a specific construct that every AI-based project follows. That process protects our assets by avoiding bias, ensuring data principles are fair, securing consent and preventing licensing violations. It’s evolving, but in the near term it remains fundamental to our operating model.

SS: Governance is critical, especially with agents being developed everywhere. Is this managed through process, technology or people?

RM: Right now it’s people-based reviews. I’m excited about dashboards to monitor performance, error handling and metrics depending on the large language model chosen. Explainability is improving, but for now it’s human-driven.

SS: Tara, pharma has always struggled with AI as a black box. How does Agentforce help?

TH: Salesforce has been in AI for more than 10 years, investing in research and building the right governance. That’s the foundation layer we continue to build on. Within our agents, we’re constantly reviewing the guardrails we’ve set up and adding new innovations—releasing updated guardrails and recommendations as we go.

From performance ratings of agents to toxicity scoring, prompt defense and data usage, we’re focused on making sure the right safeguards are in place. That includes ensuring the right data is being used, applying masking where needed and maintaining audit trails.

How can you audit to ensure what’s coming up? How do we make sure users can access that information in a user-friendly way? We already have tools on the backend. For example, if you set a threshold where an agent falls below 99% on a particular score, it can automatically be turned off. We’re focused on building the tools and guardrails that give organizations confidence and comfort in how these systems operate.

SS: Transparency is key. Let’s talk about challenges around scaling AI in commercial teams. Suman, what challenges have you faced?

SG: The biggest challenges are competing priorities for the field, lack of user-centric design and technical issues, such as data consistency and integration. Our industry isn’t native to product development, so we often lack discipline in user design. Compliance adds another layer of complexity. To improve, you need executive sponsorship, strong product design and integration, and scale thinking from day one. Pharma’s scale means that solving for one product in one market can apply across many products and many markets.

SS: It’s part of the reason CRM adoption is so poor in pharma. How do you communicate these changes? How do you measure success—whether it’s productivity or effectiveness—and how do you share that success across the organization so it takes hold?

SG: Start by prioritizing based on value and feasibility. Define categories, rank them and select your strategic bets. Measure leading indicators like adoption and early feedback, then lagging indicators like time saved or revenue lift. Share the results across the organization, and ideally have finance ratify the impact.

SS: We’re seeing some examples where patient services, patient engagement, consumer marketing and HCP engagement come together. But we haven’t seen much progress toward true integration. Tara, how are you thinking about bringing these largely siloed worlds together?

TH: Within Life Sciences Cloud, we see a unique opportunity to bring data and capabilities together through a unified data layer. Field reimbursement managers, for example, work on both the patient side and the traditional field sales side. On one hand they use tools like key account management. On the other they handle patient services and patient data. Bringing those together would empower field reimbursement managers with the right data across systems and make it easier to move between those worlds.

The same opportunity exists in the commercial and clinical space. For clinical trial patients and HCPs, the question is how to share data within the right guardrails and governance processes. That kind of collaboration is what enables orchestration and better engagement across the board.

SS: Centralization likely means different things to different organizations, and every company is at a different stage of maturity. By and large, big pharma has been engaged in data management and centralization initiatives for a long time. In the agent world, how ready is pharma’s foundation to support it, especially as data will be generated from every workflow automation?

RM: I think data is critical in this concept. We’ve tried centralizing. I think there’s a lot of work that has gone into all of this. I think things going forward will be doubling down on tagging data, the taxonomy of the data and the discoverability of the data. And from there you get knowledge graphs and the evolution of more sophisticated forms of the data foundation. These elements are the core backend of many agents, so the time and effort invested in them is critical.

I’d call it an evolution, but with the rise of agents it’s become absolutely critical. Data sharing also needs to be closer to real time, though real time can mean different things for different organizations. Think about it like vegetables: you want them fresh. Three- or four-day-old vegetables lose their value, and data is no different. Old data creates problems. That’s why enabling these constructs is so important. Suman, what’s your take?

SG: I think data, like Richard said, is the end-all-be-all. But I think it bears a deeper look—especially in the context of agentic AI. If you look at pharma, a lot of our work is on syndicated data that we buy. I think we’ve not fully talked through the implications of having agents do things on that data. Yes, today we have third-party agreements in place—for example, when a company like ZS works with Pfizer using data supplied by IQVIA. But what happens when an agent tries to do the same thing?

As an industry, we haven’t figured these issues out well enough. It’s imperative to do so, otherwise red tape and bureaucracy risk becoming bottlenecks that hold back the potential of agents, especially with prescription and claims data. We’ve also tended to sweep unstructured data aside by saying there will be a database or a file that contains it all. But the real promise of agentic AI lies in structured data.

This is more data like images, voice notes and video. How are we tagging that? How are we labeling it with the right taxonomy? What are the right ontologies that need to exist? These are, I think, critical foundations that whoever figures out first is going to have significant competitive advantage when agentic AI becomes mainstream. The analogy I use is that right now we are in the mainframe phase of agentic AI. It’s centralized. It’s with a few organizations. Tomorrow, it’s going to be on laptops. Everybody’s going to have it.

SS: Tara, scalability and agility are essential for global CRM. How are you preparing?

TH: Since we just talked about data, I have to bring up Data Cloud. Its ability to leverage zero copy and integration tools to pull in and unify data is powerful. That agility makes it easier to connect into other systems and sources. Add to that the Salesforce ecosystem—low code and no code capabilities that minimize customization and avoid the scaling challenges that come with going global. Streamlining implementations up front gives organizations the agility to adapt quickly as needs change.

Salesforce also recognizes we don’t have it all. We’re an open ecosystem—across data, data sources, ISVs and partners with complementary solutions. Not everyone is “all Salesforce,” so integration and partnership are essential to creating agility. Our open partner network, including firms like ZS, helps customers navigate those choices. It comes down to three strengths: the platform with its low code and no code capabilities, unified data through Data Cloud and zero copy, and an open ecosystem of partners.

SS: We clearly have more work to do, but our discussion focuses our attention on the opportunities and obstacles that can shape real progress. Thank you for an energizing and thoughtful exchange.