Kelly Goles

Thank you. Hello. Test, okay, good. Thank you. Thank you, LeanData team, for the warm intro. We are so excited to be up with you here today. I know for me, I’m having a little bit of a pinch-me moment. I started my career working at LeanData as a new college grad, and now we’re on the big stage. So thanks to everyone for that support.

Awesome. So let’s go ahead and talk a little bit about who we have up here today. At Nvidia, our team is Revenue Marketing Operations. We sit within a group called Revenue Marketing that’s heavily focused on how to have enterprise marketing drive revenue.

My whole team up here specializes in lead management and sales engagement. I also noticed they’re all wearing the same color palette and didn’t tell me, so that’s going to be discussed in the next team meeting. But I’ll let my team go ahead and introduce themselves.

Dante Zanotto

Hi everyone. I’m Dante Zanotto, and I handle lead management on the Revenue Operations team.

Ashley Huddleston

Hi, I’m Ashley Huddleston. I’m in charge of our campaign object and also help with sales engagement tools.

Rebecca Nguyen

Hello, I’m Becca Nguyen, and I do campaign enablement.

Kelly Goles

Great. Okay, evolution of AI. This is actually a slide that we show customers at Nvidia. It’s the way that we’ve seen AI evolve, and I think it’s also relevant to how we’ve seen the martech space evolve over the last few years.

When I look at some of the early projects that we started working on, a lot of it was about how to utilize generative AI—maybe it was ChatGPT or Perplexity—to help write sales cadences, or even in the work we do internally within our teams around emails or certain projects that way. Now we are squarely in the space of agentic AI.

If you’ve been at Dreamforce yesterday or last year, you probably saw that’s the way it’s going—physical AI with a robot. I’m not sure where that’s going to be for us, but I hope it’s not too soon in the RevOps space because I’d like to come to OpsStars again.

Last year, we presented on this slide. The idea around this customer journey was that, in traditional lead management, there were these three different components. There’s the idea and concept of some form of a trigger when you’re looking at leads, there’s an orchestration element where you’re making sure that you’re passing those leads to the right people at the right time, and then also making sure that they’re having the right conversations with engagement.

There’s been a whole martech stack dedicated to this that we’ve been putting together and working with certain teams on. Even in the intent model, there were early signs of AI, where there’s actually a developer team within the larger Revenue Marketing org that created a machine learning model that we’ll touch on.

We were starting to use generative AI around things like how to write sales cadences or emails, and we knew there would be some form of agent involvement. Last year, we just put a big box around it and said, “We know it’s coming.”

This year, we have a better idea of what this might look like. We took the same concepts—triggering, orchestration, and engagement—but broke those pieces up a bit in how we prioritize projects on our team.

We know, for example, there’s going to be a lot revolutionized with AI and agents in terms of signals, which you’ve probably heard a lot about today, and agents using those signals to help orchestrate. We also know some things will not be utilized by AI. The technical term for this is deterministic versus non-deterministic.

Deterministic, for example, is if someone fills out my “request contact” form, I know they need a response. AI can’t say, “No, don’t respond to them,” because that’s a customer experience issue.

So we’ve really been building out this concept of how to incorporate some of the traditional automations we’ve been doing, some things that are right for agentic workflows, and, on the engagement side, how to leverage fully agentic engagement paths as well as do a lot of research and enablement. How do we make sure we’re continuously learning from this whole loop we’re creating?

I’ll pass it to others on my team to explain what we’ve done and where we’re going.

Dante Zanotto

Looking at the orchestration of our lead data, there were three areas where we were struggling to improve automation performance.

Starting off with routed leads to sales, there was a challenge in finding a balance between quantity and quality of records. Next, on the opportunity front, we needed a way of recording sales and primary customer data into our CRM. Lastly, on the contact side, we needed to identify high-intent contacts that were already in our database.

So looking at these three areas, we asked ourselves two questions: how could we improve these data points, and where could we fit in AI to help alleviate some of the headaches?

For starters, on the lead front, we use a machine learning model within our data warehouse—think of a tool like Databricks—that could locate and detect junk leads and remove them from the lifecycle. Since going live, we’ve detected 170,000 leads and removed them from the database.

On the opportunity front, we use a revenue intelligence and sales productivity AI third-party platform that can look at emails and meetings between sales and customers—think of tools like People.ai, for example. From this, we identified 39,000 contact roles and added those to opportunities.

Lastly, with contacts, we used our data warehouse machine learning model to look at activity, demographics, and account strength, and also pulled in that third-party AI capability to see engagement. We could identify high-intent contacts and send them to our SDR teams. Since going live in August, we’ve seen a 6% increase in the number of contacts routed to sales.

Tying this all together with our buying committee automation, we’re now in the final phase of deployment. Traditionally, we’ve identified a buying committee by starting at the account level, where we look at three to four buying center personas. Then we use our in-house propensity-to-buy model, which looks at past opportunity data, activity, and engagement to rank accounts based on how likely they are to purchase.

Once we have those accounts ranked, we identify which contacts under those accounts to route to sales. Within this process, we asked where AI could fit in. Luckily, we already had a model built around contact identification where we could pull in engagement criteria from that third-party AI tool to identify engaged contacts.

Anyone without engagement or opportunity association could have opportunities created, linked to those contacts, and routed to our SDR team for follow-up. Now that I’ve covered that, I’ll hand it over to Ashley to talk about our sales outreach and how AI fits in.

Ashley Huddleston

Thanks, Dante. I’ll start with some sales engagement. Our data operations team built a pipeline that took data from our CRM, developer database, web analytics tools, and other data sources we couldn’t previously pull from.

Using our Nvidia inference microservices and our in-house built AI sales assistant, they generated valuable insights, detailed visualizations, and enriched our existing data. From this architecture, we delivered AI-generated activity summaries. These are generated on our leads, contacts, and accounts, giving insights to salespeople about what their prospects are interacting with—all in one field.

Next, we’re working on AI-generated sales email templates. These will be populated on the lead and contact objects so salespeople can copy, paste, and send. We’re also looking to use the same architecture and model to integrate with our sales engagement tools to improve the personalization of automated emails.

I’ve included examples here. The first is our AI summary—it gives top products, top workflows, and interesting activities from recent history. This was super helpful because we could include web activity from our higher Nvidia domain, which we didn’t normally have in our CRM.

Then we have an example of an email template. The model generating these templates was trained on our top-performing sales engagement content. I’ll pass it to Becca to go more into our AI initiatives around content creation.

Rebecca Nguyen

Thanks, Ashley. When building on top of the AI-generated email template, we wanted to enable our sales team with more than just the intro email when reaching out and responding to customers.

Our sales team uses sales engagement platforms like Outreach, Salesloft, or Salesforce Sales Engagement. These tools allow automation and streamlining of outreach and responses. One of the biggest hurdles with these tools was creating unique cadences.

As Ashley mentioned, and as you saw in the diagram, there are many different lead sources—from marketing initiatives, partner co-sell motions, and event lists. The question for sales was: how can we cater our Nvidia message to each unique customer interest and experience?

We’ve leveraged LLMs—things like Copilot, Perplexity, ChatGPT, or Writer AI—and trained the model on five things:

1. Compliance — This sets guardrails.
2. Engagement best practices — Granular details like ideal subject line length, tone, and regional language.
3. Cadence type — Whether inbound, outbound, form fill, or event follow-up.
4. Top-performing Nvidia cadences — What resonates best by region.
5. Nvidia’s full stack — Hardware and software context.

By training on these five things, the model outputs a unique cadence best fit for the situation. This process, which used to take weeks of back-and-forth between marketing, sales, and compliance, has now been shortened tremendously with AI.

Now I’ll pass it back to Ashley to talk about our agents and how they work together.

Ashley Huddleston

As we bring in all these tools and build agents, each is good at its individual purpose. That’s the bottom row of this flow. We’re finding problems, solving them, pulling data from sources we couldn’t before—but we’re still siloed.

We needed to take a step back and look holistically. How do we orchestrate across these tools and agents so they interact, improve experiences, and avoid chatbot fatigue?

We’re focusing on three main use cases:

1. A sales-focused account research chatbot
2. An Ops Studio agent experience for data transformation
3. An autonomous AI SDR

For example, a chatbot might be asked, “Show me all contacts under my accounts I haven’t reached out to yet.” That’s simple—just CRM data. But if the user then says, “Send a follow-up email to those contacts,” the bot now needs to interface with the CRM, email system, and content models.

That’s where A2A or MCP architecture fits in. MCP is how agents interface with external tools and APIs; A2A is how agents talk to each other. Different tech stacks, different back ends—but as long as they share a standardized A2A or MCP code snippet, they can plug into each other like a universal socket.

For our AI SDR project, this means integrating with customer service chatbots, marketing and sales materials, Nvidia glossaries, and more. These A2A and MCP connections improve both the customer and backend experience.

Kelly Goles

A lot of content, but there are a few key takeaways. First, as you’ve probably noticed, our roles are shifting. The idea of “citizen developers” in low- or no-code tools has now evolved—we’re managing agentic workflows.

Even at Nvidia, it’s a big shift from how we used to work. We’ve had to learn and develop new skills to meet these demands.

Second, good data in means good data out. We spent a lot of time automating data collection and cleanup, which is also why you see vendor consolidation—simplifying where data is managed.

Finally, we’re hyper-focused on user experience. If we’re building specific, purpose-based agents, we make sure the studio and sources they use are included and usable.

And lastly, regarding MCP and A2A—this is new, really only in the last year. Anthropic brought this to us, and if you were at Dreamforce, you saw that instead of AppExchange, there’s now Agent Exchange. It’s opened up a whole new world for those managing software integrations—now, agents can connect without needing full developer teams.

I’m excited to see where this goes, and I know our team is exploring how to leverage it within our existing martech stack, which many of you probably have too.

Thank you so much for having us, and thank you to the OpsStars team. Feel free to connect with us.