How Every Renewal Teaches the Next

Julia Nimchinski:

And next up, joining us, Grant Clarke EDP and GM of Atlas at Gainsight. Grant, welcome. So great to have you here. How have you been?

Grant Clarke:

Great, thanks for having me.

Julia Nimchinski:

What’s new over at Gainsight?

Grant Clarke:

A lot. I’m excited to talk about some of the things we’re learning.

In the past year, really just since May, we’ve launched our roadmap towards an AI-native service direction, so I’ll talk a little bit more about what an AI-native service is, and what we’re learning as part of that at Gainsight, I think there’s a lot of companies I was just hearing, you know, the tail end of that last conversation is, I’ll probably be saying some similar things to reinforce the same points, which is good, that’s a good signal, right?

It’s a lot of similar minds experiencing this in the open together, just because it’s so new. across the board.

Julia Nimchinski:

Great point. Great agenda point. Let’s dive into it.

Grant Clarke:

All right, so I’ll present a few things here, a few slides to make it easy. So let me pull it up. Alright, hopefully that’s coming across good. All right. Well, yeah, excited to be here, and thanks for having me.

Atlas at Gainsight, as I mentioned, is an AI-native service specifically for the renewal motion for tech companies that need help around that long tail of their own customers that they just can’t put the right type of economics behind, either through adding more people in that long tail of low-dollar renewals, trying to figure out how to deploy agents or automation or digital coverage.

And so I’ll talk a little bit of what we’re trying to do to solve that and what we’re learning behind it. But I think as a theme overall, which I understand as part of this series, just asking the question, how does your go-to-market system get smarter every time it runs with AI? And that’s something I’ll try to show on what we’re doing as part of that learning loop.

As you just heard in the previous conversation, we believe in this idea that there’s the human in the loop. There’s never really an example that I’ve seen where the autonomous agent can be autonomous 100% in a workflow. There’s always going to have to be handoffs back and forth, and how does that actually look and feel? Talk a little bit more about that, that we’ve learned. And to be upfront, we’re early.

Gainsight’s early, a lot of companies are early, we’re trying to see what we’re learning and how to take those into our roadmap. So, a little bit about me. I’ve been in this renewals space for about, well, over 25 years. I started my career very early on doing renewals. right out of college, which is kind of funny, in, like, the early 2000s.

I got my job because they said, hey, you don’t have a lot of sales experience, but if you do renewals, you can’t mess that up. That was kind of like how people thought about it back 25 years ago. So I was able to really see from the ground up what that post-sales process looks like inside of tech companies.

That led me to ServiceSource, where for 15 years, I helped lead outsourcing the B2B renewals post-sale motion for major tech companies from, you know, call it the early 2000s up to 2020. Which was a great experience being able to see behind the scenes of these companies of where they struggled around that business process. And we did it with people, and we did it with process.

And we tried a little bit of RPA or automation, which at the time, man, I wish we had Agentic AI back then, but we struggled our way through people and process for so many years. And then I landed at Dropbox and led their revenue operations team.

Which was a great way to see it, right kind of the early GPT years, and this is around 2022 when I got there, helping them launch their AI portfolio, and how do you essentially pull everything from an AI working method inside the company, a very engineering-first company. And so we really tested a lot of different models on how to do work better. And so combination of those things, what led me to Gainsight then is.

How do you bring this renewal motion to life in this recursive, iterative, human-in-the-loop process, combining an AI agent that can handle the renewal workflow with the human in the loop who’s the renewal expert and the operating model around it so that you can keep iterating on it? So, what is an AI native service? This may be a new concept to a few of you, but this is actually only about 18 months old.

Jake Saper at Emergence Capital coined this term as he realized some of the trends that are happening right now. And just to read it off, it’s a vendor who owns and delivers the outcome, not the software. The customer buys the result, and the vendor holds all the accountability as a single point of responsibility to deliver that outcome for the company. They don’t ship the software to the company.

It’s kind of this new concept of, if you’re performing AI-based work. And you have human-based work that are always changing back and forth in how you create Agentic leverage.

If you keep that as the vendor, you get the benefit in order to build improvements, and as long as you’re delivering back to the vendor that outcome, you’re able to maintain your value in the price point, and the customer doesn’t have to worry about how you do it. Because you heard earlier, it’s hard. It is so hard to sell and deploy an AI agent into a company.

And it either stall, doesn’t get adopted, doesn’t have the right data or the context. So now, AI’s native services flips that and says. You know what? We know it’s hard. Let us figure it out for you. We’ll deploy the agent inside your environment. We’ll keep the people. We’ll keep building it. And as long as we deliver what you signed up for for us to to do, that’s all that matters.

And that that’s a new trend that’s starting to happen, with other companies. And so, as we thought, or as you think about the progressions, like, how do you get to this level? You think about for two decades, you know, software vendors sold access. Technology, licensing. We’ll help you use it, you know, there’s customer success and services that kind of all come with it.

And then as you go and step to assistance, you know, they pay for training, they pay for onboarding. Professional services comes in and configures, and now everything is moving towards how do we execute so I don’t have to be involved and have the burden, and how can it actually deliver that result? And that goes, you know, back to the AIMS concept.

And the way that, Gainsight is thinking about it, just the way we sell our products, it’s not a single thing. It’s not always just the services model. If you want to buy the application, if you want to buy the agent. Great, just think about it like a car. I don’t want to worry about all this other stuff, just give me the car, I want I know it works, I can drive it off the lot, good to go.

Then there’s some customers that are like, you know what, I actually want to be a little bit more headless. I want to say, I want to configure it. I want the car, I want the base, but I want to be able to build all of the things around the car, the MCP, the APIs, anything else that’s going to get me more context and customization in how I use it. And then there’s a company that’s like, you know what?

I don’t execute well in order to do what needs to be done, let me just hire someone to do it for me. And I just… it’s like taking a rideshare, right? You don’t have to worry about the car, you don’t have to worry about the car you’re in, you’re just getting from point A to point B, and however that’s happened, you’re willing to pay for that outcome.

Okay, so a little bit of what we’re learning, and the way that we deploy our solution, in this renewal world. is we deploy the AI agents that essentially take over that personalized outreach, the first step that it engages with a customer. It eventually builds context, and when the customer replies back to the agent, it can handle that reply in categorization. It tries to interpret, okay, is this customer angry?

Is this customer… they’re on a happy path? Do they have a question? Do they have an objection? And so that AI agent is deployed and starts from a low rate of autonomy as it gets these replies, and it learns. The way it learns is you have the human expert sitting next to the agent. that can see how the customer is responding. When the agent hits a guardrail of context, it flags it as an escalation to the human.

It says, hey, I don’t have enough context to answer this question. Here’s the handoff to the human. The human then replies, takes over the workflow, answers the question. And then, as the last step, we take that as a learning loop. So we monitor everything that happens when the human has to take over.

It allows us to create the training evals for us to rebuild the way in which next time that customer responds, how do we get the agent to hit a threshold where they don’t escalate it to the human? And so you think about those… those types of the activities that happen inside of the entire workflow that we’re solving for.

Some of them are going to be low judgment, some of them are going to be high judgment, but we want to start building it with the AI agent first, and then have the human, drive over time. At the end of the day, the outcome ownership still sits with Gainsight. So we still have to deliver, you know, a performance metric, renewal rate, you know, whatever it may be, and so we own how the first two work together.

That’s part of what… What we control. We started with this renewal process, and I’ll give you this example, because you can think about maybe how it’s a blueprint for something else you’re looking at deploying AI in. But it’s critical to start with the fact that you have to have a clear outcome. We picked renewals because it’s a pretty defined, measurable event.

It’s a candidate to allow us to know that if we are operating as the agent and the human together, and we see the lagging output. There’s always going to be leading indicators we’ll watch, but as long as the lagging output is delivering that outcome, we know that it’s not just a bunch of activity that doesn’t actually result in anything.

So, deploying an AI model or inside of a workflow Just you got to know what your true north outcome is. Like, what is it you’re trying to get to? Not just I save 10% of my work because we’ve heard AI saves time, but the time gets digested or eaten up somewhere else, right? It’s a hard one to pick. Make it a repeatable motion.

Consistency and process allows you to do things like A-B testing, it allows you to understand where the variations occur, and when they at least follow a more consistent, happy path over time. It allows you to build more of the structure that you can scale, and you can take the variables and the variants aside from that process and build context where you think that variable or variation will be repeatable.

So you keep adding back into your repeatable motion, but you have to start somewhere in a business process that isn’t bespoke every time you do it, because your AI model is going to take longer to try to build that context over time. And leads to the third point, the available context matters.

Trying to layer in the different signals, you know, data… we use data from our own platform, where we do customer success, post-sales management for a lot of companies’ end users, so… How do they use the product? What did they buy? When did they buy? How much did they buy it for? Relationships, you know, how many relationships exist inside of the company that are using the product?

All those things are part of the ingestion that, of course, brings more context, but what we’ve learned There’s always the static data set. There’s ways to layer in more contextual real-time signals, so we have a platform called Staircase, which monitors all of the interactions back and forth, either by email, support tickets, if they’re talking to sales, or if they’re talking to services, it doesn’t matter.

We’ll be able to scrape the conversational contexts that are happening that may influence the way that you signal risk or opportunity. And so it’s important to think about your static context and data, and also some of the real-time context that you can gather.

And then on the human judgment side, I think going into a workflow and being honest. where you deploy an agent that is realistic to what the agent can handle, there’s always going to be shades of complexity, and I define it as low judgment, high judgment, a little bit what I talked about earlier. In our world.

Low judgment is, again, contextual outreach, personalized content, response categorization. you know, simple information-seeking requests. High complexity is negotiating, dealing with an unsatisfied customer that’s getting ready to churn, or is risking to cancel. You know, things that are sensitive or complex. We’re we’re just gonna leave it with the humans.

We’re not gonna pretend to put an agent in place that’s gonna either fail through a test process or is gonna give something incorrect to a customer. And what that helps us do is start with a baseline. to eventually decide when do we graduate the autonomy leverage up into maybe medium to high complexity or high judgment over time. Because the more that we watch how a human’s negotiating with a customer.

That will be some of the early requirements we start testing inside of the agent to eventually take that over. But deploying that out of day one, our lesson learned is you start pushing agents into complexity that they’re not prepared to handle. And you, especially when the stakes are high, when you’re engaging with customer commercial conversations, you want to make sure to manage accordingly.

Here’s another thing we learned, going back to this whole learning loop. We actually deployed Atlas as a software-only renewal agent. Where we told companies, you could buy our agent, we’ll help you install it, we’ll help you onboard it. And as it’s operating, when it hits its threshold of context limit, it escalates to an individual inside their company. So let’s say, like.

A customer success manager or a sales rep, you know, wherever it hits its limit. What we learn is that when that happens, it goes into this. black box of what happened? Because when you escalate an action or a call to action into a company, we as the vendor lose visibility of what did they do with it. Unless you go call them, or ask them, or poll them, and of course, these reps are busy.

They’re not going to sit there and be like, okay, I’ll spend 30 minutes telling you what I did. And so we lost our learning loop. We weren’t able to… take what they were doing outside of the agent to say, should you have done that? What can we take you, what can we take out of that to bring it back into the agent? It was harder to do that in an iterative development cycle.

And so, what we did now is say, alright, fine, we’re gonna deploy this as an AI-native service, which means we own the human in the loop and the agent.

And what we can do now is daily, we are seeing what the human is doing outside of the agent, and we have ability to monitor our own people in a way that helps us track their time spent going and researching a contract, or, you know, they had to click 5 to 6 different ways through history on a Salesforce opportunity.

You know, anything that they had to do. we can monitor, and we will build that back into the agent, if it’s low judgment, as I said before, and then eventually high judgment. So, the recursive loop is leverage that every company needs to think about in their journey with that AI model that they plan to become smarter over time.

And it’s hard… I think everyone realizes it’s hard to get the human to participate in that loop unless you either incentivize it for them, or you create a very frictionless process in order to capture that data. So just, like, how it would work for us, the relationship between the human and the agent.

So the agent works it, up to their guardrails, they pass it to the human, they give context to the human on why they passed it there. The human steps in, again, judgment, as I mentioned, is where it’s required for them to act. we then capture it, and then we allow us to build that requirement to feed it back into the agent. So, this recursive model, I mean, this could operate, again, daily.

We have our four deployed engineers actually sit with our people in the loop, and it allows them to ask questions when they see something pop up. We track signals on when they had to take something and why they took it. So that is a very, tight process that we do… we do together. And the way this really works, if you think about the way Emergence Capital thinks about this with businesses.

Over to the right-hand side, typically the green line is, if you’re a traditional SaaS or services business, you always chase revenue with cost, whether that’s sales cost, services cost, or engineering costs, right? You have companies that are willing to buy. But they needed to be custom, and so you have to spend money to keep going at that rate.

And what the challenge is in the AI native service world is, don’t chase the revenue that fast. Don’t chase the company or use cases that fast. That’s called Mirage PMF. And so it says in the orange line, hey, start with a few customers, slow down to invest in your AI leverage with the use cases that you have.

And as you get signal on autonomy, then go back to market and sell that, and then your revenue is boosted by the amount of scale that you have without driving the cost all the way up because your AI agents now have leverage in what they would be delivering behind that. And so that’s the way we’re thinking about it at Gainsight.

Early traction does not mean scale, so we’re being very picky around the first few customer use cases we have to get the right signal. We’re building autonomy by the task. Like, we went down very discreet by task level. What do we think is an agent? What do we think is a human-led? And you earn autonomy gradually.

You boost confidence and quality as you start seeing the right signals to avoid having something just being pushed out as low-quality work. So, wide open, this is a new, exciting, fresh area, at least for the AI native services side. So, we’re riding this as we go, which is fun and nerve-wracking at the same time. But we’re excited to see where it’s going next. So, open any questions. That we may have.

Julia Nimchinski:

Amazing presentation. Thank you so much, Grant. I’m curious about this service model AI native service model that you just presented. Are FTEs part of the model, or how does it work? Internally.

Grant Clarke:

For deployed engineers? Yeah, we think about FDEs almost… Side by side with our… humans in the loop, our renewal agents, as part of COGS.

So when we sell this to a company. we package it as, hey, there’s the AI agent, there’s the engineering that gets deployed, there’s the people we’re going to hire as the domain experts to work with the agent, and then there’s a forward-deployed engineer that sits in the pod with the human and the agent. So I see them as an extension of the team.

Because in your deployment with each of these customers, their version of how they’re seeing the outcome needs to be continually iterative inside of the product. So if we need to change a workflow because it hits a higher outcome, I don’t want to have to wait three months to go to a back engineering team, I want the FDE to sit there and just design that on the fly. And that is a shift, having that deployed.

Julia Nimchinski:

Very cool. And I’m just curious, I’m not sure if you’re allowed to share, but in terms of trends that you’re seeing, is that realm of business getting more traction than the, you know, regular SaaS model?

Grant Clarke:

It’s early, but there are some great examples, so I think there’s probably a few names people recognize. Harvey is another company that is an AI-native service that’s thinking about changing, like, legal services.

So think about, you know, companies hiring a bunch of Legal… in the legal department, a bunch of lawyers that have to review contracts all the time. it’s measurable, it’s repeatable, is the elements that I talked about earlier. And so they’re coming out and driving that as an outcome service for people to say, hey, don’t worry about that, we can actually do that for you.

So I think what you’ll see in the AI Native Services world is a lot of these vertical use cases where you think about insurance claims, you think about legal services, accounting and finance. And then in our case, you know, the customer renewal post-sale process. I think that’s where these things are starting to pop up more.

And then there’s companies, SaaS companies, that are realizing that as incumbents, and so they’re creating their own AI-native service department. inside of their own SaaS company, like Gainsight did, to drive that as a way to manage that trend. So, it is starting to pop up now, probably in the past 2 years.

Julia Nimchinski:

Thank you. And question from the audience. The renewal cycle cycles are very different by industry. I worked in satellite sector, information as a service, and software as a service. How do you address this difference within your tool?

Grant Clarke:

Yeah, just like I talked about the AI native service use cases in verticals with, these back office, we think about the verticalization inside this service as well. You think about hardware, it’s gonna operate differently than software, it’s gonna… differently than, like, healthcare life science renewals, cybersecurity.

They’re gonna have different value propositions, different engagement models, and different workflows. And so we do believe there is a universal A platform that operates and gets configured by the requirements that are specific to each of those verticals. And I think that’s probably gonna be common in a lot of different examples.

I mean, you think about the industries that a company serves, they’re gonna have those nuances, and they usually… if they’re big enough, they sometimes build their own sales departments to be specialists inside of Like a hardware specialist versus a SaaS specialist versus a cybersecurity specialist. So specialization is good in the sense that you’re helping drive more relevant context back into the agent.

And you get scale, because usually when you hire a bunch of specialists, it becomes really expensive, because you have to keep getting experts in a single field. The beauty of this is that you get the experts in the field, they train the agent, and then you need to hire less specialists, because the agent becomes specialist.

So we’re thinking about that as how do we… how do we verticalize the way that the agent has enough context in that… in that use case, or in that industry?

Julia Nimchinski:

Beautiful. Greg, I’m curious, you just alluded to the context problem within, you know, losing elements of the learning loop. You mentioned you were solving it with a human and an agent. Could you share more context? How does it work? Because there is so much, you know. A lot of the context problem just. Basically cascades down to incentives.

So how do you incentivize, you know, teams for various reasons to actually share the context?

Grant Clarke:

So… I would say it’s disruptive in the profile of the human we hire, because if we would be building this in the old way, I’d be hiring a bunch of renewal managers who are renewal experts that know how to hit quotas, and they’re incentivized on, pay me my… give me my paycheck based on the quota that I hit.

We are shifting it a bit to look for people that are of that experience, but are more system thinkers, and are incentivized on improving the system and orchestrating the system, versus it just being how much their own individual work is worth. So the agent and the person Both are the… it’s almost like having a… a partner or a co-pilot inside of your territory.

So when we bring the human on, we say, hey, look, you do have a target, but the way you can get to your target is totally influenced on how much you invest in making your agent smarter with you. And so, they’re incentivized to say, alright, well then. I can get my own personal economics better, because the more I can make my agent smart.

Gainsight’s not going to be on my back on how many hours I’m working, or how… the outcome matters. As long as the outcome matters, then the incentives align for the person to participate to make the agent smarter for the outcome.

So, we’re thinking about the profile differently, we’re thinking about the incentive structure towards the outcome, and we also want to try to make it easy for them, so we’re trying to use other software and other AI tools to monitor their work, so it’s not like this self-reporting work, which just creates friction for them.

And we’re doing it to ensure they understand that’s going to pay them back in us taking that work off their plate, and it allows them to keep, you know, being able to hit their number and hit their goal better over time. But it is a different thinking model, because usually when you hire someone who’s training a model, they think, okay, well, I’m training this model to take my job.

But we’re saying, no, there… we’ll never replace 100% of this role. Like, there’s always going to be high-judgment, complex work. Wouldn’t you rather spend time on that interesting stuff versus the low-judgment, repetitive, uninteresting stuff? So, they look at themselves as human in the loop, they’ll always have a job, but they’re just gonna make their job that much more effective in that process.

Julia Nimchinski:

100%. We just polled our community on some of the biggest challenges in, you know, Agentic GTM, AI and GTM, and, one of them is just the lack of, you know, vision of… when it comes to the art of the possible, and lack of. Speak on the vendor side differentiation. In terms of, you know. the market lands and all of the available platforms, tools, within any context.

But can you speak a little bit more about Gainsight, and how do you differentiate, what’s the… you know, number one use case that, is a winning use case, and how would you advise management to think about all things? Retention and beyond, obviously.

Grant Clarke:

Yeah, so in this example, is it, how does a company differentiate themselves in a very crowded market, and, like, how do you set yourself apart in that AI vision?

Julia Nimchinski:

How do you select the right vendor? Currently, when every day there’s a new vendor popping up.

Grant Clarke:

Yeah.

Julia Nimchinski:

new investment and just not really at this point.

Grant Clarke:

Well, I’ll give you the… biased answer, just because for me, the AI native service model is an alternative decision, because in an ability for someone making a decision that you’ve got 5 different vendors that have all popped up that probably have shades of what they think is differentiated, you still have to figure out how to Go on the journey with them, deploy… implement it, deploy it, and actually make it work for yourself.

So the use case of you still having to make whatever any of those things are possible work inside of your company is difficult enough. An AI native service solution just gives you an alternative to say, someone can do that for me. I don’t have to worry about picking the different vendors.

If I can trust that this vendor, this AI native service vendor, can deliver the outcome, because they have domain expertise, they have the technology behind it, they’ve got the way to access the data. then you kind of take away the clouded bake-off from a vendor standpoint, and you’re trusting an outcome for the AI Native service.

So that’s just another way to make an alternative option where you’re struggling to pick the different vendors. If you still are picking against the vendors. I think it’s still gonna come back down to. your data is going to inform the best context and the best custom solution you need.

I would look at it more around where can you go headless with some of these options versus it being completely out of the gate, you know, just buying it as is. Because I still think you are the best person to know exactly how you can make this agent work, or this application work. The vendor’s gonna claim they know it, but they won’t know it better than you.

So, how do you know where to start from the data that you have and the context you know that is important? And where do you find a vendor that’s willing to partner with you, maybe in a headless way, to get what you need as an early signal fast? And then it can give you a better indication whether, you know, the investment, whether the investment played out.

Julia Nimchinski:

Thank you so much, Brent. And where can our people learn more about this model?

Grant Clarke:

Sure, yeah, it, gainsight.com, I mean, just visit our website. We, we… Atlas is up there now as a way to, you know, learn more about what exactly we deploy, where we deploy it, and how we deploy it. So that’d be the best place to start, or I’m sure you can find me on LinkedIn.

Julia Nimchinski:

Awesome. Thank you so much.

Grant Clarke:

Thank you.