Turning Customer Context and Insights Into Autonomous Revenue Action

Julia Nimchinski:
And next up, we’re joined by Ori Entis, EVP and GM of Product and Engineering at Gainsight. What a pleasure. Welcome to the show, Ori. Welcome back.
Ori Entis:
Thanks a lot, Julia. Great to see you again. How are you?
Julia Nimchinski:
Super excited, I mean, you’re shipping at the speed of light, so can’t just wait, you know, to get into it. Tell us more.
Ori Entis:
Yeah, I wish we’d be going even faster. But yeah, let me, let me share this here, just a minute, and… We’ll kick it off. Okay, so, hi, everyone. Julia, good to go? Okay. Okay, hi everyone. I’m really excited to be here today. I’m Ori. I lead product and engineering at Gainsight for the CS Staircase business unit.
I also see some other aspects of the business unit, so I’m gonna… I’m gonna kind of balance today between technology and some of the business challenges, but really what I’d like to dive into is how the post-sale operating model is evolving around both humans, digital, and of course, agents, which I think is the hot topic these days.
But I’m gonna cover our three main areas, and we’ll leave some time for Q&A, so post your questions, and, you know, happy to have a chat, more than me just presenting. So, we’ll talk a little bit about how post-sales work is changing, and also the roles.
And then an AI framework that we think is applicable these days in this post-sales world, especially as, you know, we’re seeing many leaders kind of being challenged with, you know, how do I apply a strategy around AI and agents to solve a business problem?
And the last is we’ll talk about various aspects of building agents in the post-sales, and also the change management involved in using agents in the post-sales world. Okay, I’ll kick it off with a classic customer journey.
The famous customer journey, which, you know, here is broken down into 5 different steps, you know, it could be 4 and it could be 6, but really you go through, you know, the deal closes, the customer converts from a prospect to a customer, and then it goes through a journey of, you know, onboarding, an adoption, an enablement phase.
Hopefully, things are going well, and there’s a potential for expansion, and then, of course, that coveted renewal, which, you know, in order to retain the customer. Along this journey are many, many touchpoints and activities. I’m not going to go through all of them.
A lot of them have been predominantly human-led, various meetings, various enablement sessions, scoping calls. And then, over the past few years, also involving some automation and technology to help, drive this customer journey in a way that feels, you know, fluid and, of course, yields that outcome.
And this predominantly has been led for… by one persona, the CSM. And in a lot of ways, the CSM, the Customer Success Manager, has been, and to some degree still is, a jack-of-all-trades, if you think about all the different roles, this person or individual needs to kind of oversee.
From, of course, building relationships and maintaining them with the customer, overseeing the onboarding and adoption and value realization. There’s always some technical aspect, either trying to actually solve it on their own, or that internal coordination and external coordination, which can be very time-consuming.
One of the biggest aspects, of course, is risk identification in order to drive for that retention and renewal. And of course, over the past, I think, 2 to 3 years, a lot more focus on expansion and ownership of revenue. And it’s a lot.
And I think because it is so much to do, and also, especially if you’re in a company that has multiple products, and these products are shipping new features and new capabilities, the surface area just becomes a little bit too much for an individual to really successfully oversee. And what we’re seeing right now is a split of that role.
into sub-rooles. Now, this is not necessarily happening across the board in every company, but we are seeing a trend of breaking it down into specializations. In some cases, it’ll be one person that does all three. In others, we’re seeing either two or three different roles.
We’ll have a more technical, a TAM-like, or a customer success engineer, someone who’s very deep in the product, and kind of like an FDE motion that can solve architectural and technical issues. We’ll have a more kind of commercial-focused persona, like an account manager. The deal is really with the renewals, negotiation, expansion, and so on.
And then, maybe the core core aspect of the CSM remains around the adoption and value. And sometimes this can be embedded in the other two, and sometimes it’s extracted to a third. So not only is the persona, the human aspect of things changing, but also who’s doing the work, and how’s the work being done.
And again, over the past decade, you know, we’ve kind of shifted from a human-led to digital. Digital meaning we have some deterministic automation systems out there that know how to interact, you know, send various either emails or in-app engagements or other means of interaction with a customer, but still very kind of.
you know, deterministic in nature. And then, of course, in the past, you know, year, I would say, is predominantly kind of where agents are starting to pop up and help us solve those problems. Now, the main challenge right now being faced is really, how do I scale an operation without, you know, growing headcount and achieve those business goals?
And at Gainsight, from a product perspective, we like to think about things in a framework because it helps us in both build the product and also have a conversation with a customer. And what I’ll present here is a high-level framework for the post-sales, and then from there, we can talk about how that fits into an Agentic strategy.
And what I’m going to do is I’m gonna run by a very typical problem that, you know, many folks face in the post-sales world, which is the GRR, or the retention problem. And I have numerous conversations with leaders in organizations, and, you know, everyone wants to improve their GRR.
Every percentage point is usually worth a lot, a lot of money and hard work that was You know, thrown into acquiring those customers. So this is the statement of the problem, right? I have a GRR problem, or I have a GRR challenge that I want to improve.
I have acquired, over the past few years, numerous different technologies and tools, you know, CRMs, CSPs, various support technology, and even some, you know, some new AI capabilities. I also have a directive, typically within the company for my board or the CEO, about utilizing AI and without actually a growing headcount.
It’s not necessarily that I have to shrink my headcount or so on, but I have to really kind of be conscious of headcount, and I have to utilize AI. And I think where a lot of folks struggle is, like, so what is that strategy? How do I actually… You know, manifest this problem and solve it, with the newest technology.
And, the framework is, in this case, is really, based on a kind of a learning curve, or a closed loop, which is a detection of, for example, in this case, in this use case, it would be detection of any kind of risk. Diagnosing, you know, why is that risk happening?
So, an example could be, I have a drop in adoption, or I have a stakeholder that moved out, so why is… why is the adoption dropping? Is it because there’s a technical misfit? Is there a problem, or is there just something else that’s happening there on the enablement side?
To determine what is that action, and then learning if that action we took actually had an effect on the outcome, and then, you know, going back in that learning loop. And in order to manifest this, you know, in an organization, we propose a three-layer approach.
The first layer is what we call the context layer, which, you know, kind of sits right above the data layer, which many, many companies already are solving for by building, you know, their own data lake, and then where there’s a lot of raw data. So the context layer you could think of as a layer that has a first level of analysis.
Then there is a deterministic… sorry, a determination level, or A layer where we try to understand what is the next best action, and then the activation. On the context layer, we’ll look at various signals of… it could be interaction, it could be product analytics, it could be commercial signals, and we’ll do some initial analysis on that.
For example, on conversational data, we’ll look at things like sentiment, relationships, engagement. On adoption, we’ll look at trends of usage, and so on and so forth, so that there is that first level of understanding of the signal.
The second layer is probably, in some ways, the most complex one, is determining what is that next best action to take.
And here we employ a mix of machine learning technologies, as well as LLM layered on top of that, and we look at those various signals from the layers below to understand, you know, looking at the wide range of customers, where is there correlation between an action. And an outcome.
And correlation and causation are not always the same thing, yes, bear with me, but in some cases, we are able, actually, to determine causation as well, so we can help guide what is the next best action. And then the third is the actual taking the action, or the activation. And here, we have a choice.
We could have a human take the action, so an example, like, if it’s an executive outreach, that that’s the right thing right now to do. Obviously, you know, the account manager or someone else in the team should reach out.
Digital would be more of a deterministic flow, where we want, we want very little chance of deviation from a specific plan, and we want it very structured. So think about blasting kind of a campaign where you don’t need a lot of personalization.
And Agentic is where we need something a little bit more flexible and adaptive, that can respond to various signals in real time, and also drive personalization. And here, examples would be on the renewal management side, on adoption recovery, and outreach. Alright.

So we talk a little bit about the framework and a little bit about where we’re seeing changes right now in the market, and now I’ll switch into the last part, which is how we’re building agents for this reality.
And… Back to the challenges of the post-sales and, you know, hearing some of the previous speakers, you know, the post-sale world tends to be a complicated one to map into Agentic solutions, because it tends to be very open-ended and relatively complex, versus kind of a narrow use case.
And if you think about the reality of a customer in a pre-sales and a post-sales reality. I think in many cases, the post-sale tends to be more noisy, more interactions, and more options to kind of deviate from a structured flow chart.
You know, if you think about an example of what an SDR outreach process looks like versus an adoption play, in many cases, an outreach is a very scripted, you know, flow, but an adoption play could be a very chaotic one.
That means, you know, we could start off from numerous different points, you know, there’s an adoption drop because of either bad enablement, bad onboarding.
Some, you know, product issue, some movement in the team, and each one of those will throw you in a different direction, and it’s not always very clear that you can actually build that flow up front. And that’s that… what we call that fuzzy work, you know, from the term from fuzzy logic.
We tend to think that those types of use cases are better mapped into an Agentic flow versus a deterministic one.
And not surprisingly, like, I think what’s… whatever everyone out there right now is looking at from an Agentic perspective, we look at, you know, three different aspects of the agents, the context layer, the harness overall, and then I think one that we don’t talk a lot about is the delivery.
Like, that means, how do we actually get these agents to work in, you know, in the environment, in the company? And I’ll talk a little bit about that. All right, so why is context critical? I think right now, where we are in July, I probably don’t need to explain that anymore.
I think the beginning of the year, I think there was a lot of questions about context versus intelligence.
I think context and, you know, and overall the harness are probably becoming the more critical aspects and where you could really become, you know, a little bit more specialized versus of the intelligence, which I wouldn’t say it’s being commoditized, but there is… there are a lot of options out there, and most, you know, most everyone has access to the same type of intelligence.
And at Gainsight, what we try to do is we try to look at the context layer as almost like a digital twin. If you think about that term that comes from the industry, like, how do you build a digital twin of a real-world object? You know, just think about, like, an airplane engine.
You try to build a model so that you can simulate it, online, and we try to do the similar thing. We try to collect as many real-world signals around, you know, product usage telemetry, about human interaction, financial data, and so on and so forth.
And we build a semantic layer and knowledge graphs in order to be able to kind of mimic or simulate that behavior in our digital world. And this is used, obviously, to help us predict and also determine what we need to do next.
The harness is, is an area where we spend a lot of time right now, because there is a sensitivity between deterministic and probabilistic approaches, and I’ll give… it’s best explained through an example.
If you think about a renewal process, it’s typically you want to reach out to your customer, you want to notify them that it’s time to renew, and you want to be able to propose a commercial aspect to that renewal flow. Maybe you want to increase prices, maybe you want to give them a discount, maybe You want to upsell or cross-sell.
Now… you don’t want to leave all of that to the reasoning of an agent or an LLM, simply because you don’t want any kind of mistake that could affect your revenue. And I’ll give an example.
You can see an email here, there’s an outreach, it’s billed by an agent, but you can see that some of the content in this is really up to the agent to define, especially the opening. You know, the agent went and researched and found out that there was a blog post, and used that to have a bit more of a personalized outreach.
But when it comes to the contractual side of things, you know, the actual, you know, dollar amount or the discount approach, this is where you want to have very strict guardrails, because imagine a situation where your agent, you know, goes wild and gives a 90% discount to all your customers.
So clearly, you don’t want to be in that reality, and your customer, if you’re building an agent for customers, you also want to give your customer the confidence that they can control the agent. And by the way, this is not that different than how we do it in the sales world, you know, on a human front, right?
Typically, account executives and so on have a certain range of discount that they can apply, and beyond that, they need to go ask for approval. So this is a similar approach, and we’re taking that from the real world and applying it into the Agentic framework. So that balance between deterministic and probabilistic is really, really key.
And then we have some design concepts that, you know, we’ve learned across the way. I think probably anybody who’s listening is building agents, a lot of these probably, are relevant, but, you know, simplest solution, very often, wherever we can, you know, approach things in a deterministic way, we will.
And, you know, obviously utilize the probabilistic nature of the LLMs, where we can. But also a lot of effort on the operational side of validating, building evals, sometimes humans in a loop, and that really is on a use case basis. Some use cases are more sensitive, like the one I just showed. where you’re dealing with revenue.
And in some cases, maybe the tolerance for an error or mistake is a little bit, you know, you’re a little bit more lax around that. So, that really depends on your customer. on the use case, and then the overall kind of approach to the solution.
And very often, these things are not out of the box, but they involve a very close working relationship through an FDE motion with a customer in order to tailor, you know, the balance of those factors that I just mentioned.
All right, the last part really is the delivery of kind of the Agentic value and AI capabilities in the customer success organization, and then dealing a little bit with, kind of, change management and some examples here.
So, kind of going down history, you know, just kind of memory lane a little bit, and it’s not a very long time, it’s just been about a year, you know, AI has been in kind of the customer success technology for some time. It started off kind of by alerts and, you know, typical dashboards, as you can see here, so either health or risk alerts.
We moved into a more kind of supportive role for people, so where you can ask questions, and, you know, the LLM or the application, in this case Staircase, kind of goes and looks at various context layers and provides answers, so preparing for meetings, or understanding context, you know, if you’re new to an account, or if you’re just stepping in.
Going from there to a more prescriptive, you know, kind of role of an assistant, taking it more from a Q&A, but looking at things in a broad range, like at a portfolio level, and starting to identify trends within your book of business. And then switching into what we call the analyst Agents.
And these are no longer a Q&A approach, but these are analysts that are running 24-7 in the background for you, and trying to understand, you know, if this is a time for a handoff between sales and CS. identifying any risk without you having to prompt them, and also identifying any opportunities without you having to prompt them.
So these are running in the background, you know, and making sure that they can surface things to you and to other agents in a proactive manner. And just to go into a little bit of the architecture in one of these, on the expansion analysts, for example.
It’s, it really is looking at your book of business, your entire company’s book of business at times, and extracting patterns out of successful expansions. So, think that, you know, for example, you have a cross-sell. It’ll look at conversational and other signals to see what patterns seem to be common across a successful expansion.
They’ll build this, this library of learnings that then can be used by another part of the agent, the signal scout, to go and look at new accounts or new interactions and see if they have any resemblance to what the previous historical successful expansions had.
And if it identifies that, it will notify, you know, the account executive or the CSM that there’s an opportunity for expansion here, based on those signals, those previous successful signers, but it’ll also kind of give you those next best steps, or the action plan.
So, for example, it could be that the highest probability here is a selling another product because the customer explained in a call that they have this problem, but they didn’t know that you have a product that can solve that problem, and therefore there’s a high probability that you could have a successful conversation and potentially an expansion.
And based on previous conversation, it’ll also give you some you know, nudges and directions on how to have that conversation with the customer. So this kind of takes away that, you know, spray-and-pray approach that we typically had historically in the expansion world, and moving it to a much more deterministic and data-driven approach.
And then, of course, in the world of headless, you know, we’re enabling, you know, our users to access all this technology through a headless approach, and merge it with any unique data that they have, and any LLM of choice that they’re working with.
Once again, this is, you know, on the delivery front, it’s very important to meet your customer where they are, and if your customer is now moving to a more kind of LLM-centric approach. to interacting with software, you want to be there, and definitely we want to be there.
So every one of our products right now is in, you know, you can connect with MCP into an LLM of choice and use it in that way. All right, I’m gonna end it with kind of the holy grail, which is kind of autonomous agents end-to-end. So we went through different layers of autonomy, and in this example, we’ll talk about kind of a renewal motion.
And, if you think about the renewal motion, which I explained earlier, it’s identifying, you know, that there is a renewal coming up, crafting the strategy around the renewal based on the health of the customer and various signals that have come across, you know, in the past few months.
And then, deciding what is that renewal going to look like, both from a commercial terms and from, you know, from the actual interaction. And what we have here, and this is a little bit of a, you know, a heavy slide, but we have basically 3 phases, a before, during, and after.
Before the renewal, we collect data, we built that context layer to understand where the customer is, and based on that, we will build a strategy and a plan to activate the renewal, both at personal outreach and the commercial terms.
During the renewal, we will work integrated into the sales machinery to interact with the customer through a digital means of either email or voice or other forms of interaction. And then, based on that interaction, there’ll be probably a loop. You know, if it’s going well, the renewal won, that’s fantastic.
If there is any, you know, challenges or problems, then perhaps it’ll go back into the reasoning loop, and a new approach will be suggested, or it could be that you hand it off to a human, if that’s, you know, where you want the Agentic flow to end. And this goes back to my earlier term, is like.
We were very conscious that we’d have to meet our customers where they feel comfortable, and in some cases, that last mile, the renewal, you want to be handled by a human at this point, until you feel more confident with the Agentic flow. All right, I’m going to wrap it up and open up to Q&A.
You know, if we look at, kind of the Agentic areas that, you know, you want to focus on, I’ll skip this. It’s really, in my mind, you start with kind of the business problem. And in this case, like, you know, what we were talking about earlier is start kind of where do you want your GRR, to go? You know, what is the goal?
How much do you want to increase it? You want to build kind of a learning loop, a closed loop, where you could do that detection, diagnosis, action, and then decide if your action has a positive effect on it, and then learn from that.
And of course, the activation of that will be through those 3 layers that I described earlier, the context, the decisioning, and the activation. And then, at any point in time. you should look and evaluate that mix between the human-led and Agentic or deterministic-led.
Those are your three, kind of, areas where you could have people step in for the more complex or a judgment task or relationship ones, and maybe on the scale side, you do some other types of approaches.
And then the guardrails, of course, I didn’t spend a lot of time on that, but especially if you’re dealing with, you know, larger, companies and enterprises, there are a lot of questions around security, privacy. You know, role access, and many things that are really, really important, especially in the Agentic world.
Now that they have more capabilities, you want to make sure that these things are running kind of in an area that you feel comfortable and safe. Alright, let’s, let’s open it up to any questions.

Julia Nimchinski:
Amazing presentation, Ori, and really amazing to see the progress throughout the years. One of the questions here is, once the system detects risk, how far should autonomy go?
Ori Entis:
Yeah, that’s a great question. I think, so the… so I think it really depends on the type of risk, and, you know, we can… every company can break down risk into different areas. I think, some… some risk is easily, kind of directed to a human, and then some has to be directed into an automation, and I’ll give an example.
If the risk right now is that your key stakeholder left. and that person was your champion, clearly one of the things you need to do is build a new relationship, and right now, we don’t really feel that Agentic technology can kind of fill in that gap.
On the other hand, if we have a mix of users, some of them are, you know, heavy users, but a large number of them are not, this is where we could employ, you know, kind of a mix of education and some outreach. That could be Agentic in nature and very personalized.
Julia Nimchinski:
And then another one here, every business defines customer health differently. How does Gainsight help teams determine which signals actually predict success, and which are just basically, yeah, noise for… for their customers?
Ori Entis:
Yeah, that’s a great question, and health is a topic that I’ve spent, you know, probably the past, you know, 10 years on.
The answer is that you need… there are certain things that we feel very opinionated about that need to be part of your health score, and then there are things that are very unique to each business that, you know, you know that are in the nature of your product or your services, or from your own analysis.
And what we like… we have a framework that we’ve been using over the past few years, it’s called DEAR, and regardless, that framework can be fed by either machine learning, AI, or human input. And I’ll give you an example.
The relationship, the R part of it, you know, historically, a CSM would go in and fill in and say they have a really good relationship with a number of people. I heard the previous speaker say, you know, talk about multi-threading. So, do you have multi-threading? Do you have the right relationship in place?
And then a person would go and fill it in. The good news is today we have technology that can actually do that, and also assess the quality and coverage of an account. But on the other hand, when it comes to sentiment, we also have capabilities on the AI side, but sometimes we also want to have that enriched by a person.
And then, when it comes to telemetry and usage of your product, that’s where, you know, typically we will work together with a customer to build a machine learning model, and also you know, if the customer already has some solution there, you know, we can leverage that in their framework.
So, the short answer is there are areas that we’re very opinionated on, and there are areas that some flexibility.
Julia Nimchinski:
Or in one last question here. So, customer success teams already have plenty of alerts, and folks are asking, what’s your definition of an actually useful signal, and what are the strongest predictors of churn?
Ori Entis:
Okay, those are two… and I have, what, one minute? So, first of all, reach out to me offline. Happy to kind of have a deeper conversation on that. I think, yes, I think the point is that there can be a lot of noise with alerts and signals, and I think, you know, one of the things that AI is really great at is kind of filtering out the noise, right?
So. and funneling it off. So, a lot of these alerts can be handled today by simple agents. It could be an internal notification, or it could be an outreach to a customer.
And then we have the ability, since we can see which alerts are correlated to outcomes, we can basically kind of prioritize certain types of alerts based on their severity, and based on the correlation that we see with, let’s say, churn.
And then when it comes to churn, and this is a very short, you know, answer right now because we’re running out of time, but when it comes to churn, you know, there are a number of factors that, you know, no surprise, you know, you’ll have, but clearly, usage of a product is one of them, but it’s not the only one, and it’s not only deterministic.
It’s usually a multiple, you know, multiple different aspects. Including relationship you have with a customer. the overall knowledge and usability of the product, and then, kind of having a shared mutual value discussion with the customer to make sure that they’re actually getting an ROI, because everyone is focused on ROI now.
Julia Nimchinski:
Thank you so much, Ori.
Ori Entis:
Thank you a lot. Thank you, Julia, it was great seeing you again.