Turn your LLM into a Customer Expert with MCP

[Speaker 1]
Hey. Hello, everyone. How are you all doing?

Good. Good. Yeah.

Yeah. Good conference so far? Yeah.

All right. So it's probably towards the end of the session, but this is the third time we're doing this particular presentation. The first one was in Gainsight's RKO, where nobody knew about MCP, MCP was just coming out and we wanted our own team to be hands-on to experience this magic.

So it was an internal session, February, early this year. That was version one. Version two was in Pulse US, and what we've learned in Pulse US is it's a bit hard to follow for all the sessions members.

So if you are connected to Gainsight MCP, Staircase, or CS, please feel free to follow along. If not, I will be showcasing some prompts in my laptop, and you can have a look at it. I tried to make it as visual for you to take away from this session.

Now this is the version three. The second session in Pulse US, it's interesting because that was in May. We released all of our MCPs, we got probably 100K, that was the usage back then.

And fast forward five, six months, I would say MCPs taking off like crazy. All of you probably already know about MCP. So we're spending less time, I've removed all the intro slides, less time on intro and more time on actually looking at how things come up.

Quick show of hands. How many of you are CSMs who carry a book of business? All right.

How many of you are CSOps, some sort of setup who can configure things? All right. And how many of you are CS leaders in the room?

Okay, so this is gonna be interesting. No one major persona, all three personas here, okay. So I'll try to cover various versions, like I'll touch on a few things for different persona here rather than go deep into one scenario here, cool.

To start off, why has MCP become such a huge thing is in customer success from forever. I've been at Gainsight eight, nine years, from forever. It always takes more time to understand what's happening with an account than to actually help that account.

That's always been the case. And a few years ago, we built this Gainsight home as where we tried to create as much of it within one tab for CSM as possible. But it always is true that it always takes more time gathering the context than actually helping customers.

And that's where MCP came in and it's taking off. This is real numbers. If any of you remember yesterday's keynote, the number was 1.20 million tool calls across all of our products. That was September's numbers, but we are now in October. I took yesterday's snapshot. I refreshed this number yesterday.

And as you can see, only for two products, it's 1.25 already. The number that was prepared for the keynote is already outdated. This is last 30 days numbers.

This is what probably product market feels like as a PM. But it's more importantly, how this is solving a real problem for your teammates and for you as you work with your customers. That's why this matters.

And while we talk, whoever has laptop, feel free to follow our prompt. I have a micro website that I created for this session. It's in the QR code, but if you cannot use the QR code, you can also look at our MCP library.

We've shared this across. So QR code will take you to a page that looks like this. You have some prompts that you can copy paste, follow along.

Yeah, I'll go there. Now, while you connect, I want to talk about what is MCP. This is the only slide that talks about MCP, like the previous sessions had four or five slides.

MCP is giving all your models the right context. It's a protocol that helps us provide this. And Gainsight has a lot of this context about your customer already, like the conversations that you had with staircase, your CTAs, success plans, company records, information, permissions, and all these things.

And Gainsight MCP makes all of this context available to your models, whatever AI models you use, for it to actually help in your day-to-day life. And think of it like a universal adapter between your AI and your customer data. And it works with any of the tools that you're using today.

You don't need to buy a new tool. And it's included in all of your Gainsight subscriptions today. And it's not even read-only.

The best part is it's become this universal standard that every app, every tool is building towards. So for example, I can use MCP to draft my emails, whatever I want to send. I can use MCP to set up some meeting calendars, some block times for me to do my research.

In this session, I probably took five to six takeaways. If all of that were in staircase, I could have just asked MCP, create some time slots for me over the next two weeks to follow through on my to-do list, right? That's the beautiful thing about this, that it's not net new tools.

It's the existing tools, my existing workflow, all managed intelligently. And that's why we see the data. That's why it's taking off.

It's helping people do their job better, faster, efficiently. And what can Gainsight MCP do? So first, you can ask any question about any customer.

As long as it's something that CSMs can see in the UI, their LLM can answer that. For example, do I have any open cases with this particular account? If there is a cases report on, let's say, C360, it pops up.

And again, staircase has all this conversation, so you could pretty much ask a lot of questions. As a product manager, one of the questions I love to ask is, I'm looking at what are the challenges the customers are facing in this particular module, this particular feature that my team is building? Hey, we have a couple of sprints left, so I want to take some minor enhancements that I can do about this particular area in this particular product.

Help me fill up my, give me some roadmap, give me some good to-do lists, and then create the tickets for them. Like, why should I stop at research, right? Create my tickets in JIRA, assign it to the right engineering manager, and so on.

So you could ask, you could act. This is third thing for symmetry, ignore that. The fourth thing is, you are in control.

So all your permissions, when you ask this question, it knows who you are. It ties you to your user record and gain side. So all your permissions, your accounts, all of these are always on it.

And today, I will share some of these aggregated patterns that we are seeing, that I'm seeing across what people are doing. The most interesting fact is, it's not just CSMs who are using MCP. It's sales and executives who rarely log into Gainsight, who hate logging into another tool.

But they're finding real insights, especially across their accounts, but more broadly, across the portfolio, across digital segment, and things like that. And that's what I love about this. Like, it's unlocking the usage or unlocking the insights across your organization.

That's what we're noticing today. Now, not surprisingly, what I've done is I've picked three top three use cases that we're seeing today per usage pattern. And not surprisingly, the number one thing, like 66% of users are asking is, catch me up on this account.

I've used this numerous times yesterday and today. Like, I might be speaking to you and I might be asking for like, hey, I'm meeting XYZ, help me understand, what am I talking to them about? And if you were in any of the EBRs, like in the rooms there, any of those meeting rooms, most likely your Gainsight team might have gotten a pre-prep brief on it about what are the challenges, what's happening, and giving me the context to communicate better with you as I walked into the room with you.

That's know your account, and that 66% of users, two in every three users, are using some variant of that query. The second one, that's basic, right, foundational. The second one is more insights.

People gathering information, like, hey, which of my renewals? I'm a CSM, which of my renewals in the next 90 days show any rewarding signals? It's the next level, right?

People leveraging the intelligence of the models to help do their job, help assist them. Why is this account red? Why is this account yellow, or something like that?

What's happening with it? Going one level deeper than summarize, catch me up, right? That's the second bucket, and we are at a place where one in every two users are interacting with LLM to get these insights.

And what surprised me a lot is the third bucket, getting things done, updating a CTA, or closing a CTA, or logging things to timeline. That's almost like two in every five users. So CSMs who are interacting with LLMs, they're actually doing work.

I mean, it's so much easier, as I said. As a product manager, I hate interacting with Atlassian's Jira UI. Probably not their fault.

We made it, we had too many labels, too many tags. It's slow, our DevOps fault, most likely. But I love interacting with Atlassian's Jira via plot.

All I have to do is tell it what to do, and it does the stuff. And I'm sure a variant of this is true with your CSMs and your team today, right? Like, just type it in plot, it will create the CTA.

It can strike up the tasks that are done. You don't need to open a new UI, you don't need to break your flow, and that's what we see every two in five CSMs doing today. At this point, I want to tell, if you have any questions, feel free to raise hands, we can go deeper, right?

Yeah, please go ahead.

[Speaker 3]
So it keeps repeating that pattern, even if I ask something for a completely different customer. So I don't want the same answer as it gave me last time. I don't want that memory piece.

I want a new answer.

[Speaker 1]
Yeah, could you tell me which AI tool you're using that from? Like, what was your experience there? Is it our co-pilot or?

[Speaker 3]
So yes, we are using co-pilot studio. Got it.

[Speaker 1]
So there is, probably in the demo I will showcase one thing, that there is some sort of memory that all these LLM clients carry through the thread. Like some sort of information that carries through this to and fro interactions. It's usually helpful, but in scenario like what you call it, it's a limitation.

Creating a new chat thread is probably the only solution, or explicitly instructing it to do refresh is the only way forward. Any other questions? All right, let's go through it.

Now, beyond users asking, there is one very insightful trend that helped us formulate our strategy. Like the timelines, right? April, May, that's when we released MCP, like made it GA.

And end of July, what we've noticed is there are some repeat queries that are coming from some of these AI agents. They are probably about 1% of the queries, like 1%, that's what we saw end of July, close to end of July, early August time frame. And fast forward today, how many months, right?

August, September, two months, two and a half months. We see 10% of these queries are coming from AI agents. Some of these AI agents, people are using to prep them for the week, and probably end of the day, at the end of the day, get all my customer meetings, transcripts, and close out any to-do list.

People are already automating their work. It is amazing what people are doing today. At the same time, there is a bit of risk here, like we don't know if they're doing it uniformly.

Everybody is probably running their own thing here. So again, from product side, we're looking at some of these patterns. We're trying to see, like what we hear clearly is, how do I standardize this across the team?

How do I take these best practices, do it with the team? And that's where you've heard us answer that question earlier in the keynote. But all of that is primarily coming from here.

We are seeing tons of people running these AI agents, and it's real today. And the second part, this is something we can't put a number to, but people are not just working with GainSight in their queries. They're working with, they're getting the data, and then they're logging something in Salesforce.

They're doing something in Slack. One use case that we've heard a lot, we've seen a lot is, end of the week, getting the customer summary and posting it, pushing it to customer Slack channel, so that your entire team gets it. Entire team who is in Slack automatically gets it.

So we're seeing these multiple systems together patterns. Again, as GainSight, we can't get numbers there. So nothing there, but it's a real pattern that we're seeing these AI agents do today.

Yep, and most popular ones, all of these patterns, right, are available. We're catching all these good patterns, and we're working with marketing team to put it out here, right out on our website, so that all of you can use any of these to standardize your CS practices. Any questions so far?

All right, with that, let's jump right into the workshop, yeah?

[Speaker 3]
So you've seen those patterns and those standardizations, and I think that's one of the points that I still find probably part of the weakest sections, the outputs, right? So I have a system which calls data, and it calls it in whichever shape it wants to, unless you specifically. Create a set of data.

So that would be for me the next iteration, that you have templates that it automatically can apply, that you either upload because this is how your team runs, these are your templates. And therefore, when it then does its data dump, it dumps it into those templates.

[Speaker 1]
Exactly, yep, so the agents are partly doing those. The repeat queries that we are seeing is go to this type, call to actions, and get this risk call to actions. So an admin or somebody is giving that sort of a template to them, step by step format, so that the agent follows all the steps, gets the data, and fills it in the template.

That's the standardization probably we have much coming through.

[Speaker 3]
Will it eventually move to a proactive format?

[Speaker 1]
Yes. I mean, that's what we all want, right? An always on agent, one for every account, always scanning through the risks and so on.

We have parts of the today, by the way. Staircase today does it with all customer communications. Any risk signals, expansion signals, analyst agents.

We already have parts of it, but I don't think the landscape is evolving. We don't know the end state, how will it be next year? We don't know it today.

So we have some questions here. The first one, answer is always yes. Matt and yeah, we'll figure out how to do that, but we would love to highlight some prompts.

Like we shared it in the keynote, we want that on our community. What are some key prompts that people are using across products? We'll get there.

But can we switch to the workshop? We'll get to the questions next. So for people who are able to follow this, feel free to type the next prompts in your own LLM.

So I have this micro website. There is this copy button, copy the prompt and paste it in your LLM. And for people who are unable to follow, I'll be doing that in my cloud as well.

Without further ado, this is prep me for my call with an account, right? Very standard. So if you look at the anatomy of the prompt, the key things that I want to call here are, I'm talking about what do I want to see here.

I'm telling about what info do I want here from gain side. I want any open CTAs, open success plans, health score. Where can we potentially work to improve the health score and things like that, any recent history.

And staircase, if it is connected, you can get even more insights and information from that. So one thing I wanted to show with this, in the beginning, I ran it in the top, right? The answer looks something like this.

Credit gain site, multiple times, multiple iterations, and after that it created the snapshot for me. Abbott, I'm preparing meeting for Abbott. Here is how Abbott looks like, their health score.

What open work do we have, right? And what happened with them in the last 90 days? Recently, what happened with their account, and so on.

Let's go through the remaining prompts while it returns this. This is the prep me for my call. Whatever I told you as the variation that two in every three queries are coming in, it's a variation of this.

Prep me for my call, prep me, I'm meeting them, prep me for XYZ. And you can see why this is so popular. You get a snapshot of whatever is happening with their account.

And if you use a reasoning model, you can even get further insights from it. One thing to note, the more details you give, the less intelligence will be applied. So what do I mean by that is, probably there is another section which might have better insight about Abbott, right?

But because I told it to go here and look at this, AI is not smartly doing any such thing. But because I'm telling it where to look for what, it's not hallucinating. So there is a trade-off.

This is a trade-off that most organizations are not comfortable with, like how much do we precisely tell it what to do? And when you do tell it precisely what to do, how much of creativity of LLM are we missing to leverage? There's no right answer for this.

The second one, which renewals need me now? So I asked the same thing, like I'm a CSM in this account, and yeah. I told give me all my renewals in the next 250 days, and it popped up all the renewals with all the information I asked it to give.

Now, at this point, I want to highlight one thing. I'm not a very, I don't love reading through text. I get distracted going from one paragraph to other paragraph.

So table is good, table helps me focus, it helps me look at where to go. But that's why for the next prompt, why is this accounted? I asked this one thing that, can you make your response a bit more visual to a typical prompt?

Now, if you interact a lot with your AI agent, sometimes it already takes this note about you. Like your cloud is probably different from my cloud, your LLM is different from somebody else's. So it might already know that you want to look at things visually.

And when I asked it to make things visual, this is what you see in the next prompt. Why is this account red? It is able to tell, hey, overall health score dropped in a very visual manner, and what is dragging it red, right?

Like why did it lose? It's because of this particular section. AI weights, AI sentiments, and AI insights, this is what is dragging the score.

But why? Go further, this is what has happened here. So somebody rescored something, and then some meetings, like hard-nosed vendor review.

This is what has happened, champion transition happened, and so on. And yeah, that's why sentiment is low, and that's why overall score is low. It's easy when I can scan it, because nobody's reading text.

Nobody's reading the whole text today. So visual is better. When you design your agents, please follow these principles if you're creating any of these.

Now, the next question, note, I did not ask it anything about visual. But it went ahead and created something very visual for me. Because it remembered from the previous prompt.

Again, what you call as a bug, something that doesn't work, is feature. It's working exactly as intended. So this is clean up my CTAs, a lot of CSMs are using this today.

Extreme version of it, I wanna give you one example. A CSM had roughly 200-odd CTAs on their name, which they were able to close and start fresh. I hope that doesn't happen to you all, anyone in this room.

But if it does, you can do a clean slate today. Clean up your old CTAs, CTA from last year, and so on. Some alerts, probably, which should be automatically closed.

You could do all of that. You could let AI do all the work for you, yeah. And then the last one, I have a call.

So every CSM, right after the call, wants to take some notes, get some action items, send a response to customers. Now, every CSM wants to do it as soon as possible, probably. But some of them do it at the end of the day.

Everyone has their own variation. But everybody today uses an LLM to do it. Some of them rely on Teams, AI Summary, Zoom AI Summary.

But this is the number one thing we've heard on Agent Studio. I had a call. I just now had a call.

Help me draft an email. Let me bring it up in front of me. If there are any to-do lists in the call, create the right CTA for me.

Store it in the right CTA. If there are some tasks which we closed in the call, strike off those tasks. Like if we discussed about a QBR that got postponed, move the task due date.

If there is an EBR task, update the task. Tell that EBR got postponed because exec was not available. Do all these activities.

The manual updates that people hate doing, do it because all of that was discussed. We just spoke about it. As a human, I'm good at interacting with people.

I've had this interaction. Take this admin work out of my way, and you do it. That is the primary reason why I think we're seeing huge adoption for MCPs across all SaaS apps today.

And yeah, that's pretty much it. As you can see, I gave it a sample transcript. Oh, yeah, sorry.

I don't have a transcript to give. Sorry. This one, I didn't run it.

Now, the last one that I want to tell is the summary, the query, the prompt one that I did here. It's a bit sad that it didn't make it visual. I really hoped it would make it visual.

It may put a bunch of text. I don't like it. Just now, if you calculate something else, do it like an artifact.

It could do an artifact, yeah. So that's what CSMs are doing today. And each one of these, we'll go through them.

When you make it very visual, like when I make each of them very visual, this is how it looks like. Prep me for my call and a very visual state of what the call looks like. You can see some of these things that people are saying.

By the way, the difference between what you saw when I did live versus what you see in the deck is this is Opus versus that is Sonnet. So LLM model drastically changes the output. The second one, what renewals need me now?

And this is a visual there. Why is an account red? You can see what's dragging an account red and how do I push it forward.

Clean up my CTS, which CTS don't need me anymore, which are outdated. A visual step, like 26 can reduce to 18 after this exercise. And by the way, close the seven with one click.

Click it or pick individually the CTS you want to close. Right? And turn my call into Gainsight updates, right?

You can log a note about it and you can probably, if you have a CTA, you can update it accordingly. The only difference is the cost of tokens and what you see in the deck versus what you see me run. That's pretty much it.

Any questions on this prompts or anything?

[Speaker 6]
Yeah, I've made a lot of skills and prompts myself, but one of the things I'm always thinking at top of mind is when we share it with other people. And my question is, how do you make your skills and prompts as token efficient as possible, but still give you the right amount of data? And do you have any sort of tests that you do to sort of measure before and after it and things like that?

[Speaker 1]
If you looked at earlier part of the, like probably 10 minutes ago, I've covered that. The more details you give it, the more token efficient it is. If you can instruct LLM where to go, look for what, it doesn't go around scanning the entire Gainsight account data.

So it knows exactly where to go and it knows exactly what to look for. But because it is token efficient, because it's exactly going there, it's not looking at some other data which might have some other insight. So what is the right balance?

I don't know, like honestly that's the toughest to answer. It's to each, like personally I like to think like if I'm doing one of, like if I'm in a very tough call with a very important call or a workshop session with a customer where we're trying to figure out how to do things, I would actually want it to scan for more insights than what I've given it access to. Because it's a creative task, like I probably don't know what I don't know.

I cannot instruct it. But if there is a signal in some customer conversations, I want it to go, ask staircase in 20 different formats, ask 30 different questions, like all variations of the questions and give me insights. But if I am in a regular task to do, like regular day-to-day things, I want it to be very standardized.

Now for the admins in the room, I have one last thing to call out. So you've seen the QR code is for Pulse Workshop, it's here. But we also have an AI cookbook that we are showcasing and what the AI cookbook has is all the prompts that you can share, some of the skills, the same as our library, like whatever is in library, once we test it, if it is working well, you get it added here, all the skills, prompts that you've seen me speak from our library.

But also, admin CLI starter kit for CSMs and admin CLI advanced kit, sorry, for admins, for our CS admins. Admin CLI advanced kit for admins. I want to spend a minute, because of the audience in this room, to talk about admin CLI as well.

Now, admin CLI, if you've spoken about, CLI typically is a very developer-oriented thing. It's a package that you install, terminal, and so on. I haven't used terminal in probably six years, five, six years.

But with this module, I was very excited about it because for the first time, there were problems, like, I pretty much tried to solve problems for CSOps, GainSight admins, almost all my GainSight career, product career. How do I make it as efficient, as easy as possible for you to do amazing things with GainSight? That's the key problem I wanted to solve.

But one thing that I always fell short of is explaining the processes, converting that from GainSight configuration. It wasn't possible for me to solve it. And that's where LLM changed the game.

And that's what I'm very, very excited about for CLI. So we give these two kits, a starter kit and an advanced kit, as a starting point for you to experience. What it does, respectively, is it creates a local brain of your tenant, so that you can ask questions in natural language, like what you saw in the stage today.

Do we have any executive outreaches? Do we have any journey programs or journey orchestrator programs, or are we reaching out to execs any form of fashion? Like, let's say you're a new admin, you're inheriting a tenant.

You could ask that question. And it will scan through all the instance and give you the answer. And because it has a local brain of your instance, the to and fro is very, very token efficient.

So that's what we tried to do when we created this advanced kit or admin kit. And thanks to our customers who helped us figure out that just giving CLI is not enough. Some of them have spent a bunch of tokens asking questions and not getting good answers.

Together, working with them, we were able to create the best possible way for you to be most efficient, and that's what you see in our GitHub tool, right? Like the admin advanced kit, CS admin advanced kit. And yeah, I have live examples of some admin activities that you can do here today.

I want to tell one example. Let's say you have journey orchestrator programs, tons of email templates. As an admin, if a CSM ever asks you a question, why did my customer get this email?

How would you ever answer that? You need to go to the inside, look at what is the template, which template does it look like? And imagine if you're an enterprise customer, you have hundreds of templates.

That's probably half your day to directly answer that question. But with MCP or with CLI, you could just throw it in Cloud, and it will exactly match which program did it, and why was that customer selected. It can go to the next step and help you answer that question in under five minutes.

That's the key value of CLI. Like it can look at the text, string matches, and all of this. The best part is I don't need to create, build that feature.

It's available now in Cloud. Any questions on admin CLI? So I also want to share this.

[Speaker 7]
Yes, users, and then we're also doing like, we're also cleaning up accounts in our production that are inactive, because I know we're also paying accounts that are active. Can it also do those prompts in what you said earlier?

[Speaker 1]
Yes. So we have 180 tools, 180-odd tools from all across our modules that we've exposed as part of CLI. Data designer, rules, it can read only, get any object, any data.

Stale connector jobs, stale rules, you can clean some of those up. Things that are failing, you can debug. Like all these activities can be done here.

[Speaker 8]
That's nice, thank you.

[Speaker 1]
This is an example. I've connected CLI to one of the instance in my local machine. I love this view, it looks so pro, but please don't be intimidated if you've never used Terminal, it just looks good.

Cloud has built a very good, it's so easy to use, relatively.

[Speaker 4]
Excuse me, I have a question, but not on the admin side, but on the prompt that you've shared. So I'm a CSF myself, and I have around 10 to 15 customers. So do you suggest that instead of creating a new prompt for every customer and whenever I get an email, will it be better if I create a skill or a project wherein, for example, a QBR project, and any time I have a QBR coming up, I would just ask Cloud or any other LLM to help me prepare.

What do you suggest? Which is token efficient in that way.

[Speaker 1]
Yeah, absolutely. Having a pattern prompt, either a skill or a prompt, which effectively, so here the distinction is standardization across your team versus what you want to do. If there is any part of standardization in your team on how should a QBR presentation or how should this look like with my customers, within Gainsight, let's say it's a skill.

Our own DevOps team has a small unit which works with CSMs and with product to identify what is the right skill to help solve the QBR problem. That's the standardization. However, if there is a variation that one CSM likes, I'm sure like Rishabh here or Tatiana there, they might prefer their own version of it as they speak with their own clients, and you can always add that variation.

[Speaker 5]
Can I ask a question? When you were talking about, for example, in one of the prompts, trying to see where some specific field was used, one of the main challenges that we have on that end, well, two, one, if I need to retire, it will be able to remove it from all the dependencies without manual intervention?

[Speaker 1]
Lot of them, yes. All, I don't know yet.

[Speaker 5]
Okay.

[Speaker 1]
But I can tell you, across all reports, across all dashboards, it is very efficient for an LLM to remove it than for a human to remove it. I'm talking about 1,000 times more efficient or something like that.

[Speaker 5]
And the second one, it would be able to swap it, for example.

[Speaker 1]
Yes, yeah, that's the idea.

[Speaker 5]
Yeah, okay, I'm done.

[Speaker 1]
So, you tell a very real problem. The problem that I hear is, my Salesforce admin team changed the segment field. So that's the test for us.

When Salesforce admin team introduced a new segment field, Gain side admin team had to work for two days to clean that up. No, now it's five minutes' work. I think, yes, again, these are scenarios for me to test to say yes.

But yes, that's the spirit of the enhancement. It should be possible. You should be able to replace it across 10 of your reports in one shot.

By the way, to be fair, over time, these were all the enhancements that we spoke about. These were all the feature requests that I got four years ago, five years ago. I'm inheriting CS tenant.

What is running in my tenant? We had no way to solve it back then. But with intelligence, suddenly a lot of this is solvable.

With AI LLM in the loop, things are not so deterministic anymore. So that presents a new opportunity to solve all these problems.

[Speaker 2]
So I think I have an answer for the templates question and the skills we have and the other CSMs question about can I use the project and all that stuff. So we are an internal team to GainSight. And then we are also using a lot of this.

So yesterday, in the same room, we had a workshop where we are talking about the reference, the standard, and all that stuff. So if you're a CSM and if you have a presentation style, like you jumping into a call, you go with a presentation. So you can give five to six references of your previously successful calls where you had a great sentiment, reference it, and you can build a skill out of it saying that I need X, Y, Z fields to be templatized.

And then create an artifact. And when you create an artifact, think about artifact as an app. It's very interactive, it can be updated.

So then you build a skill which can query the right objects and that skill can handle that. So you are giving a brain and you are giving an interface. So now to the artifact, if you add a refresh button on the top, so jumping onto your call, you have the template ready.

You are already having a skill which can query more precisely than a prompt. So you don't have to remember all your prompts. You can just refer to the skill saying that use this skill and fill up this presentation or this artifact.

This is very less tokens as well. Like any one of your team from the same project can click on the refresh and it gets all the data populated as you need it. And that is shareable as well.

The fun fact is like if it's an artifact, you can share it as it's a visual piece. It helps your team to get a better understanding. So internally that is what working for my team.

So when I wanted to share any DCS reports with my leaders or my team who wanted to look at it, they will click on the refresh button. So which is like less tokens. So if you have a project, if you have skills, and if you have a template or an artifact, I think that is the right way to think about it.

[Speaker 1]
All right, so one last thing. Probably this is all entire SAS industry learning as well. So traditionally in B2B SAS, the product's definition is what you see within our own UI.

But when you embrace the approach of headless, as an example, let's say, if you look at Dreamforce, anyone, if you remember that, lot of what Salesforce demoed, it's not necessarily new. It's the same features they had for a while, but in a UI where people are. The key, the entire feature there is functionality is it's solving the problem in a way that is more adaptable and more usable than it's solving a problem that has never been solved before.

So the spirit of headless changes how you think of what a B2B product is. And one such thing that I want to share is, yeah, maybe what we call as CSM adoption agent and that we introduced today morning in the keynote. What you see over here is, let me see if I can zoom it.

Like, it's a plugin that my team developed which extends the definition of Gainsight product into your LLM. It works with Cloud and it works with ChatGPT as well. So imagine you are a CSM, you're having all these conversations with your customers and then as an organization, you want to drive adoption for something.

I'll give a meta example. Gainsight, I launched MCP, but then I was working to figure out how do I improve adoption of MCP. I have my scale team here.

We wanted to reach out to scale team. Now, scale, the interaction is like, we don't have a named CSM, nobody can push it, nobody's meeting these customers day in, day out. So what we ran is the same agent which looked at an individual account level, right?

Like, how are people using MCP? Did we have any conversation about MCP to them? And depending upon where they are, like, for example, if we've never interacted with them about MCP, it gave a, you might have all, like, if you are in part of this segment for Gainsight, you might have received an email saying that, hey, these are the amazing things that people are doing with MCP.

If you are, like, if you've had an interaction where there is security team which told, we need all these controls, we need logging, we need two-level controls, we need object-level controls. When we released those features in July, you might have already gotten an email. All of that were generated by an agent.

But if you are a CSM, you can do that within your book of business, right from within Cloud, and that's what I'm showing you. Like, it's a plugin you install as a CSM, run to drive adoption for any feature that you want, like MCP, for example, or any feature that you want. And across your book, it identifies which of the customers can you send an email today, and it literally drafts the right email for you to send, right outreach for you to send.

And how does it work? Let me take you to the setup part. You need to give it, sorry, you need to connect your Gainsight instance and Cloud, of course, and you need to give your tone.

How do you prefer to communicate? Some sample emails. If you want product knowledge, where does the product knowledge live?

And some safety rules, when should you not communicate? Like, do not communicate to customers who don't converse in English, right? Only English emails, only within these countries.

Create all the safety rules, and it creates, at the end of it, drafts for CSM to just approve the Gmail draft, and it comes in your email. Like, you can right away send it from here, too, if you want. All right.

So, the definition of what is product, it extends now beyond the code that engineering team owns, and all of this is done by product team, let's say, not engineering team, per se. Yeah, that's it. Thank you, everyone.

Thank you. Thank you.