From Workflows to Judgement: Designing AI-Native Customer Success

Julia Nimchinski:
And we are moving on to our next session. Welcome to the show, Tash, and… Sam, how are you doing? What’s the latest and greatest?

Natasha Evans:
Thanks so much for having us, and what an act to follow. That was, that was super interesting. I’m sure Sam, as a founding AE, is sitting there absolutely thrilled by the prospect of having his quota put up to 2 million, so thanks for that. We’ll report back to the sales leader. But yeah, we’re excited to be here. Thanks for having us.

Julia Nimchinski:
Cool. Well, before we just dive in, I’m just curious, what sparked your, I don’t know, just inspiration in terms of just transitioning to this judgment-based CS?

Natasha Evans:
I think we’re, we’re… we’re trying to approach things a little bit differently here, and hopefully you’ll… you’ll learn a little bit about this as we go through this… this presentation. when folks think about AI, we think about automation, we think about, like, lack of personalization. We think about making linear journeys easier and more scalable, and so we’re trying to approach the problem and the goal differently, and think about what makes a CSM great? Well, it’s the fact that you can review the data, you can apply a judgment, you can adapt based on the scenario, and so we’re starting to think about how do we leverage AI to do that, and what’s the impact of that when you scale it? And that’s a bit about what Sam and I are here to talk about today.

Julia Nimchinski:
Love it. Let’s dive in.

Natasha Evans:
Nice. Well, hi everyone. If I haven’t seen you on one of these before, I’m Tash, I lead customer growth at Hook, and Sam is a founding account executive at Hook. We leverage AI to help CSMs gather insight on their book of business, whether that’s to find risk or expansion opportunities, then make the best plan to go execute on that, and execute it in as an efficient way as we possibly can. And the goal here is either free up CSMs to focus where it really matters, or allow CS leaders to completely automate parts of their CS business. Either which way, we want to help scale customer success, and that’s what we’re here to Talk about today. How do we do that by really putting AI at the core of our customer success systems? We’re going to talk a little bit about where the old system is just breaking as we try and scale CS, and how, like, rethinking the problem, as I’ve mentioned already, and redesigning these systems, but with AI at the core, yes, can increase efficiency, but can increase results as well. And so that’s what we’re here to talk about today. So, what I’m going to do first is just ground us for a few minutes in the problem statement and what we’re trying to achieve, and then I’m going to pass over to Sam to bring it to life. Sam, if you, person is running slides for me, if you click on, we’re all focused on scaling customer success, but if we are just focused on scaling our current practices with AI automations, really what we’re doing is we’re scaling the effort. But we’re not really impacting the outcomes. And I think so many companies have ended up layering AI into component parts of their process, or just automating broken practices or static linear journeys with AI to help scale. It just isn’t working. And so what we’re trying to do is think bigger picture. Let’s stop just trying to scale these old, broken processes, and let’s just start again, and think about it differently, but this time with AI at the very core. And so, what’s the goal here? Well, if we’re to think about it, what we really want is for AI and for agents to help us mirror the experience, the process of a dedicated CSM, except across every single customer we have. And not just any CSM, you want it to mirror your best CSM. And so, what do I mean by that? Well, I want AI and I want agents to understand the context of the customer. I want it to know the outcomes that we’re trying to achieve and that that customer cares about. And I want it to suggest relevant next steps and actions and strategies for that customer based on all of that information. And I want it to assess how it’s tracking along the way, and adapt and change its plan, but ultimately drive the same amount of value and the same outcomes to a customer that a customer with a dedicated CSM would get, except this time it’s at scale, and it’s across every single customer. And so this is really where we’re focused on AI agents, and where AI agents really come into their own. But for us at Hook, an agent isn’t just something that’s, like, layered into part of the workflow, or just writing one-off emails, or part of your process. For us, an agent owns the end outcome and the whole process to get there, and that’s really important with how we’re thinking about, AI at the core and AI-native systems. And so what we’re thinking about is, well, we want to give the agent a goal. What’s the outcome it’s got to get to? We, of course, want to give it some parameters that it’s allowed to operate in. It can’t just go offering discounts to every customer left, right, and center. But then we want it to find the best next path. We want it to evaluate what needs to change, what do we need to do differently every step of the way, just like your best CSM would, except we can use that really smartly to run that process across hundreds or thousands of customers all at the same time. And so instead of just automating this same linear process through AI, we’re rethinking it completely differently with AI built at the center. And this is exactly what Sam’s, gonna walk you through, and talk you through a bit about how we’re bringing that to life today. Both fun.

Sam Champion:
Yeah, and so the reason we’ve gone down this path is because, what we realized when looking at the market was everyone today is trying to solve the whole customer journey problem using rule-based orchestration, and this doesn’t work for a few reasons. It leads to exponential complexity, customer behavior is complex, and so what ends up happening Is you’ll try and build the world’s most dedicated bespoke onboarding flow, but Anytime anything happens that’s not in the script, the whole thing falls apart. Like, you might set up a day one activate SSO email, a day two something else email, and if someone replies, saying, how do I actually do this? A human has to step in, do the research, reply to the ticket, submit it, etc. And obviously, if you try and get around this by just creating a really complex flow, you’ll quickly realize it’s not really possible, because after just 15 steps, you’ve got as many as 16,000 potential nodes to deal with. And so what we’ve realized from looking at this is that AI is already much more effective than rule-based logic at analyzing customer behavior, analyzing sentiment, suggesting next action, and we already have AI agents in Hook that do this ad hoc, listening to calls, tickets, emails, usage data to flag risk or opportunity. And so what we’re now also doing, which is what we’re going to be focusing the demo on today, is using this same type of premise for full customer lifecycle management. So, we now have an agent called Activator we’re going to run through, and the basic premise is that we can give these agents goals to work towards, such as, hey, you’re an onboarding agent, here’s the customer journey, here’s some resources to use to guide people along the way. the agent’s gonna abstract that down to basically goals that it needs to work towards, such as, in this example here, book a kickoff call, set up SSO, process a payment, and maybe eventually a renewal. The agent’s then gonna have those in its head. And unlike the traditional world, it’s basically gonna work through a task list. But that actually adapts to customer behavior. So, day one, send a kickoff email. But because we have access to things like meetings, we can see when someone books in a kickoff, and we’re not going to send two bumps to try and get them to book in that call. Once we see them do that, we’re gonna skip straight to the next step, which is, hey, now you need to activate SSO, and once we see we’ve done that, hey, here’s how you need to process a payment. Maybe if they respond to that. saying, I don’t understand how to do this, AI can step in, draft a great response for your review, and then continue down the pathway again. And so this is what we’re really going to be running through today, is how can we centralize data, make sense of it all, and then train agents on this data to be able to guide customers through journeys that actually react in real time to what the customer needs from you. And so, we’re gonna jump in. So, at HERC, Our agents basically do three key things, and they really mirror what a rep does, which is, first of all, they look for intelligence. So, is this customer new? Have they signed up? Are they using the product? Are they at risk? The next thing is to create a plan. So, okay, if this customer’s at risk, what can I do to solve this? And the next tool that the agents have access to is the ability to take action. So, at scale, send an email, you know, in enterprise, maybe more like book a meeting. The way we train agents to be able to deliver on these three key things is we pull all of a customer’s data into one place. This is, like, emails, call transcripts, tickets, usage data, revenue. And we do different types of AI on this to get different types of insight. For the numeric data, like revenue, usage data, metrics, LLMs basically suck. They’re really, really bad at getting insight from numbers, they’re not really trained properly on numbers. And so we use machine learning to get insight from this that we can then give as context to the LLMs. We use historic analysis of churn or upsell to basically work out what causes churn or upsell in your customer base. And then we listen to all of the other metrics and customer voice sort of aspects with normal AI to basically start building a context picture for your customer base and what indicates a risky or ready-to-renew customer. These things together form what we call ECHO, which is more of an ad hoc risk and upsell detection agent that also forms, like, basically a context setter to tell other agents what all of this data means when it comes to, is this customer happy? We’ll also go through a few other things, like Hook Chat, just to get deeper on this data, but the other thing that our agents have access to is what we call play iBooks, which is each signal that Hook detects, such as this customer’s at risk, or this customer’s falling behind in their onboarding, we create what we call a play iBook. It’s kind of like a playbook, but it’s generated by AI with full context of the account. Creates a next step plan, creates a risk mitigation plan, drafts customer-facing content. And then we give you many channels with which to either approve these or tweak them before you send it off yourself. So the idea is, if you’re a CSM, you’ve got tons of tools at your disposal to manage customers more effectively. Or if you’re an AI agent, like Activator that we’re going to run through today, you can also use these tools to detect customer needs, draft appropriate content, and send it out to manage a customer through an onboarding journey. And so this is what we’re going to focus in on the demo, is how can we bring all the data into one place, and then give AI some context about what that data’s telling you about your customers, and then again, layer an AI agent on top that can basically orchestrate all of this going forwards. So, this is hooked. Let’s say you’re logging in to look at your customer, Meta. They’ve got two products today, A and B. We’re going to use a single account just to talk through some of the mechanics of how Hook works, so how we give the AI agent some context. about your customers. how we use AI to then flag risks, to give the AI some more context about what a good or bad customer looks like. And then we’ll finish it off by jumping into settings and showing you some of the behind-the-curtain aspects of Hook, which is how we can then train agents to look at all of that data. Detect customers that are falling behind. And then queue up really concise, accurate actions for humans to review. So you could imagine a world where a human comes in, they see their action list at the start of the day, they can see all their actions created by Activator. And they can see, for example, here, I’ve got a customer that’s falling behind, AI’s summarized why they’re falling behind, it’s created a contract review term email, because it thinks that’s the right course of action, I can send some final tweaks, and then I can hit send. So that’s what we’re going to cover. Single account first, and then show how we can apply that insight across the book of business. So, to effectively orchestrate this account with AI, we need to know everything about it. So, we start by pulling in all of the key data about this account. So, this is basically the context we’re going to give the AI agents about how this account’s doing. We tend to do this at the product level, because it tends to make most sense, so if we wanted to see how Meta’s using product A, We’ve got some metrics here around integration errors, trending over the past 12 months, for example. Maybe logins. Have they set up yet? Have they met the CSM? Have they resolved many tickets? This is sort of the numeric context behind the account. But then to give the AI agent more, like, qualitative context, we also bring in all of the, sort of, voice of the customer data. So this is, like, meeting transcripts. Emails back and forth, maybe support ticket content, notes, etc. This forms the dataset that we then analyze with AI to detect when customers are at risk or ready for renewal. Such as, for example, one example is our echo agent, that for product A here is detected kind of like a risk, so we can see there’s a point of contact change. For each signal that Hook detects, such as in this case, there’s a risk within this account, because we can see Echo’s told us, senior stakeholders report data sync issues, this has led to some declining usage, and then separately, their VP Ops said that they’re leaving, and the AI’s smart enough to realize that was your main champion. this is the sort of first thing that we’re going to cover, which is how can Hook help you gather intel? So we flag when this data’s telling you a customer needs attention. We tell you where that customer needs the attention from, so we can see this usage metric, we’ve cited the source of, this call recording, this email. This is obviously a risk. But we can also generate, sort of, upsell context. So here we can see they’ve acquired a new company, they’re complaining about site errors, which is a great fit for some sort of glitch monitoring product that we could go and sell them, which again, we’ve seen here. For each signal, though. the next step, if you remember, is we need to create a plan. This is why this is really hard to do with rule-based automation, because you could imagine, like, there’s, I don’t know, 50,000 different potential signals that could exist across your customer base. You’re never going to be able to build a playbook for each one, and so we use play iBooks to orchestrate the, Actioning of our signals. A play iBook basically scans context about your business, if you remember here, to basically work out what would a great next step be relative to what we know about this customer and what we know about your business. So in this case, pretty complex risk. There’s actually four steps we need to take. AI has generated this with full context about the account. It’s a one-off, so this is dedicated for this particular customer, for this particular signal, so it’s much more accurate and effective than a rule-based sort of automation. So in this case, something quite simple, we need to send a new intro to the new stakeholder, we’ll generate this email with full context about what the account’s doing, what next steps would be great. Obviously, for the upsell signal, this is going to be more like… kind of acting like an AI SDR. So, reaching out about this expansion play, or trying to book them in for a discovery call. So, at this point, we’ve got some basic insight into our customers. We’ve generated some action that we can take. So, in this case, emails, meeting invites, we can also do slacks or in-app prompts. The next step is we want to actually take this action, and then also have Activator come in and con, what do you call it, like, orchestrate some of this. So, all of these actions are queued up for human approval over here. So we’re quite big on, sort of, human in the loop. We want AI to make humans more effective, and so you can imagine a world where someone can jump in, see all the risks detected by Echo. I can see here there’s a usage decline, I can see some actions to take drafted by AI, I can sense-check it against all the other data that I’ve got right here in this account, make sure it’s the right thing, and then hit send. This is how we deal with orchestration of, sort of, ad hoc risks or upsell signals. We detect them, we action them, queue them up here for human approval. For something a bit more deterministic, like an onboarding journey. This is where we’ve just released Activator. That has access to all of the same data, and all of the same tools that we saw before. Except in this instance, it’s got goals to work towards that involve maybe many steps, rather than the goal of just keep this customer healthy, which is what Echo’s working towards. The way Activator works is… this is obviously a demo, but we give it context about what its goals are, so it’s an onboarding agent. Maybe in the first few days, we need to just get them to set up their account, activate SSO, After that, we need to integrate some data, and then we need to start giving them some value, so viewing dashboards, generating reports, whatever your customer journey is. Crucially, we need to give agents context and resources to use, because agents are really only as good as the tools that they have at their disposal. This is overly simplistic in a demo, but many of our customers will link, like, many a help doc resource, so the agent knows what it needs to call on when customers can struggle with certain things. Many customers will link, for example, like. Templates that it can use going forwards. We then give it tools to have access to, like being able to create actions, Hit save. The hook team works with our customers to do these quite bespoke, but we’re showing you kind of how the backend works here. And then this agent will be able to see customers that fall within its purview. It will be able to detect, for example, this customer has just signed up. We need to start putting them through an onboarding journey. And then I can scroll and see all of the actions created by the onboarding guide, such as this one here, for example, that maybe is preparing for an upcoming platform migration. AI’s given you some really detailed insight about what’s going on in their implementation, what they’re expecting in the first 6 months, what integrations need going. As well as a bit more detail here, if you’d like it. maybe the step that it’s currently on is scheduling a business review to check in on how the implementation’s going. We can sense check that, hit send. But obviously, each step is only dependent on how the customer is working through its, steps. And so this is how AI becomes really, really impactful at orchestrating go-to-market, because we can jump in here, maybe it’s sending the kickoff call email. If they then reply to the setup SSO email, being like, I have no idea how to do this. Hook’s gonna then create a great AI agent reply, call on the context it knows about your business, which tells it how to set up SSO, for example, draft a great reply. queue it up for your approval, hit send, and then assuming the customer’s happy, or maybe we can see that they’ve activated SSO, the agent’s gonna move on to the next step. And so the idea here is we’re giving every customer a high touch using AI, because as everyone knows, enterprise customers, they tend to renew much better, they tend to expand much more. Some of this is firmographics, but a lot of it is because you can afford to put loads of reps on the account, you can afford to give them loads of time, and so we want to be able to give all of the customers that you might have that same amount of attention. Which is why we’re using AI to get most of the legwork away, and make them feel like they’re getting a high touch, because we’ve got agents sort of crunching the numbers in the back end, and guiding a customer along a journey. This is obviously the first step. Very soon in Hook, we’re gonna be thinking about the ability to make custom agents for pretty much any scenario you can think of, so give them goals, work towards it, because, in our opinion, this is really what defines an agent, which is that they have goals they can work towards. No matter what curveballs come their way. Otherwise, really, it’s just a curve… just more like a co-pilot, like a hook chat, for example. Which is pulling great data, can give you great answers about your customer base, but it’s not really working towards any other goal than answering a single question. If you throw a curveball at it, like, I don’t know, the data doesn’t call properly, it won’t be able to give you a good answer. Whereas agents that have goals to work towards can be kind of… They have ingenuity to work towards them, and they can sort of, go to the next level to try and get this customer to the different steps in the onboarding journey without just sort of falling over at the first hurdle any time anyone, you know, asks a question, for example. Lots of other stuff in the platform, but this is what we really wanted to focus in on today, which is the intelligent orchestration of the customer journey. Tash, anything you think you’d like to jump in and just talk through?

Natasha Evans:
No, I think you’ve, you’ve nailed it, Sam. So, we’ve given you a whistle-stop tour of a bit about how we… how we think about using AI, and how we’re trying to solve the problem of scale, but also impacting on results. And thinking differently about giving the agent a goal and letting it find its way to the outcomes and adapting along the way. Julia, I think we have time for some questions, if there’s any questions.

Julia Nimchinski:
Yeah, I have so many questions for you, but the first one, I love the orientation, you know, towards a goal and outcome. How does, what’s your philosophy in terms of pricing? Do you reflect it in the pricing model?

Natasha Evans:
Such a good question, oh my gosh. We… we’re… we’re pricing right now, specifically for some of the agents, the same way that a lot of different tech companies are pricing. We think about the value that the platform brings, the implementation, and so we… we have a fee for that, but then we’re… we’re pricing based on, credits for AI and the actions that the AI is doing on behalf of your team, or queuing up for your team. We’ve thought a bit about this outcome-based pricing. I think it’s really hard to quantify, right? And an outcome looks really different to every different organization, especially when we’re saying to every customer, you can define your own outcome. And so, perhaps we’ll move towards that model as we explore a bit more about what different outcomes mean to different customers and how our tech helps get them there. But for right now, we’re pricing really similarly to how other tech companies do with AI.

Julia Nimchinski:
Gosh, Sam, I saw among your clients in one of the slides, some of the leading SaaS vendors and GTM vendors. And we always see, you know, the success scenarios, what happens when the architecture is built perfectly, the promised land. And obviously, it’s all looks amazing, but I’m curious, in your experience, I mean, we saw this stat, 95% of AI deployments fail, what would it take for this to succeed? What are, you know, the 5% your customers are doing right?

Natasha Evans:
That’s a great quote. Oh, go for it, Sam. You go for it, and I’ll layer in.

Sam Champion:
Yeah, I was gonna say, one of the things that we’re really big on is, not just being like, here’s a platform, go log in, go figure. You know, this is quite a technical platform, we get access to a lot of data, even though the UI looks quite simplistic, there’s a lot going on behind the scenes here. And so, really getting access to all of the right data is key to a hook implementation. And so some of the things that we’ve done with our best customers, and actually, we pretty much roll this out to all customers now. is we make sure to assign two people to their team. So we assign the account manager and a technical implementation lead, and they basically partner with our customers to make sure that as we’re implementing the software, the AI agents that are within Hook are getting access to all of the right data to be able to make the best decisions. because, I don’t know, let’s say we don’t get access to usage data for product B, because it’s, I don’t know, it’s locked off somewhere in a data lake, and someone didn’t have capacity to process it for us, then the agent’s just not going to have any context about this. The churn prediction is not going to be as impactful. And the agent’s not going to be able to orchestrate a customer journey with as much accuracy. And so, for us, success is all about, like, giving it the best data possible, the right context. A lot of people think they have messy data. In our experience, it’s not really been the case. Every customer who we’ve been able to get their data into the platform, AI’s been able to make good sense of it. It’s just a case of identifying which areas of your data we should be looking at, because, I don’t know, let’s say 50% of it’s messy, you’ve still got 50% that AI is going to be able to look at and make some really good judgment calls from.

Natasha Evans:
I think the other thing I would add to that, I’m not surprised by the stat on the volume of AI tech that just doesn’t properly get adopted and does churn, because, as we said at the start, like, so much AI is just, well, let me just do this tiny little part of the process for you. Like, let me write you that one email, or just make that tiny little thing a little bit better, and… it’s then really hard to be sticky as an AI product. It’s really hard to find someone who cares about, like, really driving that in your organization, and ultimately, it’s really hard to prove value, and so I think part of the benefit of us really thinking about an agent owning that full end-to-end process and owning the outcome is it’s much easier for us to talk about value with our customers. It’s much easier to get that buy-in, because we can show how owning that full process is going to get them to an outcome that they really care about. And so then for us, the 5% that do really well is where you have a customer who really cares about driving the outcome, where you understand, like, what are we driving for you, and how do we do it together, and where we can just sit, and to Sam’s point, we pull in the data, we set up the agent to really drive that one outcome. That’s where you get real greatness with AI.

Julia Nimchinski:
Love it. Josh, in terms of, just the team structure. We talked a lot about the architecture of an AI native CSM motion and function, but I’m curious, how do you see the evolution, of CSM in general? Like.

Natasha Evans:
Yeah.

Julia Nimchinski:
Is there any sense to separate sales, marketing, and CS? What’s the future.

Natasha Evans:
well, we need a lot more than 3 minutes to answer that question. I think that where you get AI doing this really well, you’re… the goal is to free up a CSM to do the things that CSMs should be doing. Like, the future of CS is, commercially-led CS teams, CSMs who understand what value means to the customer, who can build really strong plans to get there, who can measure it, and who can turn those measurements of how we’ve delivered on that value back into commercial conversations. Like, that for me, is the future of CS. you can’t do that as a CSM when you’re also thinking about, well, what are these 80 different customers, and what do I need to speak to them all about, and what was the last thing, and what’s the next thing, and how do I interpret the data to make the best plan? Like, it’s not possible. And so this is where layering in super smart tech allows you to elevate your CS team to focus on the more commercials. Whether that means longer term that you maybe don’t need CS and sales, I don’t know, like, we’ll figure that out. I think you’ll always need some element of marketing and sales to close the loop. But it certainly should be allowing us to really elevate what a CSM does, and have them focus on spending more time with the customers that need it, and really driving value into their customers’ organizations, but into their organizations as well.

Julia Nimchinski:
Sam, anything to add?

Sam Champion:
No, Tash is the expert here, definitely.

Natasha Evans:
about that or anything.

Julia Nimchinski:
So, what are you allowed to share in terms of your roadmap for 2026?

Natasha Evans:
Hmm, Sam, well, I’ll let you take that one.

Sam Champion:
Yeah, I think there’s a few things. One thing is, we’ve got some, like, procedural stuff that’s pretty close, like redesigning the homepage, just because, as we release more agents, they need, like, a better place to live, and people need a better place to execute on what they’re advising them to do. But going forward, some of the stuff we’re most excited about is the ability to basically build your own agents, or deploy more agents throughout the platform. Because we’ve got such a great data set in the platform for which to you know, listen to customer needs, action customer needs, and so you can think of a lot more use cases beyond just risk detection or helping with onboarding. You could imagine an agent that helps with renewals, an agent that helps with account management, and Then even kind of an agent that orchestrates those agents, like a managerial way. So you can kind of just keep going with different use cases now that we’ve got this infrastructure in place. So I’d say that’s what I’m most excited about going forwards.

Julia Nimchinski:
Awesome, and last question, where can our community test drive this? Is it even possible?

Natasha Evans:
You can go find out more on the Hook website, so hook.co. From there, you can see information on all of our agents, as well as tons of other resources that we put out there to help CS and to help CSMs be more commercial. So, yeah, hook.co, please go check us out.

Julia Nimchinski:
Thank you so much again.

Natasha Evans:
Thank you.

Sam Champion:
Thank you.