In this episode of the Customer Success Playbook Podcast, hosts Roman Trebon and Kevin Metzger sit down with Dickey Singh, CEO and founder of Cast.app, to discuss the game-changing role of AI agents in customer success. Singh explains how AI agents are designed to take over repetitive tasks across the customer lifecycle, allowing Customer Success Managers (CSMs) to focus on high-value activities like empathy-driven relationship building, problem-solving, and expertise sharing. This shift enables companies to achieve complete account coverage without the need for additional headcount, which translates to millions in potential revenue gains and a transformative impact on the customer experience.
The conversation also highlights the sophisticated technical capabilities of Cast.app’s AI agents, including advanced hallucination prevention mechanisms and real-time confidence scoring, which ensure response accuracy. Secure data handling protocols are also emphasized, with data integrations across platforms like Salesforce, Gainsight, and Snowflake without synchronizing sensitive information. This robust setup has yielded impressive results for companies like Pure Storage, which generated $1.6 million in additional annual revenue, and Route, which achieved engagement rates three times the industry average.
For businesses like HPE, these AI-driven solutions have led to substantial improvements in customer engagement, with ROI increases ranging from 1000-4000%.
Through these case studies, the episode illustrates the powerful value and efficiency AI agents bring to customer success teams, helping them drive revenue and engagement while optimizing resources.
Hi, everyone. Welcome back to the Customer Success Playbook Podcast. I’m Roman Trebon, and I’m here with my co-host, Kevin Metzger. As always, we’d really appreciate it if you could rate, subscribe, and share the show with your network. Kevin, we’re talking about your favorite topic today. AI and specifically AI agents. What would AI agents bring to the table, Kevin, in your opinion?
Yeah, Roman, I think AI agents are really an exciting development as soon as LLM started getting good at conversing on topics. The topic of an agent started to kind of gain relevance. AI agents are looking at tasks that people do, which require both process and decision-making and really provide the ability to automate those tasks and enable AI to help drive them. I’m extremely excited to talk to Dickey Singh at Cast.app about this today, as they’re a leader in the space and really focused on how they’re using AI agents and customer success. Dickey is the CEO and founder of Cast.app, where he’s pioneering the integration of AI agents in customer success with over two decades of experience in Silicon Valley, including roles as CTO and VP of products at Customer Sat and leadership positions at several venture-backed companies. Dickey has a unique perspective on how AI agents can augment human customer success managers. His current work at Cast.app focuses on helping B2B companies grow and preserve revenue at AI scale, with proven results including generating $1.6 million in annual revenue for Hewlett Packard through AI-driven customer success initiatives.
In today’s episode, we’re exploring the transformative role of AI agents in customer success and how they can amplify the effectiveness of human CSMs. We’ll dive deep into practical strategies for implementing AI agents to handle routine tasks, analyze customer data, and identify opportunities for expansion, while enabling CSMs to focus on high-value strategic activities. Dickey will share insights on how organizations can successfully integrate AI agents into their customer success operations, measure their impact, and maintain the crucial balance between automation and human touch in customer relationships.
Dickey, welcome to the show. Let’s get started with the basics. What specific tasks and responsibilities can AI agents effectively take on to support CSMs in their daily work?
First of all, thank you, Kevin and Roman. Thank you for having me.
AI agents are a very exciting topic. As you remember, Kevin, we geeked out on it quite a bit the other day.
Yes. To answer your question, what can AI agents do? Every mundane, repetitive, and follow-up task. How about every task that bogs down a CSM today? You know, inefficiency is killing the CSM. And there are several examples, like, they have to manually execute playbooks, they have to triage, and they have to prepare and present content.
So, imagine a CSM jumping between three tools to understand what the customer purchased, how the product onboarding is going, usage, adoption, churn, renewals, building a deck, chasing down executives for calendar availability, praying the executives show up, delivering the presentation, and answering the same question again and again. Then emailing the presentation to people who did not show up, which happens a lot.
Let me paint a different picture for the leaders.
Now, imagine you could scale to 100 percent of your accounts without adding a single headcount in customer success, account management, onboarding, renewals, sentiment analysis, and churn mitigation.
And obviously, I don’t mean the “do more with less” mantra that is well-known to worsen customer and CSM satisfaction, growth, and churn. Let’s think much bigger. Not only engage and influence 100 percent of the accounts, but also engage and influence every user and every decision-maker, wherever they are, without wasting a minute of their time. Again, without adding a single headcount.
How? By placing AI agents in the middle of your business and your customers, backed by skilled, lean, and focused teams on the side, you can not only scale to every user, exec, and account, but also improve the team’s overall satisfaction.
So Dickey, you had me at hello there.
I love what you’re talking about here.
And I will say that when Kevin said that we were having you on the show, he was speaking very highly of the solution and how it works. And, you know, it’s, I think, a real game-changer, Kevin, not to put words in your mouth, but you were super impressed by it. Dickey, I’m curious, for a company that doesn’t have these AI agents in place today, how would they go about doing this? It sounds like there’s an operational change, maybe even an organizational change, needed to support this. How do you actually get there tactically?
It depends who you ask, right?
If you ask a product person or engineering person, everyone wants to build the AI agent themselves, right?
If you remember a few years ago, people would say, “Oh, we do this AI,” and everyone else would go, “Oh no, we’re doing AI too,” right? Without quite understanding, and it’s the same thing as 20 years ago, saying, “Oh, we do SaaS as well.”
Things have changed. There will be an AI agent in every operating system, every product, every website, every portfolio product.
We used to say, “There’s an app for that.” Now we’re going to say, “There’s an AI agent for that.” So that’s what I feel, and agents will do different things. Like how we are embedding an AI agent within the presentation so the AI agent presents and answers questions.
Similarly, there will be and are agents for SDR roles, for customer support roles, and for AE roles. These agents are going to help people do a lot of mundane tasks that require manual work today.
Following up on Roman’s question, with the agents as they get propagated, are you building agents that coordinate, or are you going to build specific agents for specific tasks? How do the agents get trained to know what to do?
So what we do is we have AI agents for customer-facing teams, right? Which, in my opinion, is the 4 out of the 5 S’s. Support being the 1st S, which we don’t focus on, but then we look at services like onboarding services, technical services, and expansion sales. Obviously, success is there, but there’s another often overlooked one, which is automated sentiment analysis. For example, if somebody gives you a score of nine or 10 on an NPS score, or a seven on a customer effort score, or a seven on onboarding effort score, you want to ask for a referral immediately—in the same breath. So we are able to do that. An AI agent across the board means you don’t need separate agents for renewals, onboarding, upsells, cross-sells, and churn mitigation. You need just one AI agent that covers the lifecycle of an existing customer at the user level.
And I was on the website earlier checking it out—great website, by the way. I was reading through the customer testimonials, and there’s a lot. So I’m curious, from your experience, for companies that are just starting on their AI journey and maybe looking to implement AI agents across their customer-facing operations, what are some best practices or things they should be considering? And maybe some common pitfalls to avoid?
There are so many to list. Every product engineering team wants to build their own agent. It takes a few hours to get an agent up and running using these tools, but it takes a long time to perfect it. That’s where the real challenges lie—you have to handle hallucinations. For instance, our AI agent had someone ask, “Who’s Morgan Freeman?” and it tried to answer. We had to train it not to answer these types of questions and built many mechanisms to manage these challenges. For example, we detect such questions and connect users to CSMs, account managers, or onboarding specialists.
Another thing is that not every dataset is available as a vectorized dataset. We work with enterprise customers, large and small. For instance, with Hewlett Packard, we have 17 different programs. Each program uses several data sources—some use Gainsight, Salesforce, and Snowflake, while others use Totango, HubSpot, and more. Not all data is readily available as a RAG-based system for real-time use, so we internalize a lot of knowledge in our AI system. Of course, we can use RAG-based systems when available, but much of the data resides in systems like Snowflake and Databricks. Our AI can write queries across Snowflake, Gainsight, Totango, and Salesforce reports automatically, much like AgentForce is doing, covering multiple products.
In short, it’s easy to get started with an AI agent, but perfecting it and deploying it effectively in the field takes time and refinement.
So, Dickey, you mentioned AgentForce and what Salesforce is doing—that’s specific to Salesforce. Cast.app actually sits outside of any specific application, pulling data from all applications to act on data wherever it exists. How do you ensure that data propagates properly in this setup? For example, if you have an action that moves a case from one state to another, how do you ensure that’s properly reflected and all data flows both in and out?
You’re touching on a mode that we have where we taught our systems how to unlearn. Case status, for example, can change from “new” to “working on it” to “escalated” to “closed.” Similarly, a user’s journey can change—for instance, they might start as a late-stage prospect, move to onboarding, then begin to use the product and increase adoption, eventually becoming an expert. Then, you may ask for a referral since they’re now experienced users.
So, AI must continuously learn and unlearn the state of each person, not just the account. The best way to do enterprise-grade onboarding is to separate account onboarding from user onboarding. Users come and go, they hire new executives, or new users join years after the product was deployed. You have to onboard each new user individually. Onboarding cannot be done solely at the account level if you want to be accurate and efficient.
Dickey, I’m curious—do you have to do a lot of client education to overcome fears around implementing these AI agents across their customer lifecycle? I hear what you’re talking about, Kevin, about handling hallucinations and training. How does a client feel confident that the AI is performing as anticipated? How much overhead is needed to ensure the AI is performing the necessary tasks, and when do you reach a comfort level where you can say, “The AI agent is good to go”?
That’s a deep question, and it could take an entire session to answer fully, but I’ll try to be brief. So, how do we detect hallucinations? We calculate an on-the-fly confidence level for the answer generated. If the confidence level is between 0.7 and 0.9, we might respond, “I think the answer is this, but you should cross-validate it.” If it’s below 0.7, we say, “I’m so sorry, I don’t know the answer to this.” We then connect them with a CSM, account manager, or onboarding specialist depending on the customer segment. For high-touch customers, we might offer a CSM’s calendar link. For low-touch customers, we generate a form on the fly. Essentially, the AI handles it, but I’m taking credit for it here!
I like that. If AI agents are taking on these tasks, what is the role of the CSM in this world? How does a CSM know what the AI has done most recently, and does the CSM’s skill level need to increase since some lower-level tasks are handled by the AI?
Exactly. We are handling repetitive and mundane tasks—the ones CSMs typically don’t want to do, like follow-up emails or scheduling meetings every month. Who wants to re-present information to an executive who missed the meeting? But there are four things CSMs excel at that AI will never match. First, empathizing with customers. AI can’t empathize. Second, building customer relationships. Third, AI can only solve previously encountered problems, but CSMs handle new challenges that AI hasn’t encountered. Finally, CSMs offer expertise to customers. If they focus on these four areas, the AI can handle the rest, allowing CSMs to focus on impactful, strategic work.
Dickey, I mentioned exploring your website and checking out all the success stories. Could you share some of your favorite success stories on how customers are getting value from the solution?
Sure, it varies by business, but let’s take a customer like Hewlett Packard. Hewlett Packard uses our product to communicate with their customers—companies like Coca-Cola, Lockheed Martin, PwC, and others. We’re not a co-pilot productivity tool, which typically yields around a 20-30% improvement. What we do is talk directly to the customer’s customer, like PwC for Hewlett Packard, delivering significant ROI—more in the range of 1000-4000%. One division of Pure Storage, for instance, is adding $1.6 million to the bottom line. Another division saves 150 hours per customer by using AI-driven education versus traditional LMS systems. Hewlett Packard’s customers rated us 9 out of 10, and Aruba saw a 6% increase in licenses. We start from pre-boarding for late-stage prospects and go all the way through off-boarding, handling every aspect in between.
Dickey, how do you handle data security in these scenarios?
Good question. We work with cybersecurity companies that prefer not to be listed publicly. These companies send information to high-profile clients, sometimes even celebrities. And like the saying, “If a tree falls in the forest…” if a product provides value but the customer isn’t informed, the product can be overlooked. Our cybersecurity approach avoids synchronization, which we’ve found unreliable. Instead, we don’t copy customer data to our system. We use LLM masking, where we don’t disclose specific customer names in queries to the LLMs, instead assigning generic identifiers to ensure security and privacy.
Just following up—since you’re using data for learning, is there a secure hosting solution for this process?
Yes, this could be a topic on its own. We don’t store customer data. Instead, we use LLM masking and dynamically generate prompts, which allows us to mask the identity of customers. For instance, instead of referring to PwC, we use “customer CX423,” and then rotate identifiers so the LLM doesn’t recognize the customer.
Kevin, any more questions for Dickey before we move to the rapid-fire round?
Let’s get to the rapid-fire round.
Can I pass on some? Just kidding.
Alright, Dickey—early bird or night owl?
If I have a customer meeting, I’ll get up early, but I usually don’t go to bed until around 2 a.m., so I wake up around 7 or 7:30 a.m.
You’re on the West Coast, so you’re going to bed as we’re getting up! Do you enjoy cooking?
I do—I have a smoker. I love making ribs or salmon, especially smoked with orange slices.
Do you have a favorite Halloween candy?
Not really, but my kids love Kit Kats. Personally, I’d go for Toblerone, though it’s a bit expensive for Halloween!
You’re in California—where exactly?
I’m in San Carlos, about 20 minutes south of San Francisco.
If someone visited San Carlos, where should they eat?
There’s a Brazilian steakhouse called Aspetas in San Mateo that’s fantastic. I always take visitors there!
Dickey, where can our audience learn more about you and Cast.app?
You can connect with me on LinkedIn as Dickey Singh or visit Cast.app. Scroll down, fill out the form, and it’ll open Calendly to schedule a chat with me.
Awesome. Dickey, thanks so much for joining us. This was great.
Thank you. There’s a lot happening in Applied AI—this is just the beginning.
Absolutely. We’d love to have you back.
