In this episode of Mastering CS: Candid Leader Insights, Irina Cismas sits down with Pratiksha Tewari, Senior Customer Success Manager at Tars, a platform for building conversational AI agents. Pratiksha came from field operations and Salesforce implementation before landing in CS, and she is also a certified Scrum Master who runs onboarding like product sprints.
She shares what CS looks like when every customer’s definition of success is completely different, how she measures health across multiple funnel stages, how she uses AI to surface expansion opportunities before renewals become a surprise, and why undercommitting and over-delivering is the one habit she keeps coming back to.
What You’ll Learn
- What Tars does and what the CSM role looks like when every use case and stakeholder is different
- How Pratiksha splits her time between strategizing and executing across a diverse book of business
- What good onboarding actually looks like when you’re building a conversational AI agent for a customer
- Why healthy looks completely different depending on the use case and how to measure it across top, middle, and bottom of funnel
- What the operational infrastructure behind CS at Tars looks like
Key Insights & Takeaways
Every customer has a different definition of success. In some cases a high deflection rate is great. In others it means the AI agent isn’t routing enough conversations to live reps. You have to know which metric matters for which use case before you can measure anything.
Measure health at every stage of the funnel. Top of funnel is how many users engage with the agent. Middle is what they’re talking about. Bottom is leads generated or tickets resolved. All three together tell the real story.
Discovery is not optional, it’s the foundation. If you don’t understand the problem the customer is trying to solve before you build anything, you will build the wrong thing and have no way to prove value.
AI as a research assistant changes how you approach renewals. Setting up automated news summaries for client accounts means you’re never surprised by a budget cut and always ready to spot an expansion opportunity before the customer brings it to you.
Podcast Transcript
What Tars Does and What the Role Actually Looks Like
Irina (0:09 – 0:51)
Welcome to Mastering CS Candid Leader Insights, the podcast where we dive into the world of customer success with industry leaders. I’m your host, Irina Cismas, and today I’m joined by Pratiksha Tewari, Senior Customer Success Manager at Tars, a platform for building conversational AI agents. Pratiksha, I’m really happy to have you here.
Thanks for joining!
Before we dive in, you’ve worked through field operations, Salesforce implementation, and now you are a senior CSM with Tars. Paint us a picture of your day-to-day there. Who are your customers and what does the work actually look like?
Pratiksha (0:52 – 2:33)
Before I give you an understanding of what my customers are, I’ll tell you a little about the product so you can draw the comparison. The product at Tars is an AI agent. Our customers come from multiple industries: healthcare, telecom, FinTech, and more. There are also a lot of different stakeholders to manage across these industries. In FinTech, for example, I would probably be managing a lot of technical POCs, while in healthcare I would mostly be in touch with the marketing team.
Industry to industry, client to client, a lot of stakeholders differ, and with that, the kind of communication changes. In some cases, if the use case is support, like in healthcare, what matters is reducing the workload of the reception staff or tele-sales users who get a lot of inquiries. The work is around how to reduce support tickets. Versus when it becomes a marketing use case, it’s around how to increase the leads being generated through the AI agent. Client to client, stakeholder to stakeholder, every day is very different. It completely depends on the kind of work I’m doing for the client.
Irina (2:33 – 2:44)
For sure we will. But let me know, in this setup, what do you spend most of your time on? How does a day in your life, a day of a CSM look like in your case?
Pratiksha (2:45 – 3:58)
Most of my time is split 50% strategizing and 50% following up on the strategies that have been built. The 50% of strategy building is planning before executing: what needs to be done in this account, defining success metrics so that I’m not fighting for renewal at the end of the quarter or end of the year. Setting the gold standards in the beginning, understanding how we achieve that, looking at the reporting, the numbers, usages, and trends that the data speaks to.
And then implementing and executing those actions. It’s not that I can do everything, there are team members who help me out with a lot of those strategies. Following up with them, following up with clients a lot of the time to see if what we have been implementing for them is working out or not. Doing status check-ins with them. So yeah, I would say a lot of strategizing and a lot of following up with internal and external clients to keep that harmony going.
What Good Onboarding Looks Like at Tars
Irina (3:59 – 4:15)
Let’s tackle one of the important processes in customer success, and that is onboarding. How does good onboarding look like in your case? When a new client comes on, where do you actually start?
And how do you know when you’ve got there?
Pratiksha (4:17 – 6:31)
Good onboarding starts with understanding who’s sitting on the other side, who you’re talking to. There could be a lot of stakeholders, some decision makers, some showrunners. The decision makers decide what needs to be done and whether the spend on the product is meeting expectations at the end of the year or quarter. The showrunners do the day-to-day or weekly tasks. So get a good sense of who’s responsible for what on the other end. Good mapping of customers and their responsibilities.
You don’t want to be asking for integration documents from a marketing person. Let them know beforehand what you’ll require in the project and who you should be reaching out to for each thing. If that’s cleared out in the beginning, it really eases up the entire process. You’re speaking to the right people, you’re not wasting time, there’s less noise. So clear out who the right POCs are.
Secondly, explain what the project stages look like, what the timelines would be, and what the current pain points are that they’re trying to solve. Once those are clear, you can start defining and measuring success once it’s implemented. If you don’t know the problem you’re trying to solve for your customer, you won’t be able to get to the solution. Speak to them, have a very good discovery call in the initial periods, understand their current pain points, understand why they want you as a product.
Once that’s clear, turn your role into that of a solution architect. You know your product best, design it based on your customer’s requirements, and implement it. Once that’s implemented, start tracking progress: how it’s performing, how it’s delivering value or not. Before you can deliver value, you have to know what problem you’re trying to solve. Get to the root of it, see how you can make your client’s life easier, and then set project timelines and stages and try to stick to them as much as possible.
What Healthy Looks Like Beyond Go-Live
Irina (6:33 – 6:41)
And once your customers are live, how do you measure whether things are going well? What does healthy look like beyond go live?
Pratiksha (6:43 – 10:06)
Because there are so many different clients, industries, and use cases, what healthy or a particular goal looks like varies. If a use case is around support, where the agent is being used to reduce the client’s workload, the metric I need to look at is deflection rate: how many tickets are not being created for the tele-sales or support team because the agent is taking care of them. But in another use case, what they really want is a lot of tickets because they want their users to be speaking to live reps through the AI agent. So a deflection rate, which is how many interactions the AI agent handles itself rather than handing off to a live person, is a measure of success for one use case, but for the other it’s the opposite. The same metric can be perceived in completely different ways, and it varies for different industries.
What I usually do in such cases is define healthy in two ways: quantitatively and qualitatively. For the quantitative side, we break it out into three parts: top of funnel, middle of funnel, and bottom of funnel. Top of funnel is how many users are interacting with the AI agent, how many are opening it. Middle of funnel is how the engagement is going, what the users are talking about with the AI agent. This gives us a lot of insights into what the customer’s users actually want. Maybe you’re selling a balloon, but what they want is party streamers. Those kinds of insights come in when you look at the middle of the funnel. Bottom of funnel is how many leads were generated or how many support tickets were handled by the AI agent.
Those are the metrics you can look at, but understanding why something happened is also important. That’s when you move on to qualitative analysis, going deeper into the data to see what happened in individual conversations happening on the AI agent. Dashboards give you high-level numbers, but it’s also important to get your hands dirty once in a while and look at what’s going on at the conversation level.
How I define healthy varies from use case to use case, but when I’m trying to define it, these are the two things I do: break it down into a quantitative and a qualitative objective that needs to be achieved.
The Operational Infrastructure Behind CS at Tars
Irina (10:07 – 10:18)
What is the operational infrastructure behind all of this? What platforms are central to how customer success actually runs day to day at Tars?
Pratiksha (10:20 – 11:32)
Customer success is around good communication. If you are communicating well internally and externally, you are maintaining relations, which is what a CSM is supposed to do.
For internal communication, we have Slack as the main channel. I love Slack, the way you can just reach your stakeholders and internal team members very easily. And then we have email and video calls with our clients.
Apart from that, we have HubSpot as a CRM. We maintain all the post-call notes on HubSpot to keep track of how the account is doing, we maintain the deals, the project statuses, and we also track support tickets. HubSpot is integral.
And then we have AI note takers as our companion, which really help with the post-call notes and action items.
Irina (11:33 – 11:47)
How is the product usage? How are your customers using your product? What features?
Pratiksha (11:48 – 12:04)
All of that is available on the Tars platform itself. We get a good sense of the usages on the platform and track that well. And actually, about 50% of my time goes on the Tars platform. I don’t know how I missed that!
How Pratiksha Uses AI to Surface Expansion Opportunities
Irina (12:04 – 12:20)
Yes. Well, yeah. What are the AI use cases are you implementing?
How do you use AI and in which conditions?
Pratiksha (12:21 – 14:40)
AI is my assistant, the most helpful assistant I’ve ever had. Starting off with the note takers, of course, that’s a big help. And then drafting good emails and difficult emails, those are just basic use cases.
The one I really found helpful was using AI for research work. We face this a lot when renewals come up. We suddenly get information from clients that they’re going through cost cutting, and that impacts us and makes us feel like we were not really prepared for this. There is a lot of automation you can do on the backend that keeps updating you, say on your Slack channel or email. We can set up automations that send you monthly or weekly updates on what’s happening at your client’s accounts on a general level, like if there have been budget cuts or if they have expanded into a new vertical.
Having an AI news summary for your particular clients coming to your inbox really helps with expanding accounts as well. If you know what’s going on on the other side, you might be talking to one stakeholder from your client’s account, but there could be other verticals where the same AI agent could be used. So having AI as an assistant that does all the research for you and tells you about untapped opportunities is very helpful.
Sometimes we’ve gone to clients and told them, hey, you’ve been using the product for service A, but we see you’ve also started doing B, and you could use this for that as well. And we create a demo to show them how the AI agent could work for their new service or vertical. It doesn’t always come naturally to clients that they would need your product for something new. It’s also good to go to them proactively and say, hey, you could use this for this as well. AI has been helpful in doing that research work for us and has helped us expand accounts.
Running Implementations Like Scrum Sprints
Irina (14:41 – 14:55)
You became a certified Scrum Master and brought back into how you deliver. And I know that you are running onboarding like sprints. For a CSM listening, who’s never run an implementation that way, where would you tell them to start?
Pratiksha (14:57 – 17:54)
For a CSM, it’s very important that the account is succeeding. What matters is a good and timely implementation so that the time to value is fast enough. Otherwise, if the implementation goes on and on, you’ve really wasted three months of a six-month contract period. Now you just have six more months to show value. You don’t want that. You need as much time as possible to have the product performing and the data speaking to the success metrics.
For a good implementation, you have to start with a very good discovery call, as I mentioned. Know what needs to be solved, know your customer’s pain points. Ask them straight up questions: what was your life before Tars and how can I fix your life with Tars now? Have those direct conversations, understand their current pain points, and then go to the problem and start solving.
When you actually start the implementation, you face a lot of challenges. The product might be functionally ready, but the integration work is not. These are two very different things that go hand in hand. There could be multiple systems to integrate with, the CRM, and many other tools. There are security checks to pass, field mapping to do, and so on.
If changes are continuously coming in and you’re not able to close the project on time, it’s best to go with agile methods. What we’ve done in those situations is give customers two or three options when things start going a little off track. We tell them, let’s push out version one, the best we can do right now, and monitor your user behavior before iterating further. Version one doesn’t need to be the best version, it just needs to solve your main problems, not all of them. If your main problem is solved in version one, push it out and keep iterating as needed.
That’s what I would suggest to any CSM out there. Go with agile methods, make changes when needed, and come up with the solution that helps you deliver value the fastest.
The Habit That Still Shows Up in How Pratiksha Runs Accounts
Irina (17:54 – 18:12)
And the last question before we wrap up, you’ve worked across field sales, operations, CRM implementation, and now enterprise CS. Is there a habit, a way of thinking or something you learned the hard way early on that still shows up on how you run accounts today?
Pratiksha (18:14 – 19:33)
As CSMs, we really are the good cop, the one clients trust. And it’s very hard to say no to them on their face when they come up with custom requests. So when you get a lot of requests coming your way, it’s very important to understand the reasoning behind them. You don’t always have to say yes upfront. If they ask for a feature, understand why they’re looking for it, do the reasoning behind it, and then decide if you really need that feature for them or not.
The second thing I found important is to never overcommit. Always undercommit and over-deliver. If the client is rushing you, tell them it’s going to take three days, even if you know you can finish it in one. And when you’ve done it in one day, tell them it’s done and they’ll be happy. Always undercommit and over-deliver. It’s good to surprise them in a good way, rather than delaying things and telling them it’s not going to work or we need more days to figure things out.
Irina (19:36 – 19:49)
Pratiksha, thanks so much for joining me and for taking the time to share your story. And to everyone listening, thanks for tuning in. Until next time, stay curious, keep learning and mastering customer success.
