Digital CX And Churn Framed

SPEAKER_00 0:22

Today, we're going to talk about Digital CX's role in churn, but maybe not in the way that you're thinking. Stay tuned. Once again, welcome back to the Digital Customer Experience podcast with me, Alex Turkovich. So glad you could join me here today and every week as we explore how Digital CS and AI can help enhance the customer and employee experience. My goal is to share the insights that my guests and I have learned over the years so that you get what you need to evolve your own digital program. If you'd like more resources, head over to digitalcustomer success.com. There you'll find access to a chatbot that's trained on this content and get information about the Digital CX masterclass where I teach you how to execute and thrive in Digital CX. For now, let's get started. Hey, welcome back to the Digital CX Podcast. It's great to have you back. I appreciate you being here. And today I have a topic that is very near and dear to my heart currently. I mean, in general, but also I've been working on this topic quite a bit in my day job. And so I wanted to bring you along for the ride, so to speak, and bring you behind the scenes of some of the work that I've been doing, specifically with regards to churn and using AI in churn analysis. So

Why Churn Reasons Get Messy

SPEAKER_00 1:40

look, I think if you've been in CX or CS for a while, you know the difficulties surrounding churn reasons, right? So the typical scenario that happens, let's say you've run all kinds of churn playbooks, you've um, you know, tried to save a customer. Invariably, either it wasn't a fit or something happened, or, you know, they're out of business, whatever the reason is. The CSM or whoever is kind of processing the churn after the fact is putting in some form of churn reasoning. In a lot of organizations, it's a drop-down, a single dropdown with maybe single select, different reasons. We all know the reasons, right? It's like budget, too expensive, out-of-business product, feature request, you know, all kinds of different things that can be put into that dropdown. Sometimes they're multi-select, which I think is actually important because, you know, it's never one reason. One of the things, one of my guilty pleasures is I like to look at like aircraft in accident investigations that happen. I don't know why. It's nothing morbid, really. I'm just an aviation fan, right? So, like, you know, I'll I'll read some of the investigation reports that, you know, that the NTSB puts out after an incident. And the thing that gets stressed over and over and over again is when there is an accident like that, there is never one reason. There's always a confluence of different reasons, unforeseen, that typically cause those kinds of things. And I think churn is exactly the same way. I don't, you know, it's very rare that you encounter a churn, a customer churn event that is caused by one singularity. So of course, the multi-select drop-down, you know, is important. But hopefully, you also have some sort of narrative field there as well. So when somebody is filling out those churn reasons that they can actually provide a narrative. At Belfry, where I lead CX, we've perhaps gone a little bit overboard on churn reasoning because we have a form that includes, among other things, was the customer a good fit? What was the implementation like? Was the customer engaged during implementation? What is the sentiment that we had from the customer at the time of churn? What else do we include? We include what their product usage was like. We include fields for understanding what uh what the team did to try to prevent the churn. But we also include a field that allows us to document what could have been done differently or better to prevent that churn. We give the CSM or whoever's filling it out the opportunity to identify other customers that might fall into the same category of circumstance. So we've we've gone, you know, pretty in-depth with regards to the churn reasons. I'm not saying that you need to be this detail-oriented. We've chosen to do that just because I feel strongly that every churn event is a massive learning opportunity. Nobody likes churn, but churn's gonna happen. It's SAS. Churn happens, but I think what's important is what you learn out of what happened to lead up to that churn so that you can then prevent that in future churn. And that's an area where a lot of organizations fall flat and you know, things get busy. You're trying to close renewals, you're trying to get people through implementation, whatever the reason may be, you know, that post-churn playbook type situation just a lot of times just kind of falls by the wayside. And one of the byproducts of having such an extensive post-churn form was that, you know, we weren't necessarily getting full answers for all of those to be expected. And quite frankly, it's it is a long form, right? But we want to try and capture the narrative as much as humanly possible. But even if you just have a multi-select drop-down with a text field for the narrative, chances are that you're getting spotty information and you're getting incomplete narratives. Why is that? Well, you've got to think about who is actually populating that churn data. It is likely the person who is working that account, whether it be the CSM or renewal rep or whoever, it is likely the person who has essentially not lost the renewal, but you know, but wasn't able to recover the renewal for whatever reason. And so there's emotion around it. There's bias around it. You know, there is a tendency for people in that situation to feel the pressure, right or wrong, of you know outing themselves or like, you know, uh entering a narrative that doesn't look good in their, you know, doesn't it doesn't work in their favor from a performance perspective. And I I totally get that. Like, you know, you you're not asking for a CSM or whoever it is to like call themselves out or whatever. But one of the things that I think is very important in an organization is to, you know, foster that sense of transparency, especially around the question of what could we have done better? And you know, going a layer deeper on that philosophy of look, churn is going to happen. We get it, but how can we learn from those things? And so at that point, the churn reason doesn't become, you know, a thing to cause a sense of like victimization or a thing to point fingers. It becomes an a part of the analysis and making things better, making the customer journey better.

Using AI To Validate Narratives

SPEAKER_00 7:58

Okay, so what does all this have to do with digital CS? Well, I think that digital CS has a fantastic opportunity to pull in some governance into this whole churn, post-churn analysis workflow. And I've recently been doing a lot of work with artificial intelligence to take, you know, the churn that has happened, do some deep analysis, and actually give me improvements for the churn narrative so that I can do some regression testing. And I'm gonna walk you through this process today from start to finish and kind of what I'm doing here. But I'll start with giving you kind of the high level, right? The first thing that I'm gonna wanna do, and and for my example, I'm, you know, I'm we're heavy Claude users, and I have Claude connected to basically everything, you know, all of our, all of our systems. And I know some of you out there don't necessarily have that luxury. And for those of you who don't, yeah, I I apologize if this is like frustrating or whatnot. But if you do happen to have Claude that is connected to, you know, the system that has your meeting transcripts and your emails and your support tickets and and all of that kind of stuff, this episode will hopefully help you help you put that framework together so that you can do that post-churn analysis using Claude. So, all of that to say, what I start with is you know the churn events of the past month or the past quarter or whatever cadence you decide to do this on. And what I'm going to do is I'm going to use Claude first and foremost to analyze the responses that I have in those churn questions. In our case, it is in HubSpot. So once a renewal deal, for instance, is marked as closed lost, the churn form will appear. It will have mandatory fields for all of those narratives. And so what I'm doing is I'm using Claude to read what those churn narratives are. But then I'm having it go through emails, call transcripts, support tickets, pre-and-post-sale for all of that stuff, Slack messages, feature requests. I'm really trying to scour every nook and cranny of our customer data when doing this analysis, because my goal here isn't to change the narrative that was entered. I want to be very clear about that because I don't want to rely solely on artificial intelligence to provide me the narrative. But what I do want to do is I want to take the narrative that has been entered and I want to first validate that narrative against what we're seeing in meetings and call transcripts and those kinds of things. I want to analyze it for bias. If there is some bias in those answers, I want it to call out what bias exists. And then I want to analyze it for completeness so that we have multiple sides of the story. So, you know, there may be a churn narrative that gets entered that is for the most part, you know, factually correct. The timeline is, you know, correct. The events that transpired are correct, but maybe there's a bit of bias in there because the person who entered it is, you know, maybe slightly emotional about the whole thing. And maybe there are key details of the customer statements that might be missing that were made on calls. Maybe there are places where there is customer data that the CSM or whoever entered it doesn't necessarily have access to, or isn't privy to, or doesn't, you know, hasn't looked at the original contract. There's all kinds of different nooks and crannies where this stuff kind of hides. And my goal in this exercise is to create a complete picture of events that transpired to lead up to this. So again, I am not in this to change the narrative that was entered because I think that is an incredibly important part. And I'm going to lean very, very heavily on what was entered originally in that form. That said, if there are customer statements that were made, I am going to look at what those customer statements are based on the analysis that comes back, that it comes back with. And I'm going to validate those. Of course, customers can get very emotional too. They can say stuff that also isn't factually correct. They can say, well, my contract says this. And then you can have artificial intelligence go and look at the contract and read the contract and kind of validate those statements, right? So by doing this analysis, again, you're not invalidating what was put into the form. The goal is to get a complete picture so that you, the human, can formulate an opinion or formulate what the actual factual narrative of events is. And then what I do after that analysis is done is I'll go through it, I'll read through it, I'll read through the discrepancies that were called out, and I will make adjustments to whatever is that churn data. I take some time doing this because I think it is important to document those events for purposes we'll talk about here in a second. Now, related but kind of a separate step because it is somewhat different prompting, is I am then going to ask Claude to do a similar exercise on anything that's missing. So if there are key bits of data or certain elements of the form that weren't filled out, I'm going to ask it to provide me the narrative for those, you know, for those missing things. One classic thing where I think this is really important is one of the things that we call out in our churn narrative is is this a customer that we want to resell to down the road? And if it's a yes, then what I actually want to do as part of that churn analysis is I want to provide that future seller who's going to do that with a narrative of what is needed, what are the things that are needed before we can go resell it, whether it's a feature gap or a process or something. You know, I want to be able to provide that AE with the data of, you know, when is a good time to approach this account. But then also a brief narrative of like what happened and what topics of conversation to avoid and those kinds of things, right? So I, you know, this is essentially a warm lead for the future. And so I want to provide as much context as possible. And I think that is where artificial intelligence is really good at building those kinds of narratives. Now, I've built Claude skills for this, and I think that is an important element because when you go and do this kind of churn analysis, you're not going to want to reprompt your system every single time, right? You are going to want to essentially kick off the exercise and have Claude know exactly what you want it to do. Now, after it's done, you may decide that you want to make some adjustments to how it's doing that, and that's perfectly okay. But for this analysis, I have created Claude skills. And in my prompting, I invoke those Claude skills to go do this churn analysis based on what our tooling is and what I'm trying to get out of it and specific kind of narrative formats that I want it to pull back. So again, step one is let's validate, let's look at what we have, let's validate it against the actual data. Step two is let's fill in any gaps and make sure that our resale narrative is up to par. So that's great. Now, after you've done those first two steps, you should have a pretty complete narrative for what happened, how it went down, what we could have done better, and what the recommendations are for the future. So,

Finding Patterns Across Churn

SPEAKER_00 16:22

how do we go use this? Step three is really about using all of this churn data. And again, this is something where a lot of orgs just kind of ignore that churn data, don't do much with it. But it is a gold mine because the more of these you do, the more patterns you're going to start picking out. And you are going to start to recognize repeated things that are occurring that are either completely preventable or that require you to change some processes, to make recommendations to product, to give some feedback to sales. And so what I'm doing as a step three is I'm then conducting some churn analysis and essentially some regression testing and regression analysis on the churn events that have happened to pull out any patterns. And I'm going to separate that feedback by department. So I'm, like I just mentioned, I'm going to give product some feedback on certain things that caused a customer to churn or that contributed to a customer churn along with the revenue impact. For sales, I'm going to give feedback to sales leadership on specific promises that were made, for example, that should probably shouldn't have been, or some weird contracting thing or whatever, so that they can decide, you know, what to do and what changes to make. Obviously, for CSMs and for support and for the post-sale, I want to make recommendations on what processes we need to change. How does our churn prevention playbook need to change? What steps did we miss? Is there an additional escalation step that we should build in for a specific instance? There's all kinds of gold that comes out of understanding how we tried to prevent that churn, or how we didn't try to prevent that churn where we actually should have intervened in a special way. So step three is about just pulling data out and really understanding what are the patterns, what are the repeated things that you're seeing in churn, and how can you implement changes across the entire organization to help prevent these things from happening in the future, especially if they're repeated patterns. The one-offs, you can kind of make a decision as to, you know, is this important enough for me to really look deeper into this? But I think it's worth stressing why this sits in digital CX. You could say also why it sits in CS ops, because you have that visibility, that cross-org visibility. You have the data and the systems access and the integrations to be able to do this. It's a very natural place for this to sit. Because again, with digital CX, I preach this all the time. It's not just a customer-facing thing, it's also an internal-facing thing. And by providing this data within your organization, it is such a valuable source of customer feedback.

unknown 21:08

Wow.

SPEAKER_00 21:09

And maybe the ultimate source of customer feedback, right? NPS surveys, the ones that actually write something in response, sure, it's great, it's good to listen to, but it doesn't provide the value that a churn event does. Side note, if you as an organization conduct uh exit interviews with your customers, I know that's hard to do sometimes. It's hard to engage customers that are leaving. But if you can do that, this is again some data that should be somewhat standardized so that you can report off of it in a standardized way, right? So again, a form would be great for that, or a you know, a survey form that gives the interviewer prescriptive questions to ask and enter. Doesn't have to be difficult, especially in, you know, today with all the note takers and the call transcribers that we have. You know, this shouldn't take an interviewer a massive amount of time afterwards to fill this stuff in. Now, step four. This is where actual regression testing comes into play. Step four is I am going to take the analysis that I've done and

Regression Testing On Live Accounts

SPEAKER_00 22:25

the patterns that I've picked out, and I am going to look at my existing accounts to try and identify those same patterns in those accounts. Uh, that could be a potential warning for me. Now, in the age of Claude and AI and all of that fun stuff, this is way easier to do than it ever has been. I think in the past we've relied very heavily, rightfully so, on scorecards to try to pull out some of this data. But now we know we have this power to crawl our entire customer communication ecosystem for this kind of feedback and these kinds of warning signals. And so the more of this churn analysis that you do, the better picture you are going to get of those patterns that exist. And you can now go and look at your existing accounts and have artificial intelligence analyze your existing accounts for those same types of issues that are popping up. Couple of examples. If you see a pattern emerging of customers complaining about a service response as part of churn, that is an area where you can go look at support tickets and call transcripts and call out where other customers may be complaining about that. Same thing and intervene. Obviously, there's a larger problem to fix there, but you know, it allows you to at least identify customers that are showing signs of distress. Some other examples might be, you know, customers who have left because of a certain feature gap or, you know, product issue or something like that. Again, you can search your customer records for those same kind of mentions or look at your product telemetry, for example, for customers that are using that particular module, that feature, or have signs of distress related to that. So, you know, every business is going to be different. There's going to be different signals that you want to listen to. But the key is you identify those patterns in your churn and you go listen to your existing customer base for those same patterns, focusing initially on your customers entering the renewal flow. But you might as well do it for the entire customer base because it's a relatively easy thing to do. And again, I'm building Claude skills for all of this analysis. I'm not re-prompting every time. So it is a relatively light thing to do. And using something like a schedule task that invokes specific skills in Claude, it's a no-brainer. You can actually run this stuff automatically. If you want to do this kind of churn analysis and regression testing on a regular basis, you can just schedule it. If you want to do it monthly, if you want to do it quarterly, whatever that cadence is, you can now do that relatively easily and just have this stuff handed to you on a silver platter. It's just going to take some upfront work to do that. Now, what you're not going to be able to escape, and what doesn't become magically easier with artificial intelligence is what you actually do about those scenarios. And that's where things kind of tend to fall off, right? Because you can identify all this stuff, but you need to follow up with the playbooks to actually address those things. It could be one global standard playbook that addresses everything the same way. It could be playbooks that are specific to the type of issue or the type of problem that's identified. That's kind of up to you. That's where each organization is a unique special butterfly, right? But that's where your cross-functional relationships

Turning Signals Into Real Playbooks

SPEAKER_00 26:18

become extremely important. Because the better those relationships are, the better that you can roll this feedback without seeming accusatory, without causing negative sentiment, but also so that you can collaboratively work on the playbooks needed to actually address some of those things. So again, not your typical digital CX kind of function or thought process, but I felt it was really worth mentioning on the show because I don't think a lot of people are doing it in a systematic way to this level of detail. So again, four steps. First, validate, look at your data, make sure that you're capturing churn correctly. Step two, fill in any gaps, make sure that you have a complete churn account per account. Step three is you're then going to do some analysis, look at patterns across that data, see what's happening on a repeated basis, you know, look at your cohorts and all the different ways that you kind of segment your customers and try to figure out if there's a problem with a specific area of the business, whether it be a department or a cohort of customers or a vertical or whatever it is. Then lastly, you're going to do some regression testing with that data, with that analysis to look for those same patterns within your existing customer base. So I hope this has been helpful. Love to hear from you if you're doing something similar, if you're doing something differently than I've just described, if you're doing something additional, let me know about it because I'm constantly kind of refining this process myself. Uh, and I think it's critically important to look at your churn very, very closely. And I'm willing to bet that you may be one of the only ones in your department or in your organization doing that. So let me know your feedback. Until next week, have an awesome day. Thank you for joining me for this episode of the Digital CX Podcast. If you like what we're doing, consider leaving us a review on your podcast platform of choice. It really helps. Don't forget, digitalcustomer

Recap And How To Reach Me

SPEAKER_00 28:30

success.com is where you can find extra resources, a chat bot trained on this show's content, and the Digital CX masterclass that helps you master digital and AI in CX. My name is Alex Turkovich. Thanks again for joining, and we'll see you next time.