The AI Race Has a Customer Experience Problem

In this CX Today interview, Rhys Fisher, Associate Editor at CX Today, speaks with Muj Choudhury, CEO of RocketPhone.ai and former Salesforce Director, about the uncomfortable questions surrounding enterprise AI.

As organizations race to deploy AI agents, Choudhury argues that bigger investments and flashier product names do not automatically produce better customer experiences. This is a candid conversation about data, workflow design, agent coaching, risk, and how to find AI use cases that improve service rather than simply chasing automation.

AI may be everywhere in CX, but Choudhury says the industry needs to get far more honest about where it delivers value and where it creates new problems.

Drawing on his Salesforce background and his work at RocketPhone.ai, Choudhury joins CX Today’s Rhys Fisher to unpack why enterprise AI success depends on more than an LLM, a chatbot, or an ambitious product launch.

Why Salesforce’s acquisition-heavy history and changing AI strategy create a difficult integration challenge, especially as smaller, less encumbered competitors move quickly.

Why customer data, including voice and contact center conversations, is the essential foundation for relevant AI experiences.

A real-world field-service example showing how AI can help engineers arrive with the right part, reduce wasted visits, and improve the customer outcome.

Why AI should support human coaching and risk management, not replace experienced managers or introduce unsafe, non-compliant service interactions.

Choudhury’s message is clear: start with people and broken processes, then apply AI where it produces a measurable result.

Watch the full interview, then assess your own AI roadmap: Is it solving a specific customer problem, using trustworthy data, and keeping human accountability where it matters?

For more Customer Experience tech news visit https://www.cxtoday.com

Transcript

SPEAKER_00 0:08

Hello and welcome to CX Today. I'm Reese Fisher, Associate Editor, and today I'm delighted to be joined by Mudge Chowdhury, the CEO of Rocketphone.ai and a former Salesforce Director. Mudge, thanks for joining me today. How are you doing? It's my pleasure. Thank you.

SPEAKER_01 0:24

I'm pretty hot, but apart from that, everything's bad late.

SPEAKER_00 0:28

Yeah, I think we all are in the UK over what feels like the last month at least. We're not uh we're not built for this weather, are we?

SPEAKER_01 0:35

Absolutely, absolutely.

SPEAKER_00 0:36

But we'll we'll try and persevere. Um, because you know we've got a really interesting chat today. I think it's gonna be I think it's one of the one of the big trends in the space right now, to be honest with you, this is we're gonna be talking about this idea that with all this substantial AI investment in customer experience, in customer service, is it really sort of delivering on its promise, I guess. And given your experience that we touched on with Salesforce, we're gonna maybe use them as a little bit of a lens for this discussion, especially uh considering the recent news that Morgan Stanley actually downgraded its price target for Salesforce. And I know the framing around that was uh timing and commercial proof, more than perhaps outright AI skepticism. But in your opinion, where do you think the gap is between the AI story that Salesforce is currently selling and the value customers can actually reliably realize today?

SPEAKER_01 1:30

Yeah, it's it's a good question. Um, and it's quite interesting because I think it goes back um way beyond this uh Morgan Stanley sort of reaction to what's happening at Salesforce right now. I think if you look back at Salesforce's history, um they've made a like a ton of acquisitions over the years, um, and uh and all those acquisitions kind of have an impact downstream on future strategy, how they um how they address the market, how how they address um upcoming needs and requirements, but also how they have to deal with things that they didn't foresee. And it's that last one I think it becomes quite problematic for a lot of these organizations because when you make large um acquisitions and very large investments, you're kind of putting a very, very big stake in the ground um around what your trajectory is going to be. And that trajectory has to follow that stake that you put in the ground, right? So if you go back and look at the acquisitions over the last 10-15 years or whatever it is, I mean, when I was there, um we acquired, we had some big acquisitions then, the exact target, I think that two and a half billion dollars and demand were like you know, nearly three billion dollars, and then there was like MuleSoft and Tableau, and I think more recently um owned company. Um, and all these companies kind of plug gaps within the Salesforce stack in in some shape or form. Um, and then in fact, and and those are more like data acquisitions because the exact target thing is about marketing and reaching the right prospects and customers, and and and mule soft is all about the integration of multiple systems, and you know, we used to call it enterprise application integration back in the day, but now it's just I think it's just generally called integration and Tableau is the the front-end visualization aspect of all this sort of back end stuff, and then and and that was all data-oriented, marketing-oriented, but but they made some pretty pure, what I would consider to be pure play AI um acquisitions as well. So they bought Relate IQ. This is all off memory now, so I might have the names wrong, but Relate IQ was uh um was a way for using deep learning to map um, you know, customers and and and predictions. Um prediction I own MetaMind uh were sort of pure play AI AI acquisitions. Um so what they've got is this big smorgasboard of like technologies, right? And um and I think the challenge that I see time and time again, especially as custom companies mature, like Salesforce, is that um you you you make an acquisition, right? And there's a lot of fanfare about the acquisition, there's lots of build-up towards it, and lots of planning, and everything gets really, really excited, and then the acquisition happens, and it and and then who's left holding the baby, and it's usually the product teams. And the product teams have this horrible job of you know stitching all together, making it work and delivering on the original promise of the acquisition. And and so long answer to your question about the Morgan Stanley situation and and the current sort of perception of sort of Salesforce in the AI space is that I think they're pretty saddled with a lot of stuff. Um, and and I think they're struggling to figure out exactly how to make good of those investments um uh and and the acquisitions. And the investments aren't just like the the acquisition, it's also all the internal work that they've been doing around you know the Einstein. Because the Einstein was like uh let's pull together all these prediction IO and meta-mind acquisitions and call it Einstein, and we can use for like predictability scoring, and then it was like, okay, we've got this data cloud thing, we're gonna buy that, and then let's acquire Informatica as well. And and so now we're gonna rebundle it all and call it Einstein GPT because we've thrown LM into the mix and and we're gonna let's call the AI cloud for the time being, and then now let's call it Asian Force, right? So it's I I I I think so much has happened um with Salesforce over the last just in the last five years, it's very hard to predict where you're gonna go from this point on. And I I talk about one stake in the ground, they put like numerous stakes in the ground, and some of these are a little bit disjointed if you look at it from a sort of you know strategic perspective. Um, so I I I think the worry that the market has is um A, how do we leverage what we've done over the last few years? Um, how do we do this in a way that stays ahead of all these smaller companies that are completely unencumbered, but they're you know, able to sort of start start from a start from a blank sheet completely. Um and and how do we remain relevant and nimble and and you know and and innovative um in the in the face of an industry that is emerging to be super important to everybody, given that our background is in enterprise applications and and workflows and user experience. Um it's a very different world that they've come from to the world that we're seeing coming ahead. And and the expectation is really high. Like Salesforce revenues are whatever they are, and the market expects them to be significantly higher. Some of some of the tricks are the MITA pools, not tricks, marketing tactics, like you know, let's let's rebrand existing products and call them agent force service, agent force sales when they're really the same old products and and then recategorising the revenue around that. I I think I think the world, you know, sees through that. Um and so it just creates some challenge around, you know, how you're gonna plug that revenue gap, um, how you're gonna how you're gonna stack up in you know um in in this new uh set of expectations and and this new sort of market landscape. Um and then you know what does the strategy actually look like, right? Because you know there's that saying, like, you know, when when your only tool is a hammer, everything looks like a nail. Like I I I saw what they were talking about with Slack, and Slack's gonna become the new headless, you know, user experience layer, right? You know, yes, we spent billions and billions on making Salesforce.com super slick and ease to access because ultimately what is CRM? It's just a database of customers and deals and customer support cases, right? Uh the real value there is how accessible is that database, and you make it accessible by having a super slick user experience, right? So you invest billions in beautiful workflow processes and workflow designers. And if that stuff is all going away, where does that leave them? Do they become relegated to become a system of record vendor? Uh, and are they really going to replace that user experience layer uh with Slack, right? And and call it headless, but it's not headless because you're replacing one head with another head, which is Slack. So it's all a bit sort of convoluted and it it's a bit confusing. Um, I I I they'll find their way, it's never easy. And even for a smaller software company like us, it's never easy. But you know, there's there's a lot of smart people there, and I I think it will resolve itself in time.

SPEAKER_00 7:58

Yeah, yeah, it's really interesting hearing you talk about kind of the the acquisition situation there. Because it in my mind immediately it was kind of parallel with CX and customer service agents right now, which hearing a lot about kind of tool fatigue and a lack of connected systems, and having all these tools is almost more of an issue, and you're almost pointing to a similar thing there where they've got all these acquisitions, all these options, but they're perhaps struggling to knit it all together clearly.

SPEAKER_01 8:22

Yeah, absolutely. Absolutely. I think uh it's an interesting place to be right now.

SPEAKER_00 8:28

Yeah, for sure. I also want to bring up what you mentioned there around kind of a lot of these acquisitions being linked to data, you know. As again, as you've said, Salesforce is clearly investing a lot of time and resources into Agent Force, but the backbone of that is obviously the data. What perhaps are I guess some of the common readiness gaps that enterprises maybe underestimate before it comes to actually deploying AI agents?

SPEAKER_01 8:55

I mean, I I I think you you've hit the nail on the head. It's um it's a data aspect of it. Um I think the biggest the the there are a lot of areas of sort of preparedness that you need to be sort of ready for. So there's you know, the all the organizational stuff. Uh I think there's a lot of fear um, you know, in in companies right now about AI taking jobs. And am I gonna be left redundant? You know, my my family is full of very high-profile lawyers, and uh even they tell me, uh, you know, I chatted to my sister last week, she said she was saying um we don't actually need junior lawyers anymore because the stuff that we used to ask them to do, I can just ask ChatGPT to do it. And it's really unnerving. And uh and in fact, at which point does it then eat into um our relevance, you know, in in this field and how does it. So there's a lot of there's a lot of fear. Um and and I think interestingly, like, you know, before um, you know, try related to change uh was often it was often sort of confined to um areas of industry that weren't causing the fear. But I think uniquely for us, you know, we're in the tech space and people in the tech industry are themselves in fear of what's about to come. Um and and and so I think that's a very big part of sort of being prepared is set the expectations correctly. Um I I think from a technology perspective, the data has to be good because without good data, you're not going to be able to deliver the level of experience that you wanted, Liberty of customers. Um, one of the examples that I use is um customers often ask me, like, oh, why do we need your technology at Rocket Phone? Because you do all this clever AI stuff. We could just be using Chat GPT and and and it's just it's a very, very different, um, a very different set of circumstances, right? Because you can't go to Chat GPT and say, hey, you know, I need a new car insurance policy this year. What's your recommendation? Because it knows nothing about me as a person, it knows nothing about my driving history, what car I drive, all that sort of stuff, right? So whatever the response is gonna be very, very generic. So um it, you know, I I think companies, enterprises in particular have a pretty unique um, and especially larger enterprises, uh, but they have a pretty unique um advantage in the sense that they have a lot of historic data. Um, and they're also capturing data every single second of the day, whether they know it or not. And and preparedness for me is um how do you leverage what you have today? Okay, but more importantly, how are you gonna tap into the data that's being generated? Like in the customer experience world, a lot of you know engagement with customers happens over, you know, in the contact center. Um, if you're out in the field, it's gonna happen over mobile phone calls. And and those conversations are in themselves data. And how are you tapping into that and and making that part of your part of your strategy? And and I don't think a lot of companies have even thought about that yet.

SPEAKER_00 11:43

Yeah, yeah. No, it makes uh makes a lot of sense. I suppose I was interested in, you know, once we get to position perhaps where you have got your data in line, you've got it where it needs to be, and then you know, the next step is that actual potential AI implementation. I think quite often AI tools and CS and customer service, they're perhaps they're limited to cost-saving accesses or cost-saving tools. What do you think are maybe some of the other measures of success that could maybe sit alongside those cost savings? Because that is obviously still a very important issue for companies.

SPEAKER_01 12:17

Yeah, um, so I I think this is gonna sound strange coming from somebody who runs an AI-driven software company, right? But AI isn't everything. Um, it's never the panacea that's gonna cure all your ills. Um, I remember when I was at Oracle and we were selling, you know, SIBL products, you know, the SIBL was the CRM Goliath back in the day. And uh, you know, our sales teams would pitch it as like, you know, it's gonna solve it's a platform that solves everything. It's gonna hand all your customer success, uh, customer support, sales, marketing processes, and so on and so forth. And it never did because it was just technology. Um, and in order to be successful with anything um like this, you you have to sort of go back to the basics and sort of ask yourself all the all the usual questions, like, you know, okay, so let's look at the business holistically. Um, let's look at where the challenges are, where the bottlenecks are, what can we improve, what's not working, you know, how many steps are there in this process? Can we can we optimize that? So you've got to just go right back to basics and say, right, what is it that we're trying to improve? Um and and some of it is really boring, right? And some of it doesn't need AI, you know, wall to wall. Some of it like you got to look at your enterprise and you say, right, let's start with the people first, because you got to go in order of expense um and and and you know, and and efficacy. So you all start with the people because you you're your your people uh are are the biggest sort of drivers of change in the organization and and and and the biggest levers that you can pull in terms of efficiency, right? So let's put technology to one side for a second. Is this stuff that we can do with the people, retrain them, educate them, you know, change the way they work, uh, you know, the way they interact with customers, the way they handle particular, you know, uh internal functions. And if we can improve those things with our people, then are we gonna get a much better outcome? And often the answer is yes, right, because you know that's one of the best investments you can make before you even look at the technology. And and then you've got to look at your internal processes, right? So this is classic business process optimization. What are we actually doing? In every organization, you you're gonna find that you know there's been a lot of uh uh you know uh creep in in terms of how they get things done, like uh processing an order um 10 years ago versus how you do it today is is gonna be very different because you over time you keep adding and adding and adding because every customer, you know, you you st you you launch your business as a subscription-based software company, right? Speak from experience, um, and and you want annual payments up front because that's what we're used to having worked at Salesforce and other companies. But now we have customers who want monthly payments. So now you have to modify your order processing systems to accommodate monthly payments, and every company evolves in this way. And and you know, if you look at your processor, you will find that there's a lot of inefficiency in there. So address that stuff, right? And and then some of these things you might need technology to help you achieve or deliver those things, and that's where tech stacks come in, that's where you know um adding technology can help speed things along, make things smoother. And that's when you start looking at technology, and it's not just AI, AI AI can be a part of it, right? Um, I'll give you an example, right? You know, so so so we we're working with um a very large manufacturing company based in France, but they have offices all around the world. And uh, you know, most of us in the UK, because of freezing, you'd never guess looking out the window, but it's freezing here most of the year, right? So um we have boilers to heat our houses, and and boilers don't work sometimes. So you so you call whoever to come and fix it, and they're gonna show up and they're gonna say they're gonna have to look at it, inspect it, and they're gonna say, Oh, I've got to come back in two weeks because I haven't got the pot that you need to fix it, right? For me to fix it. I'm gonna let me order it, I'll come back in two weeks. And then you're in this sort of like horrible, sort of like it's three months before everything gets sorted out. So we have a customer who, you know, they make boilers and they have a field team who goes out and fixes them. And uh, and and and so the they're using Salesforce and a couple of other systems to handle all those interactions. We've done a very simple thing for them. You know, we've given them um our SIM card. So the field engineers have like a mobile phone each, and and and they have to make calls to their customers anyway before they go and see them. So, what they do now is they have the conversation, what's the problem, what are the symptoms? And the AI in our system listens in real time to the conversation and he'll say, you know what, Mudge, um, it sounds like the heat exchange is broken. You haven't got one for that particular boiler in your van. Don't even bother going. I'll I'll get one in order. What you can do is go to the next job. And you've just, you know, um, you've just solved a whole bunch of problems, right? The the experience of the customer is gonna be fantastic now because for the first time in their lives, they're being told, I'm not gonna come, but I will come next week when I have the part that you need. And and and then helps another customer who would have been waiting maybe a couple of days before they got to see the engineer. So and and and and the AI portion of that is about 20% of the overall, because there's still Salesforce, there are still workflow management systems, there are still GPS and GIS systems at play, um, and and calendaring systems and scheduling systems. But the the small portion of AI in there, which is like 20% of the solution, which has uh a significant impact. But to to get to that inefficiency, we had to just sit and just spend a lot of time with them and and literally go out with their drivers and observe what they do day in, day out. So you can't just go online and sign up to Claude and think that's going to solve all your business problems. It's never that straightforward. You know, you still need people to sit there and really think about the issues.

SPEAKER_00 17:32

Yeah, that's really cool. That reminds me, it almost sounds a bit like uh kind of like a triage system in a doctor, you know, obviously you've got someone checking over, so we'll go and see the cardiologist, or we'll go to see an x-ray or whatever. Yeah, it's a really it's a really interesting real life example of it. Cool.

unknown 17:50

Cool.

SPEAKER_00 17:50

You also mentioned, which I thought was interesting, you picked up on kind of training as maybe an area that people don't pay enough attention to. And this is something we hear a lot, but I suppose kind of the catch 22 of the situation, I think we hear a lot, is a lot of companies agree they need to do more training for their agents, but the call on top is we need AI because everyone's talking about AI, let's get AI. So what's actually happening is we're getting more tools, and we weren't trained up enough on the tools we already have, and now we've got to incorporate more tools into the into our daily routine. I just wonder if you could talk a little bit about that. Is that something that you come across, or what would maybe be your advice to help agents in that situation?

SPEAKER_01 18:30

Yeah, so we come across um sort of uh agent improvement, sales rep improvement, you know, um quite a lot because uh we we have systems that can record conversations, whether they're cell phone or fixed line conversations, and uh you know, coaches want to use that, user recordings for training purposes, as we as we all know. Um and and and then you know, and then w when they speak to us because we're we're like an AI-first software company, they say, okay, can we automate all the training, right? And and uh have an AI do all this sort of stuff, right? And so you can't, you can't, you know, we'll we won't be there for a very long time because you know every business is is heavily nuanced. Um the way people work is heavily nuanced, and it it requires an experienced, you know, human being with empathy, right? Who knows how to deal with people to be able to you know coach correctly and properly. And and if if that person comes with um their own experience um of having done that job successfully, then they will carry that credibility and and and they'll be more convincing and people will take to them more, right? So you can't just replace these people with AI, it's just not possible. What we do do um is you know, we can use AI to um highlight areas for improvement, right? So um so during conversations, you know, we have systems in place today where the AI is trained to listen out for you know sort of things that matter to that company and and what matters to each company is different because financial services, you're very worried about compliance and regulations. So if your rep is saying something completely out of compliance, it might land you in a lot of hot water. So you want to catch those fairly quickly. And there are there are levels and degrees of of severity as well. Uh so some things, um, if if a rep is saying something that's super um uh risky to the company, you might just want to stop the call altogether. That's like a severity one, and and a severity two might be, you know, you want to pull in an experienced hand just to help that person out. And severity three is just advisory, it's just like, you know, okay, here's some advice on what you should do next time you have this conversation. Um, and what AIs are very good at is sort of categorizing the level of severity, um, listening out for things that matter to the business, um, and then flagging those things for a coach to then pick up later on um with that rep. Um and that works really well. Um I I I I'm I'm again, this is gonna sound really odd coming from a CEO of an AI software company, but like, you know, the the best AI is is where you combine you know the human effort alongside alongside the automation um so that the humans are better informed um and they're better equipped and they have a bit of help so they can get through a bit more than they normally would be able to. Uh and that's kind of how we approach training uh from an AI perspective.

SPEAKER_00 21:18

Yeah, yeah. It's uh yeah, it's good to hear, like you said, people in your percent in your position who are obviously heavily invested in AI but still recognizing that human element, which I think maybe leads mostly to my final question I wanted to ask you. And this is something we've kind of picked up uh in our in-house analytics lately, as a bit of a trend we're seeing across our audience, and it's something that I want to be asking more people about. But with AI obviously becoming more and more prevalent in customer service experience, do you think that AI is making CX better or is it just making CX cheaper?

SPEAKER_01 21:52

Oh, that's a really good question. Um I I I think um you can make Okay, so I don't think it's making it any cheaper. Um I I I think there's an economic argument to say that whatever you're paying for your AI tokens right now is artificially low and expect that to course correct um in due course. Um I I think uh I mean any any future rises in the cost of AI might be mitigated by scale. So uh or economy to scale, I should say. Um but I don't think that's gonna be the case because a lot of these larger companies who OpenAI, you know, Claudia, whoever, Google, I think the economies of scale are built into the pricing already. Um but I think the cost for them to run the services is probably um higher than they're letting on, which is why we have all these sort of challenges around data centers and um you know being able to have enough compute capacity around the world. Uh so that's that's a very separate economic argument. Um so if if you think it's cheap now, it probably won't stay there forever. Um is it making customer experience better? Um I I think it can do. Um you can have very quick wins. Um I've seen you've got to be really careful though. I've I've seen a number of uh instances where you know they think that they can stick a stick an LLM to a chat bot and then hey, Presto, you have this like you know, super cool customer experience, which is very human-like, but it's all being done by a bot. And uh and I think organizations can be very sort of like hasty to deploy those things, and and they uh they ignore things that are really important. For example, is there a health and safety aspect to the response that you respondes that you're giving out, right? And this is obvious, right? With boilers, for example. So what you don't want to do is ask a consumer to open up the front panel of a boiler and start removing components, right? Because AI knows no different, because they're not they're not certified to do that, and they could very easily blow themselves up, right? Uh, and that's a very, very obviously obvious example. But I are you, you know, uh look at your industry, look at your business, and you might be financial services. And and are you sure your bots aren't making recommendations that could be perceived or construed as advice, and it might be construed as bad advice, right? In which case you're definitely going to get a massive slap on the wrist from the FCA. So there are lots of nuanced things that you really have to think about, and um, and and and those things can deliver a very quick hit um in a narrow context. Um, but done in any irresponsible way could land you in a huge amount of trouble. Um, and in that case, the customer experience uh becomes um worse and it also becomes a lot more expensive. Um, so it is a really good question, uh, and I think uh the answer is not super straightforward. I think there's a lot to think about there.

SPEAKER_00 24:50

Yeah, yeah, I think that's probably probably a fair assessment. Thanks, budge. I really, really enjoyed this chat. I really kind of appreciated your uh your your honesty and your candor. I think you're probably the most AI skeptic owner of an AI company I've met, which I think is really it's really good for chats like this. So I think it uh yeah, it shows I think that gives far more insights for our audience than people who are just gonna sing the virtues of the product. So yeah, I really appreciate the chat. I think it's been really insightful stuff. Okay, dokie. Great. I did also just want to quickly uh thank our audience as well for tuning in. If you enjoyed this, and I'm sure you did, please do remember to like and subscribe to the channel and head on over to txtoday.com for more stories like this. Until next time, thanks for watching.