[00:00:00] [00:00:00] Teaser --- [00:00:00] Ankur: Executive, uh, regularly reviewed AI initiatives, but rarely discussed basic functionality problems, So what we did, we implemented foundation first approach, pausing all AI projects for 120 days to fix the fundamental journeys gaps. This results were remarkable, uh, addressing basic automation, improved adoption by 28%. Generating 4.2 million in additional revenue. When we later implemented AI engine on the Stronger Foundation, the effectiveness increased by 41%. [00:00:29] In This Episode --- [00:00:29] Phil: What's up everyone? Today we have the pleasure of sitting down with Ankur Kothari, adtech and Martech consultant, who's worked at big tech names and finance consulting firms like Salesforce, JP Morgan, and McKinsey. [00:00:41] In this episode we cover AI personalization that actually increases retention, a step-by-step approach to AI personalization with a maturity framework. Why you should fix the stack that you have before you buy another tool. How to balance AI personalization with human context and [00:01:00] predictive loyalty. Engines are coming to a MarTech stack near you. [00:01:05] All that and a bunch more stuff after A super quick word from two of our awesome partners. [00:01:09] ​ [00:03:26] Phil: Ankur Thanks so much for your time today. Really excited to chat. [00:03:29] Ankur: Thank you, Phil. Thanks for having me. I as a mar enthusiast, I love watching your podcast. [00:03:35] Phil: Awesome. I appreciate you saying that. Yeah. We've had a few episodes recently where we focused a lot on AI's role in acquisition and outbound type of efforts. But today I'm excited to chat with you about maybe the more overlooked intersection of AI personalization and how to use MarTech to integrate all that stuff and improve retention through personalized experiences. [00:03:59] A lot [00:04:00] of folks focus on the outbound acquisition 'cause it's easier to kinda like portray, but retention is just as important in activation. Um, maybe we can start with the practical example. [00:04:10] AI Personalization That Actually Increases Retention --- [00:04:10] Phil: what do you think is actually working right now when it comes to personalization? Can you maybe share a real example where AI driven personalization has meaningfully improved retention metrics? [00:04:21] Not just identity metrics? [00:04:25] Ankur: About two years ago, I was a consultant to mid-sized, uh, e-commerce company in the home, good sector that was struggling with, uh, customer retention. The churn rate was hovering around 68%. They were losing approximately, approximately 2 million in annual recurring revenue. [00:04:39] Due to one time shopper never returning. The traditional marketing approaches were not working because they were treating all customers with the same generic messaging, regardless of their browsing behavior, purchasing history, or engagement patterns. When we dig deeper into the data, we discovered that the customers are abandoning the brand, not because the product [00:05:00] quality or the product issues, but because of the shopping experience felt. Impersonal and disconnected. The real issue was not just the acquiring new customer, but about understanding the existing customer needs at much more granular level. what we did is, uh, implemented AI driven personalization strategy to website interaction, purchase history, email engagement, and customer service. We deployed machine learning, uh, algorithms to analyze the behavior pattern and predict purchasing, uh, intent. by creating 27 different personalized customer journey based on the AI prediction, we automated the content delivery across email, website, and mobile app. Based on the individual preferences that focus on creating individualized customer journeys, we consolidated first party data within six. months. We saw remarkable improvements by key business metrics such as, uh, customer retention increased by 42%. Average customer lifetime value grew from dollar 1 27 to [00:06:00] 203. Repeat purchase, uh, rate improved by 38%. Overall revenue increased by 3 million, which is across 22% of, uh, of that client. The ROI on the AI personalization investment was one is two, seven is to one. our specific, uh, example was, uh, customer segment interested in sustainable products who typically made one small purchase and disappeared after implementing a customized personalized journey for such, uh, customer segment. One, such customer went from single dollar 45 purchase to over 600, uh, dollar purchase over the next one year. That's more than 1200% increase in the customer value. as, as, again, the key lesson is effective AI personalization is not about having technology. It's all about creating genuinely helpful customer experience that deliver value rather than just pushing products. [00:06:51] Phil: Yeah. Yeah. I love hearing those, those stories and, and taking folks through that, um, scaling Mitch phase, right? Like [00:07:00] I. Small companies have a lot of opportunity, you know, different sizes of companies do as well. And, and it's funny that we talk about that because like I, I actually just signed up for, a Google Ads account for, for doing a bit of YouTube advertising and, um, that meant like creating a Google Ads account and spent a lot of time creating my first Ad Instream ad. [00:07:20] And the next day I got a welcome email from, from, from YouTube or, or Google. And I was like, oh, I'm, I'm excited to, to see the personalization from this big AI giant company. And I was so disappointed. Anchor, and I'm sure you've had that experience too, a big company not even doing like simple rule-based automation. [00:07:41] Well, like user signs up, wait 24 hours, did the user create their first ad? If no, then send him the email that Phil got. But since Phil did create his first ad, don't send him an email about creating his first ad check to see what his average CPV was, and if it's greater than this, send an email about that, about improving budget parameters, [00:08:00] like just the simple based behavioral type of segmentation that even Google wasn't doing in, in my welcome email. [00:08:07] Um, it's easy to think like we should be doing, you know, step zero to step five. Some of the bigger players in this space can't even get step one. Well, a lot of the times it's about data and time lag and stuff like that. Right. [00:08:21] Ankur: Yep, you are absolutely right. Uh, to the point of this connect, uh, in YouTube onboarding process, it's a prime example how even tech giants can falter when it comes to implementing basic personalization strategy. Uh, I, I saw this firsthand, uh, at the retail bank in, uh, bank investing $2.3 million in AI loan recommendation engine. [00:08:42] While they're basing onboarding EMS were broken, sending WIL [00:08:45] messages weeks late, or promoting product customers already had. This disconnect cost them 3 million, uh, dollars annually in missed cross-sell opportunities. Uh, and what we found is the three common causes. First is [00:09:00] organization siloing, uh, the in the marketing team and the marketing operation teams, the innovation teams working separately with different goals and metrics. Secondly, legacy, technological depth, uh, their basic system. The legacy system ran outdated platforms make, uh, making simple changes very difficult and complex. While AI projects use modern architecture and they do not talk to each other smoothly. Uh, third is visibility issues. Executive, uh, regularly reviewed AI initiatives, but rarely discussed basic functionality problems, basic journeys, uh, personalizations, which is which they have a lot of gaps. So what we did, we implemented foundation first approach, pausing all AI projects for 120 days to fix the fundamental journeys gaps. This results were remarkable, uh, addressing basic automation, improved adoption by 28%. Generating 4.2 million in additional revenue. When we later implemented AI engine on the Stronger Foundation, the effectiveness increased by 41%. This experience taught [00:10:00] me as successful financial institutions are not dosed with the most, uh, advanced ai, but those who execute flawlessly across all technology levels for simple automation to cutting edge applications. [00:10:11] Phil: Yeah. Yeah, the super great advice, so like [00:10:14] Step by Step Approach to AI Personalization With a Maturity Framework --- [00:10:21] Phil: if we're trying to figure out where to start with like AI implementation for personalization, let's say maybe we have some of the basics in place. The foundation is there are, if this and that rules, simple rule-based automation, hopefully faster than 40, 48 hour sync on, on some of our data to do, behavioral based automation. [00:10:37] Let, let's say the basic is in place there and, and someone is trying to figure out how do we move to AI personalization. Like, what's your recommendation for maybe a first project that balances out the potential feasibility and some of the complexity that comes into play here. [00:10:52] Ankur: Yeah, that's a great question Phil. Uh, I recently advised on one of the regional bank with the rich customer data, but no AI [00:11:00] implementation. We avoided a common trap of starting a overly ambitious project. Instead, we focus what I call high value micro journey using [00:11:09] AI to enhance the mortgage preapproval process. Uh, this was a mortgage banking, uh, form, so. How we started is clustering algorithm that analyze the transaction patterns, pre, pre utilization, savings, behavior, and identifying the customer's likely to be home buying space. This allows us to create personalized, uh, pre-approval communication with tailored rate and terms. Uh, this project really succeeded well because of three reasons. First, it utilized the existing data it already had. Uh, the bank has already, uh, core banking system and core banking data, which, uh, which was underutilized. We don't need to invest anymore in getting more collecting data. So we were utilized. [00:11:53] Uh, so that was the best thing. And we had this data was really of good quality. Second is contain, it had [00:12:00] contained scope focusing on the specific product of a specific mortgage, uh, journey. Instead of, uh, attempting enterprise wide personalization, which always, uh, uh, leads one of the reason to get failures. So we wanna really focus, which will bring the most value to our customers and to organization. And then a combination of these two actually delivered measurable results quickly. Within 60 days, mortgage application rates increased by 21% and the approval to closing conversion improved by 12%. This, for a financial institutions focusing on the any of the mortgage product or any financial services product, I recommend identifying the single high value product journey where the personalization can meaningfully impact both customer experience and business outcomes. This builds organization confidence while establishing the foundation for the broader implementation. [00:12:55] Phil: Very cool. Yeah. If, [00:12:57] Personalization Maturity Framework --- [00:13:09] Phil: if you were to like map out the maturity phases of this approach that you're taking and you're kind of seeing like a variety of clients at different maturity stages there, like what? What does maturity implementation path look like for, for personalization messaging? Maybe, maybe we focus on like the, the post purchase use case. [00:13:18] Um, I tear, tear my mo my maturity path apart here, but like, for me, step one, super basics, simple behavioral rule-based stuff. Assuming that the behaviors are, are coming into your messaging platform fast enough that you can do stuff before 48 hours. Step two to me maybe is we start bringing in personas and use cases and we're having multiple journeys within that, like simple behavioral rule-based stuff. [00:13:46] We're adding like past purchase history, um, use cases for cross sell, going into like different products. If you have multi products. Step three is dynamic content. So the emails that you're sending out or the text messages or the push [00:14:00] notifications, um, you're dynamically changing the content of that based on the persona, the use case. [00:14:06] We're creating performance management systems so people can pick and choose what type of messages they want to like receive. So we're almost like creating content modules, Lego blocks within, uh, the emails that we're sending out. Then we're getting a bit more complex. So like step four is implementing AI in this. [00:14:25] So like step one to three didn't really have AI in any of these. And step four we're introducing propensity models and uplift modeling into our messaging. So thinking about like the next best action, instead of like having rules determine the dynamic content that shows up in the message to the user. [00:14:45] We're letting a machine predict the right next best action to suggest to that person based on a bunch of historical data. And then finally, maybe like a step five is like the marketer is feeding content to a [00:15:00] machine and we're letting that machine fully orchestrate unique messaging journeys for individual people based on the channel they engage with the most and the right time to be sending them that message. [00:15:14] Is that, is that the right way to think about it? Like tear, tear, my maturity phase, uh, apart there incor. [00:15:19] Ankur: No, uh, I, I, I agree up to a certain extent and it definitely is always evolving a whole about your maturity curve. It has to evolve. So, uh, I totally align what the way, your maturity curve. I will share what works best for me and let's very much close what you said. So, I, I always start with the behavioral takers. [00:15:39] That's the simple way to start understanding the customer needs and provide the right nudges and right, uh, direction to your customers. What. How we can serve them better. So we start with a simple rule-based notification when the customer completed their first, uh, account setup and that increased the follow-up, uh, visits or the follow-up, uh, [00:16:00] uh, purchase criteria. Then we can go with the segmentation enhancement. And this is the where you, you point, where you have collected enough first party data and you have combined with your second and third party data, where you can evolve your dev demographic segments to behavior based, uh, microsegments incorporating transactional patterns, digital, uh, engagement. This allows us to create distinct journeys for conservative, uh, conservative savers. I'm taking examples of any financial services from, so, uh, conservative savers who, uh, who, who want to invest in more conservative way. And there are some active, uh, investors who want to, to do major, uh, risk, uh, taking capabilities. [00:16:40] So you have a different journeys for them. Once you are done with the basic, uh, segmentation enhancements, you can move to the content modelization. We, what we suggest is a component based content system with dynamic elements that adopt, uh, based on the, uh, relationship you have built with the customers or specific, uh, financial [00:17:00] needs based on, uh, preference, like, uh, uh, how much frequency you want to communicate, what are the topics you're interested in. [00:17:08] Some customers like frequent communication. Some customers do not want to, uh, get frequent communications. Uh, so you need to respect that, uh, preferences and, and build your content according to, uh, according to their needs. Once we are done with this basic infrastructure of work, then we can move to the predictive personalization, and that's where we come with the, uh, models, machine learning capability to predict likelihood of, uh, specific financial needs. Based on the life events, like if somebody's getting retired or having pks or getting married, they, those, uh, lifetime events will help you to learn your models better and then you can build your next best action approach. Uh, offering relevance from, uh, improving your, so in my example, I was able to improve the relevance by 42% and conversion by 28% by moving into this step four. Uh, and this itself is a good [00:18:00] amount of, uh, personalization we have done. Uh, most of the organizations are not able to complete this, optimize this first four step, and they've moved to [00:18:09] the next step, which is the AI orchestration. But I strongly recommend to, uh, focus first on the top four and then move to AI orchestration where we can deploy machine learning to continuously optimize journeys based on the real time engagement and financial behavior. Uh. Creating truly individualized experience that improve retention and, uh, by 24% and, uh, will improve the share of wallet by 29% for the, for the customer I worked with, uh, recently, what I would, uh, see that measure your each step at each step, measure your, uh, uh, measure how it's getting aligned with your KPIs and the organization goals before you advance to the next step. [00:18:52] Phil: Hmm. Yeah, it's a really good point in this actually, like AB testing or, or creating holdout [00:19:00] tests or some type of incrementality where you can see the net impact of moving from step number two or step number three from the previous step. And maybe that's even like a, an interesting element to introduce to this. [00:19:14] Like at what stage do you implement simple AB testing in some of these like journeys or orchestrations versus. Automating some of those testing frameworks. Like I've seen some tools that allow you to have, like, before you decide to go from like step three to four or or three to two, you can do a simple test and you run it for like three or four months or whatever, and then you see the net incremental dollar impact and you decide to go all in on like the next step. [00:19:44] It makes sense for us to do it. Um, but sometimes some tools have it just baked into the process where, um, we're gonna work with you to get to step four or like, we're gonna jump from step one to step four right away. And everything that we're gonna do in step four is gonna have a [00:20:00] holdout group. There's gonna be a control group in every message that we send, and we're gonna be able to give you the incremental impact of, of, of every message that we're doing in step four. [00:20:11] So you'll see the results along the way. What are your thoughts on like instructing the human based AB tests versus. Adding the automation, uh, testing framework as part of that. [00:20:23] Ankur: Yep. And this, this get asked multiple times and, uh, most of my clients have this common question, uh, when to study v testing. And in my experience, I always recommend timing. Depends, uh, of your v testing depends on the personalization, maturity and resources you have. So I recommend introducing AB testing at the step two of their personalization journey. [00:20:44] Immediately after implementing basic behavior takers, we, we can start by testing subject lines or call to action variation. Uh, in, in our, any of the, uh, email communication we reveal, uh, in the, in one of the [00:21:00] recent, uh, engagement, I realized that after first implementation of AB testing, manual AB testing, uh, we found there's a 31% performance gap between the A and b, uh, variation. And, uh, that helped us to understand a lot about what our customer likes and, uh, do not like and getting the right engagement data. However, we waited till step four, predictive personalization step to implement the automated testing frameworks. Uh, by this point. And the reason, uh, by this point, you want to have sufficient traffic volume to across segments to achieve the statistical significance quickly. Establish, uh, content modules that can be systematically tested, data infrastructure to support, uh, for, uh, so that you can analyze the res, uh, result efficiently and, uh, more effectively. Then by the time cus uh, organizations come with a clear KPIs as well as, uh, initially it's very broad. Now you have a more clear defined general level [00:22:00] KPIs optimized for the customer segmentations. Then, then I move to the automated framework after the step four, uh, where we can use a machine learning to continuously test content variation across different, uh, customer segments, automatic, automatically relocating traffic to the winning variations. This increased conversion, this actually will increase the conversion by, uh, 20, 30% depending on the, uh, what financial services firm you, uh, you're referring to. Uh, but it, but I have seen 37% increase by just doing the manually testing. The key, uh, lesson was starting simple manual test to build a testing culture before investing in automation. Many finish, uh, many organizations, uh, rush to automate before establishing organ organization discipline of hypothesis driven testing leading to sophisticated system that generates insights nobody acts upon. [00:22:58] So first, [00:23:00] focus on the data you're collecting, how you want to engage it, make it sure that KPIs are clear, your data infrastructure is there, content is there. Then you go with the AV variation more in automated way. [00:23:11] Phil: Such a great point there on, on the data foundation piece there. I think that, you know, for a lot of folks listening, and maybe there's like some marketers on the team that were just like, we wanna start doing step five or, or even like step four or three, and the marketing ops team, the marketing technologists knows the state of the data right now. [00:23:32] And even though we could get to that stage from like a tech standpoint, the data that we have in place right now just isn't structured or we don't have enough pieces associated with it. It's not enriched enough to be able to do like step three, step forward or step five. So like. Before any of this, like crazy, like step three or four or five that we talked about, um, there are like these data foundation pieces that, that need to be in place [00:24:00] before it's even an option, right? [00:24:01] Like what, what does the data maturity look like for this? Like when do you know, in your opinion that you're ready for step two to five? [00:24:11] Ankur: You know, I learned the critical importance of data, uh, foundation, really hard way. And, and this happened during my early, uh, early in my career. I was helping one of the wealth management firm, uh, and I was asked to implement the personalization strategies using AI driven and make it happen. Now, uh, despite, uh, this happens all the, all the time, you want everything to be done, uh, right away. [00:24:39] So despite my concerns about the data readiness, I felt the pressure to move quickly. So we launched personalized investment recommendation program using what data we had. The results were really bad. Uh, we, we sent inappropriate retirement planning offers to the client who are still in thirties or suggest a call to the customers [00:25:00] who children have already graduated or then. So we realized so many mistakes and uh, we did end up in the personal journeys and email communication we did, or. Uh, the, the way we want to interact with our customers went really bad. So this painful experience led to me develop what we call a data readiness roadmap. Uh, we took a step back, uh, and focus on fundamentals for our basic triggered communication. [00:25:26] We first ensured 90% of my identity resolution across channels, uh, uh, and clean transaction data to achieve less than 12 hours latency. Before attempting segmentation, we spent six months building customer profiles with consistent behavioral history, standardization of attributes across a retail end of wealth divisions. So we really know what our customer is instead of treating customer from different, different angles. And all this six months of work that patients really paid off. When we reload the personalization PR [00:26:00] program, after a few months, client engagement increased by 43%, and new investment account opened rose by 28%. The valuable lesson I learn and I share with my team all the time is, uh, the quality of data foundation directly determines the ceiling of your personalization capabilities. So, as you said, data foundation is very important to begin. [00:26:23] ​ [00:28:35] Phil: The, the, the MarTech landscape. Uh, like I, I was looking at LinkedIn earlier and um, you're a big fan of Sharon stuff from, from Scott Brinker. We had him on the show a couple years ago. Um, I listen and, and read to to Scott all the time and to, you know, everyone that's seen the landscapes, uh, they're notoriously fragmented. [00:28:54] Right. And I feel like with AI and Gen AI and now AI agents, like [00:29:00] it's just blowing up and it's becoming unwieldy. [00:29:03] Fix the Stack You Have Before You Buy Another Tool --- [00:29:03] Phil: How do you approach integrating AI personalization capabilities with existing tech stacks with your clients without creating more silos or even more of these like tech islands? What are your thoughts there? [00:29:17] Ankur: Yeah. And, uh, we all have faced a similar challenge over the time. Any big organizations, enterprise organization because of multiple reason. One could be merger and acquisitions and, uh. Adding Lot Tech stack, uh, to, for different use cases. Uh, we commonly found, we call it Ty Diagram, Ty Architect, where you have many tools, they are somehow connected to each other with, uh, with, uh, not perfectly working, uh, data. [00:29:48] That data. [00:29:48] is. So in the recent, uh, enga engagement, uh, I, when I was with Salesforce, so one of the CI was advising one of the CMO that how this figety [00:30:00] uh, architecture is disconnected marketing technology and the critical customer data is strapped in the silos. Uh, our ambitious, uh, AI personalization goal seems impossible given this fragmentation. Now, it's very difficult to rip apart everything and recreate the tech stack when you have a. AI and personalization already on, on hook to get implemented. So instead of that, what we built is called a connective tissue strategy, where we started implementing a CDP platform, customer data platform that sat over the existing tech stack creating a unified customer profile without disrupting any of the operations. what we did, our approaches had three phases of it. We created API connections between the, our core banking systems, CRM, and digital banking platform to establish a single customer view. Second, we implemented decision layer, uh, decision engine that could access a unified profile and determine the nest best actions. And the third was [00:31:00] the delivery integration to execute these decisions across channels. While more focusing on the omnichannel experience. This breakthrough, uh, came when we connected our wealth management system to this ecosystem. Previously isolated client information become accessible for personalization, increased investment product cross sell by 34%. The key lesson, uh, was prioritizing integrations over replacement. By creating a data orchestration layer above your existing systems, we can achieve a personalization goal without a risk of, and cost of complete tech overhaul. The, this approach, uh, delivered ROI in almost nine months versus two year plus years of replacement strategy. [00:31:43] We, we might have considered that. [00:31:47] Phil: The orchestration layer that you mentioned there is, is really interesting, like [00:31:51] Everyone Wants to Own the Orchestration Layer --- [00:31:57] Phil: something that's happening in MarTech right now. I, uh, chatted about this with, uh, the episode that dropped today. Um, every MarTech vendor is integrating [00:32:00] AI and agent capabilities right now, and they're like recalibrating their vendor, their value propositions. [00:32:06] Product marketers are rewriting the home pages to include, um, ai, GTM and like AI orchestration, like all the MarTech tools are doing this right now. How do you see the relationship between all these different platform categories like iPASS, marketing automation, customer engagement, CDPs, like all of these major ecosystem players evolving in an AI agent centric world? [00:32:32] When everyone has AI agent features, what are your thoughts there? [00:32:37] Ankur: Yeah, that, that's so true. Uh, if you see a MarTech stack, uh, includes separate platforms for data integration, C-D-P-C-D-M, campaign management, engagement orchestration, and each vendor is rapidly adding AI capabilities, claiming to be a central brain for customer experience, uh, in one of the. [00:33:00] Brainstorming session, uh, within another steering committee, uh, one of our client CTO asked the same question. what we approach is in five years, which of these platform will actually exist as a separate categories, or they're going to merge together, or they're gonna collapse, or, [00:33:15] and that's, things are change and things are changing rapidly. So that led, that led us to develop a strategy, we call it convergence strategy. Rather than betting on individual platform, we focus on creating a flexible architecture that could adapt. As category merged, [00:33:31] we identified three key trends, reshaping the smarttech landscape. First is, as you said, the collapse of the traditional boundaries between IPAs, CDPs, reverse a TL As data, uh, movement has become more commoditized. Second is emergence of AI agents as a new orchestration layer sitting above execution platforms. The third is the shift from tool center approach to the use case centering. So solutions. So we restructured our contracts in [00:34:00] shorter term so that we have a flexibility to change the platform as as frequently as we need to, and we prioritize open API and data portability. [00:34:08] So there will be no tech debt or lab difficulty in adopting of new tools and solutions in our tech stack. Uh, this approach of paid off when the CDP and the marketing automation vendors merged in last couple of years creating integration challenges and called, but the competitors, like other organizations were facing, challenges like how to adapt to this new merge or collapsible CDPs are coming in the market. So keeping your architecture flexible. Uh, more focused on o open APIs. Keep, uh, keep an idea of uh, data operability helps you to. Overcome this, uh, or ever-changing dramatic landscape. [00:34:49] Phil: Very cool. Yeah, I, I like that, that approach there, making sure that you're rethinking your contract. Tracks and setting up the stack in a way that you can be flexible instead of trying to [00:35:00] predict who is gonna win the race to like that, that orchestration layer, figuring out, you know, being flexible and whoever ends up winning that race, whoever ends up buying other players in that race too, right? [00:35:13] Like, um, I feel like one thing that's, that's top of mind, especially for enterprise marketing operations team, is specifically like the complexity that will come in a year or two before there is potential consolidation or before there is like a specific category or tool that is the clear winner and this is the best practice and we need to do it that way. This explosion of AI agents from all these different vendors with their own orchestration logic, like you said, we need to figure out a way to have a layer on top of that, but on top of the execution layer. Like how, what advice do you have for folks that are in this enterprise world right now and need to figure out how, in this meantime, before we know what is the best practice and where we're [00:36:00] going, what do we do with this spaghetti right now of orchestration? [00:36:03] And every tool wants to do AI agent stuff, especially in an enterprise environment. What advice do you have? [00:36:10] Ankur: Uh, again, common pitfall, common challenge. I will give you example. When I was an architect at one of the FINA major financial services firm, we faced the exact challenge. We had implemented AI agents for across multiple functions. Conversational AI for customer service, agent for the fraud detection, another for investment recommendation, and there's a marketing orchestration agent. Initially, this agent's work, operated in silo, uh, but, uh, and they were working fine, but as the use case evolved and they're interacting more with the customers, they need to collaborate now, because that we can give the consistent message across all the platform, in fact, the complexity become more apparent when the wealth management clients started receiving contradictory messages. Our, our marketing agents were promoting high risk investment while our advisory agent were recommending conservative [00:37:00] approach. So this get more confusing for the customers, and this was not a great experience for their customers. So what we did, uh, we developed, we call it, uh, agent, uh, governance framework with three key components. First is centralized knowledge craft that provided single source of truth to all these agents. And, uh, this not just, uh, uh, confined to only two agents, but only kind of, uh, in-person conversation as well. So even the salesperson or the advisor or the h wealth management advisor, they will have access to the same common graph knowledge. Second is priority hierarchy, uh, that determine which agent's decision will take precedence in conflict situations. So we always know which is the most, uh, the best advice we want to give to the customer. Third is the unified customer and tech service that ensure all agents had access to the same real time customer information. This breakthrough came when we implemented, [00:38:00] uh, this meta agent architecture, agents that, uh, orchestrated other agents. This approach, uh, con, uh, helped us a lot to avoid this conflicting actions. Uh, we were able to reduce 87% of, of our conflicting actions and improve cross-functional consistency. Again, the insight I would say is, treating AI orchestration as an enterprise, uh, architecture challenge rather than a vendor management issue. Most of the institutions can succeed this environment by establishing a clear governance models and how agents will interact rather than allowing each vendor AI to operate in isolation. [00:38:39] Ideal Team Structure for Scaling Personalization --- [00:38:39] Phil: What does the ideal team structure look like for you as you're like thinking about advising clients on moving up that maturity scale that, that we talked about of, of personalization and, and introducing AI once they have data foundations in place. Do you think that lifecycle slash marketing remains different teams [00:39:00] than the data team? [00:39:01] Do you have them kind of integrated together? Is it a decentralized or a centralized model? Do you have different pods with a bunch of different, like cross disciplines in there? What are your thoughts on, on team structure to make this thing happen? [00:39:14] Ankur: Yeah. Uh, I actually, uh, tried all structured [00:39:19] and I will share what works best for me. So. As we advance through our personalization maturity journey, I, uh, I call it, uh, you can have a centralized team, as you said, then you can have a decentralized team. And the third one is called hub and spoke model. And I will, uh, we all understand centralized and decentralized, but what works best is hubs spoke. So initially we created a center of excellence, centralized, personalized, uh, team that served all business units, while this improved consistency, but it creates a bottlenecks and distance from the business objectives because they are very focused on their central team. And, and each of the line and teams have a different business goals, which is, uh, which needs custom journeys and custom [00:40:00] interaction. our major breakthrough came when we implemented I call hub and spoke model. We maintained a core data and technology team, the hub centralized that manage our CDP platform, identity resolution, and model infrastructure. Around this hip, we created cross-functional, uh, customer journey parts aligned to specific lifecycle stages. Uh, for example, acquisition or voting growth retention. and each pod includes a product owner, data scientist, contented strategist, technology specialist who collaborate daily. And this, uh, by this acquisition hubs spoke model and each acquisition pod in, uh, increase the new conversion, uh, customer conversion by 28%, while our retention pod reduce attrition by 23%. for personalization to happen correctly, you need both specialized technical expertise and deep business context, which is very difficult to manage at the centralized team because each business function, uh, has a business, different business goals. So what. [00:41:00] Succeeded is maintaining centralized data foundation while embedding analytical talent directly into the business teams where they can develop domain expertise and measure the impact directly. [00:41:11] Phil: Um, yeah, we kind of touch on this a little bit already. Like we hear [00:41:14] Stop Pretending Your Data Is Ready for AI --- [00:41:14] Phil: data quality and data quantity constantly as like two of the. Objections or things that hold back teams on figuring out what is good enough data that actually allows us to do effective AI personalization and, and do stuff that's, that's a bit more complex. [00:41:33] Where do teams often like misdiagnose their data readiness? Like how do you know when you have enough data and good enough data to make that change? [00:41:43] Ankur: most of the times we think data readiness is all about we have good quantity of data and then, and the data is accurate and complete. That that's good enough data, but. But what I have seen, if you don't have a context of why we, why we have [00:42:00] this data and how we are going to utilize this data, data, it's actually help us to define are we really ready for personalization or do we have good enough data? So, uh, what I do, and, uh, it's basically three hidden quality gaps within the data, which you have not identified yet, and which is very critical to measure how much data is really good enough data. First is identifying with the data has a consistency there. Most of the organization have collected data, but they have a, they have, we have discovered the temporal inconsistency, the transactional data was there, but it was collected in different parts of the timeline. We could not able to stitch together and, uh, get insights out of it. So inconsistency of data is, It's one of the bigger challenge. Second is we, uh, we know what customers are interacting, what customers are doing with us, with the purchasing of the Goda transactions or the various engagement data. We know what they're doing, but we don't know [00:43:00] why they're doing, uh, why did they buy, why did they went to our visit? [00:43:04] What are the different channels they visited? Why they switch their preferences. So that why is very important behind the, uh, data, which I was saying the context, uh, why we collected this data. And the third is the, relationship blind blindness. Uh, it's very common in the financial services, uh, uh, when it comes to management of money, the head of household and, uh, not able to connect the other life, uh, other family members into one, [00:43:30] uh, that relationship, blind blindness. Because most of the decisions are collective decisions by the family. And, uh, by making, identifying the head of household, you can provide the better, uh, recommendation or communications we developed a fitness of purpose framework that assess data quality specifically for personalization use. For example, we determine with that effectiveness based offer, personalization requires nine months of consistent behavioral data, 85% of the identity [00:44:00] resolution, across channels and less, less than 48 of data latency. Uh, this breakthrough came when we focus on enriching relationship data, connecting household members within our system. This single improvement, increased personalization effectiveness by 31 30 7%, even though we haven't made any algorithm, algorithm changes. So the good enough data is not about perfection across all attributes, but rather sufficient quality in specific dimension that drive the personalization use case. [00:44:31] Phil: One of the things that holds back a lot of teams, especially enterprise teams, and especially teams in regulated industries, like a lot of your experiences in the financials service sector, we have a lot of folks that work in healthcare and. You know, we're, we're dealing with PHI PIIs very sensitive data, right? [00:44:54] And we don't want AI to send the wrong thing to the wrong person [00:45:00] and share sensitive information with, you know, a machine and, and it gets in the wrong hands, right? Like there's, there's a lot of sensitivity, uh, when it comes to moving up that maturity phase in scaling personalization. And I, [00:45:13] Balancing AI Personalization With Human Context --- [00:45:15] Phil: I wanted to ask you about like, that human touch in there and, and this idea of balancing the human touch with letting the machine orchestrate a lot of those decision making things like where have you seen the most successful balance when you're striking between, uh, algorithms making decisions and having this like human creativity or even like guardrails coming into place and making sure that, you know, the sensitive, uh, information isn't, you know, being shared in the wrong places and we're doing QA and all that stuff. [00:45:46] Maybe chat about that for a bit. [00:45:47] Ankur: Yeah. So, uh, working recently with, uh, one of the asset and wealth management client, we implemented AI personalization engine for high net worth clients [00:46:00] that analyze portfolio performance, market conditions, behavioral patterns to deliver highly personalized communications. The data showed impressive engagement metrics. [00:46:08] Open rates increase by 42%. Client portal logins rose by 37%. Yet in our quarterly client, uh, satisfac satisfaction survey, we saw alarming trend comments like, uh, fields, mechanical fields, uh, missing the personal touch. That appeared frequently and our relationship strength scores declined by 18%. We, when we analyze which communication resonated most deeply, it's combined through AI driven insight with the authentic advisor perspective, consistently, uh, uh, perspective. And that consistently outperforms purely algorithmic content. For example, market volatility alerts that includes brief personal video from the client's advisor had a 3.5 time higher engagement rate than the automated alerts alone. So this is what we call develop a human in the [00:47:00] loop framework, where the clear guidelines for when human involvement is essential. Some major lifestone, uh, my life milestone events like retirement inheritance, that required human touch outreach, uh, investment recommendation over certain thresholds needed advisor review. And that's what you need, human in the loop. So talk to customer, make him comfortable that we are not just algorithm or mechanical well, uh, communication, driving, uh, market volatility, uh, including personal context from advisors, uh, the most successful interaction I have seen. So [00:47:36] Phil: Very cool. [00:47:36] Ankur: yeah, it's a balance between the AI and, uh, how you handle personalization along with the human in the loop. [00:47:42] Phil: Yeah, I've heard that Human in the loop, um, used by, uh, a couple different folks, uh, on the show. It's a nice analogy to, you know, let folks that have those hesitations about I. Um, you know, how, how are we gonna implement AI and all this without forgetting the humans that are gonna be part of the, the [00:48:00] decision making process? [00:48:01] Because it is a journey to get there. We can't like flip on that switch if you're in those early phases of, of that maturity there. Um, I got two last questions for you. Uh, [00:48:12] Predictive Loyalty Engines are Coming to a Martech Stack Near You --- [00:48:17] Phil: wanted to let you chat about emerging tech and, and like some of the bold predictions you have, uh, when we were trying to figure out topics for the show. [00:48:19] Um, uh, I wanna let you chat about behavioral intent predictions, uh, predictive loyalty, and you're also excited about empathetic ai. Uh, so yeah, I'll give you the floor here. Uh, walk us through those. [00:48:31] Ankur: first I would like to talk about predictive Will replace, uh, that will replace the reactive rewards AI system. I. In future, we'll anticipate the customer needs and deliver value before customer even recognize those needs themselves. [00:48:46] Phil: Hmm. [00:48:47] Ankur: In, uh, in the early prototypes, uh, we have seen system can detect subtle, uh, patterns indicating a customer might be considering alternatives proactively and intervene with personalized experience [00:49:00] that can increase retention by 53%. that would I see as a, as a major enhancements going to happen in personalization. And the second is, uh, empathetic AI that will emerge as a key differentiator. Beyond behavioral data, AI will understand emotional context and life circumstances to create truly human-centered, uh, experiences. Our research shows that the system incorporating emotional intelligence, outperform traditional personalization by 72% on long-term loyal loyalty metrics. I'll give you a great example. Uh, we, uh, we created or we did a pilot, uh, test case on healthcare provider that could detect subtle linguistic and behavioral patterns indicating patient anxiety. The system analyzed communication tone over the phone appointment, scheduling patterns of portal usage to identify the patient experiencing healthcare anxiety before that explicitly can expressed [00:50:00] come, from the patient [00:50:00] Phil: Mm-hmm. [00:50:01] Ankur: For these patients, AI can automatically adjust communication frequency, simplified medical information, high anxiety, cases, facilitated human outreach from the care coordinators with before even customer patient ask for it. Uh, I see that would really bring lot value to the customers. [00:50:20] Phil: Super cool. That, that's really exciting. Um, analyzing, uh, the tone of voice in, in those calls and how they're interacting with portals. I feel like my, my Health Gulp providers can, can learn a thing or two from that. And this has been a super fun conversation. I got one last question for you. Um, [00:50:37] Protecting Your Sanity While Scaling Martech --- [00:50:37] Phil: you're obviously a MarTech leader, ad tech customer experience leader, a digital transformation expert. [00:50:42] You're also a father of two and a pickleball fanatic. One question we ask everyone on the show is how do you remain happy and successful in your career? But how do you find balance between all the things you're working on while staying happy? [00:50:54] Ankur: Yeah, so my secret to happiness in MarTech space, I treat my personal life [00:51:00] like a high priority e test that always outperforms my work metrics. Whenever my smart, uh, my smart, uh, watches or smart, uh, devices, buses with campaign alerts, or that happens all the time during the dinner time, I simply ask whether this notification improved my family engagement rate. [00:51:19] Phil: Hmm. [00:51:19] Ankur: is always no. Uh, apart from that, I, I try to declutter my li life over the weekend with, from a technology, uh, device. Techn tech devices keep my phone away, or the technology devices away when I'm playing the pickleball with my, uh, with my friends or family. [00:51:36] Phil: Yeah, definitely better at pickleball when, when you don't have, uh, your phone in your hands there. Uh, yeah, I empathize a lot with, uh, with that. I think, you know, since I went like Solepreneur route and I work for myself, it's, it's easier said than done to like fully turn off work mode and put the phone away, but, um, you know, my daughter doesn't care how busy I am or what's going [00:52:00] on at work and sometimes we're like kicking the ball in, in her playroom and she'll like tell me to put my phone down and I'm just like embarrassed sometimes. [00:52:08] 'cause like she catches me in the act and I'm just like, I don't want to be that dad who grows up with like, you know, the phone always on, always in work mode and. She's a good reminder of like detaching and there's a world outside of work. [00:52:23] Ankur: Yes, I can understand. [00:52:25] Phil: Awesome. I really appreciate your time and course this is super fun. [00:52:28] Um, yeah. Thank you so much for, for being here. [00:52:30] Ankur: Thank you. Thank you.