Matt Wilson on AI Strategies and Innovations at Alkami

Speaker A: Welcome to the podcast. As a listener, you're invited to claim your comp pass for our next live full day Software Oasis virtual AI event with the world's top B2B AI experts. By attending, you'll join our engaged community of founders and executives from over 200,000 leading B2B global organizations. AI won't take your place, but, uh, a competitor leveraging it to drive results will learn the latest in AI first Thursday of every month. Tap green button@softwareoasis.com or use show notes link to secure your past. You know, one of the fascinating things about Matt Wilson's journey to Alkami is how he pivoted from negotiating contracts at Walmart's IT department to leading operations strategy at Al Kami. And he's seen the winding roads of tech working in digital banking and risk management. And now he's all about AI and automating business outcomes. And here's the kicker. Al Kami has become a powerhouse in digital sales and service, thanks in part to strategic acquisitions that Matt's been a part of. Um, I'm your host, Michael Bernzweig, and this is Software Spotlight. I know a lot of the organizations that are listening in today have, um, executives and uh, even, even founders that have had an interesting journey along the way getting to where they are. And it's not always a straight path to success. So I was hoping you could share a little bit about your personal journey, uh, prior to joining Alchemy and this then a little we can talk a little bit more about, uh, you know, your role at Alchemy and some of the, uh, interesting, uh, journeys along the way after, after arriving.

Speaker B: Yeah, sure. Thanks, Michael. It's not always a straight path. I think you, you nailed that. It was a very winding path for me. I, I went to college as a computer scientist. I, I stopped one year short of graduating with that and moved to management of information systems. And I did that because I love technology, but I loved more on how it applied to the business. I started my first real world gig at, uh, Walmart's IT department negotiating contracts and somehow meandered into the financial space. So I've worked for bbva, I worked for regions ranging from their digital banking sites to running some of their risk management teams. And shortly before Alchemy, I was part of a risk management startup that focused on credit and risk reporting. And so AI really aligns very well with my goal of automating things with business outcomes.

Speaker A: Now I know, uh, clearly the space is one that has a lot of sensitive data and um, so many organizations over the past couple of years have, you know, run different AI pilots and internally and not found success. And they have so many things on the shelf collecting, uh, dust. But what, what is the, um, transition point between organizations that might be listening in that have tried to incorporate AI unsuccessfully versus those that you've seen find success in the space?

Speaker B: Yeah, great question, Michael. I think there are two approaches to this. Uh, one is you can have AI live where your data lives. So if you're running databases and all of your data is in aws, you can, in a somewhat straightforward way, leverage AWS AI. But maybe it's not the tool that you want. The second approach really requires what I'd say is a creative workaround. So some tools, for example, you might want to connect to a system, uh, enterprise knowledge management system. But to do that, you would expose pii if you can find a tool that allows you to put in custom connectors to that data rather than the ones that come out of the box, and AI can help write these connectors. It's not as big of a lift as it sounds. You can effectively purge the data you're concerned with before sending it. That might be as simple as stripping out all digits. So you're removing account numbers, you're removing Social Security numbers, email patterns, address patterns. You can bake all of that into your tool to prevent exposure. And then there's a second layer of security tooling, whether it's netscope, Zscaler, Cloudflare, whatever your flavor of that tool might be at your company, you can write similar rules there that prevents accidental AI exposure. Training will go so far into what can and can't be used. But if you want flexibility to really use whatever AI, uh, tool feels best for you and your company, you're going to have to be a little creative on the data, not even cleanup. Ah, but the data transition and what you're sending to those AI tools.

Speaker A: Interesting. So as far as, um, successful rollouts, it sounds like from a strategy standpoint, really just having the right foundation or the right, um, strategy in place before embarking on one of these projects is important.

Speaker B: 100%. 100%. And you can even take. Our data and marketing product takes an approach where we create a unique identifier for each user that isn't sensitive. And so if you can send your unique identifier with other information that's not sensitive and return results, then you can bump it against the sensitive data at that point. That's probably been Alchemy's most successful use of AI and again, just a creative way to not expose data you're not comfortable with, but still be able to leverage AI in a way that gets the outcomes you're after.

Speaker A: Now, the alchemy solution overall, um, has been around for some time, well before the advent of AI, uh, maybe even into the machine learning days. But, um, can you talk about the solution overall as far as the types of organizations that have been using it and where AI, uh, comes into play with the solution today?

Speaker B: Absolutely. So Alchemy did start as just a digital banking company and that was the only thing that we provided. We've made several acquisitions and we've really pivoted into what we call as a full digital sales and service platform. So digital banking, fairly straightforward. I'm assuming most people listening use either a mobile or desktop app. We are the number one provider for credit unions across the US and we're working to do that for banks. But the rest of digital sales and service really comes in two parts and this is what we refer to as anticipatory banking. So number one, share of wallet is key for any financial institution. Having the ability for people to open accounts online is a pivotal part of that process. And so we made an acquisition with a company called Mantle, and that's part of our servicing platform. And we made another acquisition with a company called Segment that really lets us anticipate a user's needs before they have it. And we do that based on transactional data, not them filling out a survey and saying, oh, I might want to open a certificate of deposit. No, we do analysis and it is machine, uh, learning style AI we run on their account. And so that lets us use data as the oil for the engine to help financial institutions really drive that share of wallet from start to bottom, uh, and including all the ongoing servicing, whether it's balance check transfers, all of that in between.

Speaker A: So for the organizations that are using alchemy today, the current solution, um, what types of organizations see the most benefit from the solution?

Speaker B: Really all financial institutions. I wouldn't say that there's one particular mold that it fits in. As I mentioned, we are the number one credit union provider. We're striving to be that for banks, we see success across all of Alchemy clients, whether they've only selected the digital banking or whether they're going with a full digital sales and service platform. That said, if they go with the full platform, we do see higher success rates for those clients, whether it's on the account opening side, whether it's digital issuance. Once They've opened an account where they're getting their debit card on their iPhone. And so that makes them top of wallet. Right? Where users aren't just moving money or an account becoming stale or really not leveraging the fact that you have a relationship with these folks. It creates a spin, a spin wheel or flywheel, however you want to think about it, where it serves the institution to service these customers in the best way. And it's rewarding for those financial institutions because they ultimately get a bigger share of wallet from each one of those users.

Speaker A: So if I'm sitting in an executive role at one of your clients, what might my world look like prior to using your solution? And what does my world look like, you know, sometime after, you know, uh, integrating and being fully up and running?

Speaker B: Yeah, great question. On, uh, on the digital banking side, we really aim to create a seamless platform. Most people that interact with their digital banking don't realize how many vendors those financial institutions actually have running different things. For example, it would be very rare for your financial institution to be the ones that are actually generating your electronic statements or accepting your remote deposits. We aim to make, and it ranges from about, uh, 35 to 50 third party vendors that these financial institutions have behind the scenes. We aim to make that as seamless as possible. And so before alchemy, a lot of times it would look like totally disparate systems. I open up a new window that's not really well branded, the styles don't match whatsoever. We, uh, try to make it look like a seamless all in one platform. On the other side, specifically for marketing, what it would typically look like is demographic based marketing. Right. So, uh, I'll just use an example, maybe. I see that you're between the ages of 22 and 27 and I think there's a chance you might be looking to buy your first house. So they would just run a targeted demographic, see very low hit rates and very, very little yield on their marketing campaigns, which are expensive to run with our platform. It's targeted and it's targeted based on data, not demographics. All the way from the user and key performance indicators around the user to daily transactional data, which is actually really complex at times to comprehend and understand. Folks may not know what WM M042 is. Well, that's the Walmart in Bentonville.

Speaker A: Right.

Speaker B: And so we piece together the pieces of data and we run targeted marketing campaigns in an anticipatory way, not based on a, uh, cookie cutter style of. I think these people might be interested. No we know based on data that they are very interested or they will be interested shortly. And so rather than marketing teams spending a lot of time trying to put together campaigns, we fully automate that. And so there's one for certificates, there's one for mortgages, there's one for refinances, and that frees up a lot of time dollars and actually yields more dollars to those financial institutions.

Speaker A: So as you going through with a new client and helping them get up and running, what does that process look like? What is that journey?

Speaker B: Yeah, obviously we start, it starts before they sign the contract. Right. And I think any good vendor tries to truly understand the needs and where the customer currently is at. Right? What vendor are you using right now for digital banking? What third parties are you using to do all your integrations? And where do you feel like from a strategy perspective, you're really falling a little bit short today? And that could be marketing, it could be opening new accounts. Maybe, uh, maybe they're actually seeing account loss and they're seeing consolidation and they're worried about a potential merger because things aren't going well. So we try to identify the big strategic targets that that financial institution needs to hit. I would say they're relatively the same for a lot of financial institutions, but they're not always. You have digital only banking brands now and then you have others that still rely heavily on branches and branch expansion on top of their digital presence. And so number one starts with understanding pre sales. Once they sign with Alchemy, we go through a very rigorous scoping process with I would say, a variety of team members. And that is from an executive level. So what are you thinking more strategically all the way down to people who actually have to run the operations at the institution? Because it's important to us. We don't want a financial institution to have to hire more people to support our solution. We want to free up time, not be a burden on them. And so we have to understand things like how do you actually process transactions on the core to how does your customer support team work and work more effectively? And are there partners to us even, Are there, uh, are there potentials for us to bring partners we have to the table you don't have today that can even impact your operations in ways that you weren't imagining even when you

Speaker A: selected Alchemy now in the space, obviously an organization is only as good as their weakest link. So when you're bringing together so many diverse integrations into one final experience for the consumer on the front end, uh, do you Sometimes find areas of non compliance, uh, along that journey that you need to point out to the organization you're working with.

Speaker B: I wouldn't say non compliance, but suboptimal would probably be a good way to say it. Uh, we have. What's nice about Alchemy is we've done several hundred of these integrations at this point. And so for us to see a vendor for the first time is very rare. And we can actually use the data. We've seen the experiences other financial institutions have had and we can recommend best practices for what you currently have today. Separately, we might know that there's a better option that could solve a lot of pain points for you. And whether they're a partner of Alchemy or a non partner, no revenue versus uh, we couldn't care less. We drive with a. We want to create the best experience for the users of the financial institution first and for the people that work at the financial institution second. So we absolutely would recommend best practices. And even if at times that's bringing up a new potential vendor which does require work for the financial institution. But if you have the discussion around the potential benefits and let them weigh their options, a lot of times they do realize this would be a better result for our end users.

Speaker A: And you know, best case, worst case, um, you know, depending upon what the existing infrastructure and integrations and needs are, um, what is the fastest that you see clients come aboard and worst case, what does that look like?

Speaker B: Our shortest, uh, implementations are typically about six months. We can typically move faster than that, but financial institutions typically can't. Uh, as I mentioned earlier, there's a lot of different roles, the training. You can't just implement something and launch it and then not be ready. So they have to understand the money movement team needs to understand how money moves on Alchemy and how that affects their core system, their core banking system that they have. At worst it's usually a year long project. Those are ones that are much higher in complexity and might actually even entail the financial institution is switching out their core platform. And when I say core platform, if you ever go into a branch and you spin around the screen and you see that very old mainframe looking thing that they may have, uh, they are changing out the wiring under the hood and we will do that beside them. Um, and sometimes actually coincide the launch to the new core with the launch on Alchemy, those are much more complex and can take up to a year.

Speaker A: So uh, I guess in wrapping up. So for organizations that are considering Alchemy or a solution in this space, what are the key considerations that they need to uh, have? Top of mind.

Speaker B: Yeah, I think top of mind really has to be what is their strategy for their digital presence and how important is that to them? All of these. Whether you go with Alchemy, another provider, there is a lot of effort required for a lot of team members at that financial institution. And you can either stay with outdated technology or you can get onto something that has great user interfaces, creates a great user experience whether you're on the desktop or mobile, and ultimately can drive more revenue to your organization. And so I think it's a decision point about how much change do they think that their financial institution is willing to handle, weighing that against the ultimate benefit that their users will see from a technology switch.

Speaker A: Well, I really appreciate your taking the time out for this deep dive on uh, the solution and on the space. For a, uh, lot of the individuals listening in this week, I think uh, a lot of these details are something that uh, that they need to get uh, their hands around and I think this is a good starting point. Matt, I'm really intrigued by how you mentioned the creative workarounds for AI at Alkami. I'd love to hear more about your four part framework for AI success, especially in a highly regulated industry like finance. Can you walk us through how you define fine policies and guardrails? Let's dive into that.

Speaker B: Thrilled to be here and thanks for that introduction. So today I wanted to share a practical, effective approach for leveraging AI and this is really based on what we've seen be successful for Alchemy. I'll focus less in this presentation on some of our internal product or some of our external product usage and more on the internal side where we built efficiencies and seen successful business outcomes. From our perspective, this is basically a four part framework that has helped transform approaches across many different teams at Alchemy and that boils down to these four key guidelines. First and foremost, defining the policy and building the guardrails. Second, truly understanding the potential need. Third, choosing the right tool or tools for the job and then targeted training that reinforces success. So let's start where any good AI, uh, success story starts. In a highly regulated industry, which is governance and policy. According to a policy, uh, according to a study From Microsoft, around 75% of users are bringing their own AI to work. So governance is key, especially in the financial space. If you aren't building governance, folks are bringing tools anyway. So the first point is in highly regulated industries we really require A well thought out policy and governance structure. Step one is always involving the right people. That goes from the obvious folks like the chief information security officer, your chief legal officer, to operations and strategy leaders, to even the builders who help build the AI tools that you'll use. Step two for building the right policy is what I call making a baseline of the non negotiables. For example, for companies like Alchemy that work in the fintech space, we have some things that we're absolutely not comfortable with using AI for, especially on a public model, like exposing pii. Step three comes down to understanding the potential use cases before you finalize the policy. This is what I call a reality check. So we've identified a few high value use cases. Can we actually perform any type of AI against those use cases with a policy that we've set? It's better to be overly specific with your policy based on this early data than it is to have a vague guidance structure that really allows for interpretation. The second thing is governance is definitely not a one and done. I think this goes without saying, but AI is constantly evolving at a very rapid pace. And we see the request for tools and use cases match that rapid pace. Your governance structure should really plan accordingly. For example, we have weekly committee meetings. Folks can submit a form at any point in time with use cases to get on that committee meeting. And this should be broader again than just the people building the policy. It should include the tool owners and builders, all the way down to leaders of these individuals who are stakeholders in AI usage and its success. The second point, we really should allow for proof of concepts and experimentation before going full bore. So rather than roll out ChatGPT to your entire enterprise, you probably want to make sure that it's actually usable, especially across the many different roles you might have at your company.

Speaker A: Tell me a bit about how you enforce policies to protect AI data while enabling real experimentation. What are some of the guardrails you put in place?

Speaker B: And keeping a very, very tight loop with those proof of concepts and your governance committee is ultimately what leads to success. The third piece of advice is wherever possible, build guardrails to enforce policy to both protect your AI data and to enable real experimentation. The thing that you should really expect is people will make mistakes. That is one fact of life. You can build protection around that. For example, if you have netscope or Zscaler, you can build in rules that catch potential PI exposure and outright reject files that contain pii like account numbers, Social Security numbers, or anything like that from being uploaded to any AI. System where you are not comfortable with that data being sent. Understanding how and where to leverage AI is of the utmost importance because a cookie code or approach to AI will never see as much value as truly understanding how and where to leverage A.I. uh. My, my view on this is every role in the company can really benefit, but the way they benefit may be vastly different. So just to give a very real world example, if you have a marketing team or tech writers, their needs are likely vastly different from a software engineer, a uh, marketing or a tech writer. They benefit from context. They benefit from style guides, previous articles, product information. A software engineer also benefits from the context, but that context varies vastly from style guidelines. Right? They need code guidelines, they need integration patterns that have been deemed to be safe for use. The way in which you integrate the tool may also be vastly different. So maybe you actually use the same AI provider for both of these teams, but a marketing and tech writer team ultimately probably are okay with a web based ui. If you presented that as an option to your software engineers, you would not get a positive response. They need something baked into their integrated development environment. Understanding the highest value use cases is of the paramount importance and deep exploration, and I do mean deep exploration, is required to make the best decisions for impactful AI use. Before you begin purchasing tools for any part of your organization, you really should understand the needs of every single role across the company. You want to cast the broadest net possible to understand these use cases, and that may be from conducting interviews, asking for survey feedback, or just researching where AI uh has made an impact in an industry similar to yours. The third point is you likely already have advanced AI users in several of your job functions. You should use this to your advantage. If you ask an individual that's not familiar with AI uh how could you benefit from AI in your role? They're not going to give you the best answer because they simply don't know. If you find champions in each role who are already familiar with AI, you can get answers that are less theoretical and more concrete, which lets the actions you take ultimately see real world benefits. It's also critical to have the right tool for the job. So let's talk about how we determine that.

Speaker A: I'd love to hear how you manage external providers when it comes to AI and data safety. How do you ensure they adhere to your standards?

Speaker B: After you truly understand the needs of each role? Weighing the value of each use case versus the cost is the next biggest factor that I look at when determining what is the right AI uh tool for this job. So Using our marketing example, if they have a $200,000 contract with an external provider that helps write content and they're not really doing anything that we deem unsafe, right? Marketing doesn't have account numbers or Social Security numbers. What they're writing is ultimately going to be publicly available. You can save $200,000 in a contract and prevent a three to four week turnaround with content generation by using something as simple as Cloud or ChatGPT. But broad rollouts do have a real cost to your company and you can offset those costs by prioritizing the things where you know you'll see an immediate roi. There's really no value in choosing tools that don't work with your policy. So you have two, uh, options. Choose tools that either fit with your policy and those could be privately hosted in something like aws. Or choose tools that allow you to creatively back into policy. So let me explain what I mean by creatively back into policy and I'll use a non alchemy example. I'll use a bank or a credit union. Let's pretend that I'm the Chief Digital Officer and I've decided I want to figure out, uh, why are people calling my call center? Maybe it's because I want to hire a vendor to do automation. Maybe it's I need more training. Whatever the reason might be, there's a, there's a real world use case where you could take the last months of call center transcripts, put them into something like deep research, and easily identify the types and frequencies of each request. That said, uh, if you've ever called into a call center, you know that it will contain pii. You have to verify yourself. You have provide information before you can check your balance or before you can make a transfer. You can actually use AI to help creatively back into your policy, which may not allow those things to be sent. So for example, you might ask your AI, can you write a program for me that removes all digits from the call transcript? That gets rid of Social Security numbers, account numbers and identifying parts of the address? Maybe there's a potential for email. So you also ask an AI tool, can this program remove any string that looks like an email address from those transcripts? And last but not least, maybe you're worried about including their names. You could ask your team to run a simple database query to get a list of all first names and last names and completely strip those from last month's call transcripts. What you're now left with is are call transcripts that are within policy versus Ones that you wouldn't be able to use previously. Now I can run deep research and identify where can I get the biggest bang for my buck with either training or omnichannel automation tools to do things like check a balance or make a transfer. Last but not least, it is very important to understand tool overlap and prevent A.I. uh, tool sprawl. Obviously this has a cost benefit, but there's several other things that play into that. The cost benefit is obvious.

Speaker A: Tell me a bit about the role of your security team in building these guardrails. How do they contribute to creating a safe A.I. uh, environment?

Speaker B: Most of the vendors in this space work on economies of scale, meaning if you sign up 50 users vs 5,000 users, your per seat cost is usually less. But you don't want to pick a tool that has only value to the 50 versus the 5,000. A less obvious benefit is your security team. I talked earlier about building an allow list or block list style approach really to build the guardrails for what you can and can't do. If you have a hundred tools versus ten, their job becomes infinitely more complex and harder to maintain. Last but not least, uh, the ability for your organization to organically learn, I do believe goes down significantly if all departments are on different tools. So if I have one department that's learning to write agents and another that's not on the same tooling, it's very difficult to look at was successful with one team and translate that success in a different way for a different role and different team. You should really only allow different tools where those tools create obvious value at a role based level for people within your teams. Last but not least, knowing how is not the same as knowing when to use AI uh. So as I've stated previously, a cookie cutter style approach to AI is really only going to yield cookie cutter style results. Most executive teams that I talk with are pushing their organizations as hard as they can to leverage AI. But most of the time this is with zero clear direction on what they mean. I refer to this as the field of nightmares approach rather than the field of dreams. To use a real world non AI example, uh, outside of my career here at Alchemy, I'm a dad of 4am A volunteer soccer coach. All four of my kids play different sports. Running or weight training is probably helpful to every single one of my kids in the sports that they play. But the demands of each sport require vastly different approaches. You wouldn't just expect to buy a basketball goal or a soccer goal and people be better at uh, those sports But a lot of organizations are making those types of decisions when it comes to their AI strategy and investment. There is a great amount of pre work required to understand each role and how they can truly maximize their AI results. A per role strategy with targeted training has to happen. If you build it, they will come like the field of Dreams. If you don't, uh, you'll see a much slower usage. To give an example, at Alchemy we leverage an enterprise search and knowledge tool named Glean. There is an enormous difference in understanding how to search for knowledge in Glean and when you should search. We invested more time and effort in training our customer experience group at Alchemy than several of the other teams that also have access to the tool. And even internally we saw the team that got more training and had agents pre built for specific parts of their job actually had over 300% higher usage than the rest of the organization. This equated to a real world savings of just under two hours a day for over 300 individuals each within our customer experience group. By building specific things that can help, you can break things like linear hiring patterns or even equip your teams with automatic answers as a starting point and with a confident score rather than them starting from scratch on every support ticket.

Speaker A: I'd love to hear how you maintain ongoing training for AI. How do you ensure that your team stays updated with new use cases and techniques?

Speaker B: For an example. The second point is AI doesn't require one time training. It really requires constant reinforcement for maximum results. There will always be new use cases that pop up and having a mechanism to share successes helps permeate that success throughout the rest of the people who hold that role within your organization. AI is always going to follow a bell curve pattern without influence, meaning there's always people who are going to be really good at it. And those that are questioning how do I really use this. Some basic things like offering weekly AI office hours on demand training and regularly monitoring usage with a specific look at individuals with low usage and providing them more training will result in more benefit for your organization and for the folks that are that are working with AI, uh, like before, uh, if you offer role specific training it helps immensely. I always say people are naturally hardwired to use the path of least resistance, but they can't use the path of least resistance so they don't know it exists. A lot of organizations in the fintech space are using what I call the Hansel and Gretel approach, which is we'll leave a few breadcrumbs and Hope they can figure it out. You can't do that. You have to build a core, crystal clear path and I would say go as far as to baking how you use AI into your onboarding and providing on demand training. Not something that was updated a year ago. Last but not least, you really want to encourage organic growth and adoption. Your training team is never going to be as effective as a teammate on each team who understands the demands of the job day in and day out. AI will continue to evolve and some of your best ideas will come from individuals in your organization. Not strategy leaders, not AI leaders. Building a group of internal AI champions is enormously helpful and you can look to them to help guide where we should focus and even where we should train others. You can ask each team to nominate an AI champion and make sure they have time baked into their week to educate and share with others. Last but not least, you should reward and recognize the individuals who help their teammates and and ensure that their leaders are bought in on this being a real part of their role. Not a, uh, not an after gig or a nighttime activity. In conclusion, AI is not about chasing hype or trying to automate everything overnight. It's really about creating the right guard rails, identifying the right opportunities, choosing the right tools and helping people build the habits to use them.

Speaker C: Well.

Speaker B: The organizations that win with AI won't be the ones that move the fastest with zero direction. They'll be the ones that are intentional, practical and consistent. If we do that well, AI, uh, becomes more than a tool. It really becomes a lever for scale, efficiency and better business outcomes.

Speaker A: If you have questions for Matt related to this presentation, you have a Q and a box right in front of you. Just type in your questions and we'll do our best to get to as many of the answers to those right at the top of the hour on our next Q and A panel. Now here's the live panel from our recent Software Oasis AI Summit where Matt joined other industry leaders to answer real questions from the audience. It's always fascinating to hear these insights straight from the room. Okay, I hope everyone has been enjoying the event so far. For everyone that's been with since 8:00am this morning, uh, I hope uh, everyone's having a great time and picking up quite a bit of knowledge. I know we've started off the, the event with uh, well over. We had 1083 people early in the morning and I see a lot of, a lot of you are still troopers hanging out with us and uh, so I just want to introduce this next panel. Um, we have, um, starting up at the top of the screen we have Josh Haas with Bubble. He's the co founder and co CEO. Or is that founder and co CEO Josh?

Speaker C: Yeah. Well, I founded it with a partner and we built the company together.

Speaker A: There you go. Okay. And Matt Wilson, who's the operations, strategy and delivery lead over at Alchemy Technology. Welcome, Matt.

Speaker B: Good to be here. Thank you, Mike.

Speaker A: Finally we have Ethan Giffen. Ethan is the founder and CEO over at Groove Commerce. Welcome, Ethan.

Speaker D: Thank you, Mike. Excited to be here today with everybody.

Speaker A: Same here. And that makes all of the speakers you've heard over this last hour on this panel. So let's get right into it. I want to see if we can cover as many questions as we can get through as we have quite a few questions that came in from the, uh, audience overall. So let me start with the first question here. That came in for Josh. And that question came from Portland, Oregon, from Lincoln. And uh, Lincoln is asking, how do you coach founders who want to add AI to their product to be honest about whether it's truly core value or just a marketing checkbox?

Speaker C: Yeah, great question, Lincoln, because everyone is adding AI to their products and some of it is just a marketing checkbox. Uh, I think the thing to keep in mind and the thing I tell founders is if it is just a marketing checkbox, it's not going to move the needle that much. You win as a founder by focusing on core value. Right. And uh, sometimes that means not being super AI centric. Like maybe your role in the AI ecosystem is to provide an API or an interface or something that other AI agents can call into. And you don't have to be the AI expert. You can be providing a valuable resource for AI. Sometimes your role in the ecosystem is to be that AI expert. And I think being honest when yourself about which it is, helps keep focused on how do you actually provide value to customers. Because building customer value is ultimately what you're here for as a founder.

Speaker A: Fantastic. And the next question came in from, uh, this is for Ethan, and it came in from Wyatt, who's just outside of the city in Cleveland, Ohio. And Wyatt is asking, when B2B leaders say we want to feel like Amazon, what is one painful checkout or catalog moment you try to fix first so buyers actually notice the difference?

Speaker D: Well, yeah, I think, um, you know, I think the first thing when someone says that to me, I try to say, first off, you know, we want to be like Amazon, but you can't be Amazon because Amazon has more innovation in an hour than most B2B companies will have in 10 years. Uh, you know, in terms of manufacturing and distribution, um, and so, but what, what often that means is, you know, they want to figure out how do we allow, make it easy for our customers to buy from us. Right. And so when I work with an organization, it is trying to understand are there one, two or three things that we could quickly accomplish that make it easier for your customers to buy from you. And so creating less friction, um, is the name of that game. So it could be um, creating an online system, you know, an online e commerce system for an organization for your customers to buy from you. Or if you have an existing system, it could be removing friction from that process

Speaker A: makes a lot of sense and I think that that is great advice. It's uh, a challenge that so many organizations in the B2B space are, are working through. And I think a lot of organizations um, have seen all of the slick uh, checkouts and solutions in the B2C space. And I think a lot of stakeholders uh, are asking for that same kind of functionality. Um, Matt, the um, next question came in for you and it's from, let's see here from Charlotte, North Carolina, from Isaac. And Isaac, um, is asking when you draft AI governance for a beck, what's the single clearest never do this example you use so employees immediately grasp the risk line around personally identifiable information.

Speaker B: Yeah, great question, Isaac. Um, I would say your AI policy needs to align with your general security policy. So if I was to use a very concrete example is you have a policy that you don't email account numbers or tax identifiers. That is a key indicator that that should be part of your AI policy. And you can extend that same sentiment, whether it's Google Drive, whether it's how you share documents internally, they have to be encrypted. Those same guidelines absolutely have to apply to your AI policy. You can try to be creative and strip PI, uh out, but from a baseline in an AI policy, it needs to match the rest of your governance structure.

Speaker A: Uh, this next question comes uh, to us from partway around the world. This is from Australia, from uh, Sydney. And uh, this is uh, from Amojan. Uh, and this is for Josh. So Josh, Amojen is asking, um, as you think about Human in the Loop at Bubble, what guardrails keep you from pushing too much autonomy to agents and accidentally harming trust creator workflows?

Speaker C: Yeah, that is a fantastic question, Red, because it's speed versus trust. And the way we think about it is what are Irreversible destructive actions. Right. Like what's something that destroys data? What's something that involves something, you know, going to a customer, an email being sent out, you know, a transaction that would be hard to unwind. Right. And those are the starting points. You start putting guardrails around because that's where it has to be. Right. And once you have guardrails around the uh, sort of irreversible actions that creates a safe sandbox for AI to play in. And that doesn't mean you can ignore what happens in the sandbox. You need some transparency, you need human monitoring and oversight to see if it's actually creating value. Because if you're left with like a mess of sort of indescribable, um, uncomprehensible AI created work product, um, that no one has their head wrap around doesn't help you. Right. Um, so you need to pay attention to it. But it's those hard guardrails that allow you to really protect, trust and know where to you have to loop a human in.

Speaker A: Makes a lot of sense. And the next question came in for Ethan and we're, we're back to the U.S. this is from um, Leah who's in looks uh, like right, right in Philadelphia proper, Pennsylvania. And uh, Leah's question is how do you convince sales teams that a more self serve commerce experience won't erase the relationship but will actually surface better qualified engagements?

Speaker D: Leah, that's a fantastic question from the city, City of Brotherly love. Um, fantastic question. Um, and that is I have spoken at national sales meetings for sales teams and you see the looks of despair and nervousness on their faces when you begin to talk about online buyer portals. And so it is something that requires alignment from the top down sales leadership, the CEO, um, coo, um, it needs to be kind of driven from the top down in terms of that. And what they need to understand is the less time they spend water taking and the more time that they are able to spend actually selling and relationship building it will help them to increase their sales. Do they have long tail accounts that's in their book of business that they just can't get to because they're so busy taking orders for repeat things that your customers just want to come and be able to buy Those things at 10 o' clock at night or on a Sunday morning or whenever they feel like it. And so it's making sales be a part of the process from day one and actually piloting these systems and including the salespeople and the customers in those conversations as you're starting to, you know, get out of the development phase and into the pilot phase to kind of build that. You've got to build trust there because they are not going to trust you. So you've got to put a lot of work into that, from the top of the organization to the bottom, and pay attention to those questions and concerns that they may have.

Speaker A: This next question comes in from, um, San Jose, California, from Diego Format, and I love this question. A couple of events ago, for those of you who have been following along, we had our event on, uh, keeping AI projects from dying in it. And um, so many organizations have, um, launched all kinds of initiatives over the last year or two and many of them have never made it to production. And I think this question is great. So, uh, Matt, what, uh, Diego's asking, uh, if an executive wants to buy AI tools before mapping use cases, what quick exercise do you walk them through so they see why that's a field of nightmares related to risk?

Speaker B: Yeah, that's a fantastic question. Uh, and I do think it's a field of nightmares. To use an overly simple analogy. You can't just buy a weight room and expect people to work out or expect results. They have to know how to train properly. They have to see positive benefit. Uh, I look at any tooling purchase and AI is no different as an ROI play. And so if you don't know the potential roi, there's no other tool that people would typically invest in. I know a lot of executives want to buy A.I. uh, first and figure out the use cases later. But if you do that, the chances of aligning a good tool are totally incorrect. So I like to be a historian at times. I'm sure we all have some history in our organizations where you can point to failed rollouts and you can point to things that did not equate to roi. It may not include AI in that software, but it's the same premise. And so if you have any of those past examples, you can certainly use those. And I would go further and say oftentimes a small proof of concept with AI, even if it's not wired in a system systems, if it's wired from output from those systems before you go through full, full bore and wire it, uh, up, proving the ability for the AI to do what you want to ultimately return value is the goal. And so that's where I would typically focus.

Speaker A: Great advice there. Um, Josh, this next question came in for you, um, from Alpharetta, Georgia. So I believe that's right outside of Atlanta and, and this is from Sienna and Sienna's asking, looking back, what is one pricing experiment around AI features you wish you had run earlier because it would have clarified customer willingness to pay?

Speaker C: That's a great question, Sienna. Um, we've had a lot of experience with pricing at Bubble, both AI and non AI and we've actually had some pretty rough rollouts because we've had to start switch pricing model midway through the business which is always a very, very tough thing to do because when you change your pricing model it impacts customers differentially which, which is a tough situation to be in. So I'm a big believer in always getting your pricing thought through upfront. With AI in particular, there's sort of three models right now. There's the like, pay for token model, there's the like pay um, you know, per user sort of fixed subscription model and then there's the sort of higher, like a virtual user model where the mental model you're giving your customers they're paying for like a part time employee and the AI is the employee. And I don't think any of the theme models are correct but like for all cases, but they are better or worse fits for your individual uh, product because it really depends on the price point your customers are coming in, uh, at and the mental model of how they consume what you're doing. So I always advise a highly experimental approach where um, you take your pricing and show it to real customers and that can be sensitive. Right? But it's much better to get that honest feedback um, on a prototype or mock up of your privacy page than post launch. So that would be my advice that

Speaker A: makes uh, a lot of sense and nothing like experience, uh, to be a great teacher. Um, Ethan, this next question comes in from um, Ottawa, Canada and it's from Elise. Um, Elise's question is if a manufacturer has messy legacy pricing, what's the minimum data cleanup they should tackle before trying to layer AI recommendations on top of their catalog?

Speaker D: Yeah, I think um, another fantastic question, I think um, figuring out like that's a deep question in terms of what exactly is wrong with the pricing or what's incorrect or what could be optimized with that. Um, sometimes there is just some level setting that needs to happen organizationally to say we need to simplify our pricing because over years, over the years, especially in legacy organizations, you can add all these rule after rule after rule and sometimes you almost need to kind of figure out how to, how to just wipe the slate and start fresh, uh, and figure out um, how to do that. Um, because you might have customers that are grandfathered into buckets and other areas. That's one of the catalog. Generally the catalog in the ERP and the sales question that I just previously had are the three kind of core areas that are significant challenges within these projects. Not necessarily the technology, it's the internal alignment on that. So figuring out how do you simplify the catalog, um, and um, simplify that pricing. That's generally the first, um, the first step. If you can get past that internally, um, then you can get into, hey, we can leverage AI to make better product descriptions or leverage AI to search through product spec sheets or CAD files or whatever we're going to do with that.

Speaker A: Makes a lot of sense. And I think uh, data cleanup is an important part to uh, transitioning from any system to another system. And obviously not uh, creating what you have in a new system is uh, sometimes a great, great portion of the rollout, figuring out what it is you need. So um, the next question came in from Matt and this is from, hopefully I'm pronouncing your name correctly. From Calgary, Canada, I have Aliyah. And um, Aaliyah's question is when you compare teams that had role specific training versus generic AI training, what behavioral differences show up first in their day to day work?

Speaker B: Yeah, great question. And I love the aspect of what behavioral differences because that's what I really think proper enablement does. Without proper enablement you're just given a tool and expected to figure out how is that tool going to be beneficial. For me, with proper training and enablement at a per role level you can be very thoughtful and considerate about what are the pain points of this person's particular job that AI does a good job of solving. And if you're able to deliver prescriptive level enablement, you do re pattern people's behavior because they see immediate benefit. So for example, if uh, I'm an engineer and I'm just told I can use AI, that is a very, very different concept from saying here is how you can benefit from AI is people are always going to find the path of least resistance. And if you don't make that path clear for them, you're just hoping that they find it and take it. If you establish the path for them and you make it clear how they can benefit, you'll have similar results which we had, which was uh, 3x, a little over 3x. The difference in time saved just on one tool alone

Speaker A: makes a lot of sense. Um, this next question came in for Josh from uh, looks like Sebastian is in Just outside of Austin, Texas. So um, Josh, uh, Sebastian's question is what's your advice to non technical founders who feel behind on AI? Where should they invest the next 90 or so days so they can talk credibly with both customers and engineers?

Speaker C: Great question Sebastian. Bubble actually focuses on serving non technical founders. So uh, I've actually had more experience working with them than technical founders. And what uh, I found is first of all as a non technical founder, don't sell yourself short. Like a lot of people who classify themselves as non technical are actually pretty, pretty savvy at this point from working in the tech industry. So I always encourage people to get hands on at the technical level. That makes sense to you, right? M. So for instance, I think trying to have AI build a prototype of what you're working on is a no brainer move to do for any non technical person. Even if you have a team executing against it. Build a prototype yourself using tools. Right. And going from prototype to production often needs engineering support. That's kind of the problem that we're trying to solve for at Bubble. But uh, even the process of building the prototype will give you a lot of intuition for what those tools are capable of. And it's always better to enter those conversations, you know, with hands on experience rather than theoretical understandings of AI.

Speaker A: And I think that's a great perspective of experience and really interesting to hear. Um, the amount of experience with non technical founders I think that's very comforting for most to hear because uh, if you listen to the news too much, you think that everybody uh, knows AI inside Node and is super technical. So really fantastic. This next question is for Ethan and it comes to us from looks, uh, like just outside of Nashville, Tennessee from Cooper. And uh, the question is this, how do you handle, um, let me see if I can. Okay, so I think what he's asking is what, what metrics in portal usage among clients tell you that buyers are genuinely shifting behavior versus just logging in once and going back to what they used to do?

Speaker D: Yeah, um, another great question. Um, and so I, we have about a dozen different core metrics that we track in terms of B2B and we do track in, we track login frequency. So we're looking not only for the first login but the repeat login. We're looking at time between orders. Um, and so that you know, there are um, you want to start tracking um, metrics like that so you can understand um, how often people are logging in and how often they're buying how many SKUs they might have in their cart, how many total items they have in the cart, what your average order value is. And then we're trying to build game plans in order to increase those. Generally there's some friction involved if it's really low. Often, um, one of the most unutilized things in B2B E commerce portals is email marketing automation. So, um, we leverage tools like Klaviyo, um, to create, um, the same tactics that you might use in direct to consumer, except, uh, focused on B2B buyers. So, you know, your abandoned cart, instead of doing that, say within an hour, which might be a direct consumer best practice, uh, it might take somebody all day, um, you know, to uh, build a cart with, you know, 5,000, 10,000, $50,000 worth of, worth of items in it. And so it might take them all day to do that. So we're gonna fire that abandoned cart off 24 hours, um, later and not, not the same day. So it's, it's figuring out what, what the, you know, how your shoppers buy and then aligning core metrics to that and then marketing strategies to basically remind them to come back and reorder.

Speaker A: Okay, well, I want to try to sneak in one, One more question before this, uh, Q and A panel ends. Matt, I'm going to throw this last question to you. Um, Grayson from Jacksonville, Florida asks, what advice do you give CISOs who are overwhelmed by AI tool sprawl and need to cut back without killing momentum for advanced users?

Speaker B: Yeah, really, really good question. I think there's two aspects I would look at first. The first aspect I would look at is usage. And so with tool sprawl, it usually means there's one part of the organization using a tool, another part using another tool. Those are usually not an even ratios. And sometimes those tools were actually built and almost are bought in a proof of concept manner. And so are they actively being used and how many people are using them? I think that that's key to understand. The second thing I would look at is overlap. So if you talk about code generation, we could take up another 25 minutes listing the possible tools that you could buy for engineering to use. Uh, so it's understanding overlap. And then once you have that picture, it becomes usually a little bit more clear which tool is actually being used, maybe which one's not. And if you understand overlap, you can get feedback from people who are using, let's say, two competing tools tools and have them have a direct conversation because usually there is a winner between the two. But it may have just, uh, A contract was signed sooner because someone, someone had the idea initially. So there's a, there's a good chance to revisit in any tool sprawl to make sure is it actively being used. What overlap do we have? And out of the overlapping tools, which one's actually the most effective for our organization that helps any CISO with governance and uh, I promise you, your CFO would also appreciate that when it comes to renegotiating contracts and getting better deals on that front.

Speaker A: Fantastic. Well, Josh, Matt, Ethan, really appreciate the fantastic presentations at the event today. And for joining on this live Q and A panel, I want to keep everybody on track for um, Melissa Jenner's uh, presentation coming right up. Uh, she's with AKFO and she'll be speaking on Designing Mid Career Navigation. Our next panel is coming up, uh, just a little bit later in the event, uh, right after 3:00'. Clock. And uh, once again thank everyone on the panel for joining today. Well, that was a deep dive into AI strategies with Matt Wilson from Alkami. I really appreciated how he explained the importance of governance and the creative ways Alkam uses AI to drive digital banking. It was really great having you on, Matt. If you enjoyed the show, don't forget to subscribe to Software Spotlight for more conversations like this. Thanks for listening. Thanks for tuning in to the Software Oasis Podcast network. With more than 200,000 listeners each month, you're invited to join our engaged community of B2B founders and executives at our next live full day Software Oasis Virtual AI event with the world's top B2B AI experts. Reserve your compass today. Get ahead with real AI strategies. Join the first Thursday of every month. Tap green button@softwareoasis.com or use show Notes link to claim your spot today.