{
  "video_id": "pcAtJDBO3hw",
  "channel_slug": "openai",
  "channel_handle": "openai",
  "title": "Customer Ignite Talk: Ravneet Shah (CTO, Allica Bank) & OpenAI",
  "duration_seconds": 910.0,
  "url": "https://www.youtube.com/watch?v=pcAtJDBO3hw",
  "upload_date": "",
  "transcript": "[music]\n>> Hi everyone. Very nice to meet you. Um,\nmy name is Clem. I'm part of the\ngo-to-market um, team here at Open AI.\nAnd I work with um, amazing customers\nlike Allica Bank here. And thank you\nRavneet for joining us. So, if you want\nto introduce yourself.\n>> Thanks. Thanks uh, for inviting me here.\nI'm Ravneet, CTO at Allica Bank. Um,\nAllica Bank, if you don't know, so we\nare an established SME bank. Uh, so uh,\nwe uh, have been uh, operating in UK\nsince 2019 when we got a banking\nlicense. And uh, we are building\nspecific lending and business current\naccount and deposit products for SMEs in\nUK. And our unique selling uh,\npoint is like we do it through\ntechnology and relationship banking.\n>> All right, that's great to hear. And\nit's been one of the fastest challenger\nbank uh, growing in the in the UK. And I\nthink like the organization you lead is\nvery special because you're applying AI\nacross every steps of your business from\nuh, product development to relationship\nmanagement like you were saying. And\nalso um,\nlending operations, right? So, um,\ntoday's conversation I think is mainly\nto\ntry to showcase how\nmodern regulated banks operate with AI\nand how they augment um, their team to\nsimply go faster, better, and uh, better\nserve their customers. So, let me ask\nyou that first question.\nAllica scaled quite quickly\nas a digital bank. Um, how are you\nthinking about the AI in this current\nphase of growth, but also in what's\ncoming next?\n>> So, uh, we've been on this journey like\nI think we started back in 2023. So, uh,\nmade a lot of mistakes, tried few\nthings. I think that was the time when\nwe didn't know what to do, what are the\nuse cases, trying to establish where we\ncan fit it. But over the years and I\nthink building from where we started, so\nI think we have a clear idea how we can\nscale and we want to scale it as well.\nI think the first thing what we learned\nis like adoption across the\norganization. A lot of and\nmyself, even though I come from tech\nbackground, there's a lot that we can we\nneed to unlearn before we start learning\nthe new technology. I think the first\nmotto that we had was like to scale this\ntechnology, we need to make sure that\neverybody within the organization is\nadopting, they're changing the way they\noperate, be it in operations,\ndistribution, technology, product,\nfinance, everybody. I think that was\nthe drive that we went through. So we\nincreased the adoption across the\norganization. So last year I think it\nwas like 25% so we the today we have\na median work day of 77%. So that's what\nwhat we are approaching. The second\nthing that was quite key and I think\nit's quite key for any technology\norganization is how we are actually\nbuilding our product.\nSo that is more about our product\nengineering\norganization, our operating model there.\nSo we've been working through what what\nneeds to change to be actually\nensure that we are leveraging AI in the\nbest way. So obviously what works for\nthe rest of the organization\ndoesn't work for the product engineering\nsquad. So we had a very specific\noperating model there so which we looked\ninto and the third part was like our\nproducts itself.\nSo I think the way we have been using\nproduct historically, traditionally\nthrough software engineering\napplications\nis significantly changing. So what we\ncould actually do can not could not do\nwith the traditional machine learning or\neven software applications. I think we\ncan do it with AI. I think that added\nanother layer and I think it's just like\na motor within the\nthe business. We keep on repeating\nwithin the organization. Um as in we\nneed to think differently how we are\nbuilding our applications using agentic\napplication. That's the third third way\nthat we used.\n>> And that that's actually very\ninteresting. I think one of the things\nyou mentioned there is a specific\noperating model, especially the one that\nhas allowed you to I think um do more\nthan like 3,700 deployments last year,\nwhich is quite impressive given the size\nof the team and how you've been working.\nUm\ncan you just tell us a little bit more\nabout like how you've organized the team\nand build the team to foster that\ninnovation, but also allow them to build\nfast?\n>> Sure. So, uh we are not a huge team. So,\nwe have like 100 engineers uh or less\nthan around 200 uh colleagues within the\nproduct engineering team including\nproduct data, everybody.\nUm so, we historically operated in\nusing Spotify model\nuh where we had cross-functional squads\num\nproduct, data, uh software engineers,\nback end and front end and as that's and\ndesigners working together. So, um and\nthat that model worked perfectly well\nfor us uh previously and we've been\nshipping uh\nuh our products using that model. With\nAI, we\nsaw that that needs to change and we've\nbeen observing over the last year as\nwell. So, what we changed is like we had\nlike a blueprint of our squads\npreviously, which we have changed\nsignificantly. We've reduced the size of\nthe squads. We call them squadlets now.\nAnd then the squad structure differs\nfrom one team to another based on the\ncomplexity and the nature of the product\nthat they're building.\nSo, we also uh\nfound out that the hand-offs that were\nrequired previously is no longer needed.\nSo, we believed in the concept of\nT-shaped model. So, having a deep\nspecialization specialization skill, but\nthen you're building your adjacent\nskills. Obviously, it's\nskill that's coming the new term that's\ncoming.\nAnd I think that is something that we\nencourage within the organization. So,\nrather than having back end, front end,\nand separate tested, so we combine the\nrole into one.\nAnd then we had the product role as like\nwe used some in some of the squads we\nused to have like product owners and\nproduct analyst. We combine the two role\nthe two roles into one again. And even\nwe evolved that role\nwhere we have in some squads we have\nproduct representative could be a\nproduct owner or a product\nanalyst, but then we had some squadlets\nwhere we just have product engineer.\nSo, someone who can do product as well\nas engineer at the same time. So, I\nthink that's that's how we are evolving.\nI would still say we are in a journey.\nWe've experimented that in some of like\nvery few squadlets, but that is the kind\nof methodology. But\ntowards the end of the year we are going\nwith\nan objective that probably our product,\nour design, our engineers all engineers\ncan ship the code today, but our product\nand designers will be able to ship the\ncode to production by the end of this\nyear where it it is happening in some of\nthe squad not at the same not to the\nextent that we would like to, but I\nthink we'll see an improvement and\nand over the year. And what works well\nis like everybody is excited about that\nchange.\n>> And thank thank you for that. And I\nthink when we're catching up before um,\nto talk a little bit about like\nthis content and what we're going to\ntalk about like, what one thing that I\nhave loved and I love to hear is, um, I\nthink Arica's a specific approach that\nallows them to scale\nfast be and avoid the governance drug\nthat usually happen with AI because of\nthat, um, pod system that you have when\nyou have those groups where everyone\nthat needs needed for the decision or\npushing the product is in the squad. So,\nthe product can go from creation\nto publishing\nwithout needing to leave the squad\nunless absolutely necessary. And so, it\nallows you to go faster and to iterate\nfaster. And I think that's something we\nneed to remember as well.\nWe need to unlearn, but we also need to\nunstructure the way some of the teams\nhave been structured to be able to\nrebuild them to move faster with that\nnew technology.\nUm,\nwhen you think\nabout another area of use talked about\nabout the landing and underwriting\naspect of the business, um,\nI'm curious how is AI changing that\nwithin your organization right now?\n>> So, that's quite an interesting one. So,\num,\nlending is our core business,\num,\nand\nthe customers that we deal with and the\nwhole lending process is quite complex.\nIt's not automated. So, we deal with\nintroducers, brokers,\num, and within asset finance segment, we\nreceive a lot of applications through\nemails. We have portals, but our\ncustomers or brokers or introducers,\nthey don't like to use that. And even\nwith the broke with the portal itself,\nyou sometime get like half-baked\ninformation or there's missing\ninformation. And I think that's probably\nwhere, uh, rather than just forcing our\nend customers to, uh,\nchange their approach, we changed our\napproach slightly.\nUh, so we've introduced a kind of agent\nwhere, um\ncustomers can send us emails, the\nbrokers can send us emails, but then the\nagent actually look into the information\nthat comes through\nthe email, and they can\num\nfind the missing information, so they\ncan ask more information from the\nbrokers before the information gets fed\ninto our portal. So, I think that\njourney is kind of automated, and by\nusing the combination of deterministic\nand non-deterministic agents, so we've\ngot to a position where some of those\napplications, we've reduced the time to\ndecision to like less than 7 minutes or\n12 minutes. So, that's what we are\nheading towards. I think we use the\ntechnology to actually improve the\nprocess rather than just changing the\nexperience of the customers. There are\nmore use cases that we've been using. I\nthink the the the idea that where we are\ngoing is like we introduced we've been\nintroducing technology over the last few\nyears already, but now where the\nprocesses were manual\ncould not be solved through software\napplications, we are just rethinking\nthem and see like, \"Okay, where can\nagents help?\" Obviously, there has been\nharnesses in in place, but I think\nthere's there's a drive, and we see a\nhuge opportunity there.\n>> And I think it links as well to what was\nsaid right before we joined the stage\naround like the the example of voice and\nbanking, and\nthe idea of like some customers are\nwilling to change, some are not, and so\nit's important to meet the customers\nwhere they are, and to make sure that we\nimprove the overall experience using the\ntechnology that is accessible to us.\nAnd that's what we build the technology\nthat is not accessible to us. I think\nthat's one of the things that has been\ntalked about by everyone is using tools\nlike Codex\nallow you to build technology that was\nnot available before and easily as well.\nThat leads me to just another question\nthat's because you've mentioned\nrelationship banking,\nand I know that it it adds to core of\nErica's business model.\nUm\nwhere did you see AI having an impact in\nthat area as well?\n>> So, we had number of debates internally\num on this as in like how can we use the\ntechnology there? Obviously, there's a\nlot of conversational\num chatbots that we could have\nintroduced, but considering relationship\nbanking was a a unique selling point, we\ndidn't want to change it. Uh we didn't\nwant to replace relationship banking by\na chatbot. Uh\nI mean, we have messaging, but that's a\nkind of core functionality that we need\nto uh that we wanted to bring. So, we uh\ncame up with the idea, and we're still\niterating and developing on that. So,\nrather than replacing relationship bank\nuh relation- relationship managers by AI\nbots, what we wanted to do was like\nactually\nsupport them with the insights and the\ninformation that they need to have about\ntheir customers. I think that's what how\nwe are thinking about it.\n>> Yeah, and I I think it's augmenting the\nemployees so they can do more and focus\non the work that matters. And I think\nit's spending more times\num not necessarily more time, but\nbetter, more quality time with their\ncustomers and where they're trying to\nbuild relationships.\n>> And I think it's also about like better\ncontext as in so they so that they don't\nspend a lot of time understanding and\nbuilding information about their\ncustomers, but rather we provide them\ncontext using AI enough so that they can\ndrive that conversation.\n>> All right. Um\none last question um before we wrap up\nthe session, and I think it's something\nthat is um maybe a bit idealistic, but\nif you're looking ahead the next 12\nmonths, next 24 months, like we all know\nat the crazy pace at which the\ntechnology is evolving.\nUm\nwhere do you see Ali G Bank taking their\nAI journey and where do you see\nAI taking\nthe banking or the lending journey that\nyou're part of?\n>> So we are evolving. We are on a journey,\nI guess, but I think now if you had\nasked me that question 6 months ago, I\nthink we were still experimenting, but I\nthink we have a clear direction like\nfrom product engineering world what way\nwe want to get to. So you mentioned like\nwe had 3,700 deployments last year\nand we are we want to double it, but\nthen doubling it just could be playing\nwith the numbers or gaming the system,\nbut what we actually want to get to is\nlike increase the product increments for\nour customers\num both just customer-facing\nproduct increments and internal-facing\nincrements as well. So that could be\nrelated to risk and compliance and\nsecurity. So I think that's what we are\ntargeting at the moment. So obviously we\nwant to keep the speed of our service.\nWe want to make sure that the quality of\nour service\nstays and improves while we're doing it.\n>> That's great. So better serve the\ncustomer and go faster and being more\nefficient while doing so. Like it\nreminds me a little bit of what we were\nsaying about Open AI at the beginning.\nIt's like better models, more\ncost-efficient, and just building for\nyou. So first, thank you very much for\ncoming here today, but also thank you\nvery much for building on our\ntechnology. Thank you for the feedback\nbecause I think we've talked twice and\nyou've already gave me like two three\npiece of feedback of things that we need\nto get better at or we can help you\nwith. So thank you for that. I\nappreciate it.\n>> [applause]",
  "transcript_chars": 13847,
  "ingested_at": "2026-06-18T10:32:24.113921+00:00",
  "source": "channel",
  "yt_meta": {
    "view_count": null,
    "like_count": null,
    "channel_id": null,
    "categories": null,
    "tags": null
  }
}