Tech Exec Talks - Episode 1
===
Kulvinder: [00:00:00] Welcome to Tech Exec Talks, where we sit down with leaders of the tech industry to discuss their experiences, perspectives, and vision for the future of technology. This episode's talking all about data strategy. Every organization creates data, sometimes knowingly, sometimes unknowingly, but the desire to utilize that data or use it better impacts every organization at some point in their evolution.
Kulvinder: So where do you start? That's the subject for today. That's the question we're gonna be answering. So I'm Kulvinder, I'm your host. And this episode we'll be talking to Jon Cooke. So Jon is the owner and founder of Dataception, a data analytics and data product company, and the creator of the data product pyramid, which is an adaptive data product operating model. He's a data product storyteller and technology specialist, and he spent the last 30 years delivering data analytics and product capabilities and marketplaces for analytic use cases across many companies. So [00:01:00] Jon, thank you for joining me today. How are you?
Jon: Oh, thanks Kulvinder. Thanks a lot for having me on the podcast. Yeah I'm great. I'm certainly looking forward talking around data in this episode. So yeah, thanks for having me.
Kulvinder: Okay, let's get into it. Let's start let's talk about all things data. So I guess for our listeners and also for me I'm of curious as well how'd you get into data as an area to specialize in and why was it interesting?
Jon: Yeah, no it's been a very, quite an interesting, journey for me. It's like over the 30 years, I graduated in the mid nineties. Actually worked for a research company in inside British Aerospace doing virtual reality, believe it or not. But then Went into gaming and then sort really, started getting around getting involved in message orientated middleware, low latency messaging.
Jon: Did a lot of service orientated architecture on, in the early two thousands, low latency web services, that sort of stuff. And it was interesting again, very data orientated, but much more kind of around the software software space. And then I got into investment banking in mid two thousands.
Jon: And I was sort of front office trading and risk models and all that type of stuff, which is. Hugely data and analytics heavy, doing Monte Carlo simulations on a billion risk points and all that type of stuff, and low latency market data messaging [00:02:00] very much involved in around using data to actually drive analytics for investment banking type portfolios.
Jon: I then moved from there , got into a consultancy and also when start looking at big data, so Hadoop was like 2010 Hadoop and 11 that was around, and we're trying to build risk models and risk capability. On top of Hadoop and big data. So again, massive amounts of data, lots of low latency, that type of stuff.
Jon: Ended up building a practice in data practice in the consultancy and then moved on from there to work to build data and AI practice in pwc. So got much more around into the AI and analytics side of it, doing lots of data strategy. Data, governance, that type of stuff, driving business initiatives with data.
Jon: I left there and then went for Databricks, who everyone, I'm sure everyone knows who they are fantastic company. Was there for very early days of 250 people I think it was. So I was in, in EMEA and built out the solution architecture team there. And yeah, getting involved in AI and data, of course that's their bread and butter.
Jon: And I left there in sort of 2019 and formed my own company. Cause I, I thought there was. Databricks and these other companies are really good. But there was a space in the middle around linking business data and technology really driving [00:03:00] that kind of product mindset. And then I founded the company in 2019 Dataception to do that.
Jon: And that's exactly what I've been doing for the last sort of four years. It's very much that kind of data product, that product mindset with data and analytics in the business. Yeah, it’s been an underlying theme all the way through my career, but very much a focus in probably the last fifteen years on data itself.
Kulvinder: Fantastic. And that's probably a nice segue into, what for you constitutes a good and successful data strategy. What are the key elements of that, would you say?
Jon: Yeah, no, it's a great question. So I think I've been involved in many data strategies and many data initiative, and I think there's two or three key pieces. There's the specifics and general. And the specifics are can you can data and analytics now, and again I tried to use the terms of data and analytics, cause data itself is like the fuel, but the analytics is really the vehicle for driving this sort of stuff.
Jon: So I'd see the the data and analytics strategy really been around specifics, how the company does it and are there specific initiative that data needs, needs to drive. So I'm working with one customer, they're building a new analytics app. They're trying to put out the 4 million customers, have they got the data to, to drive that and actually make that successful.
Jon: [00:04:00] So there, there's some specifics that need to be addressed, but then there's the more sort of general capability, and split those into kind of two areas. There's the transactional plumbing, are your data processes. Fit for purpose, for doing things like, your reg reporting for updating your sub-ledger to doing your risk models, to doing your, your sales forecasting. good hygiene in those sort of things. And then there's a kind of the really interesting bit around embedding data and analytics into the lines of business and being what I call, data driven or scientific driven, scientific methods, experimentation, testing measuring with customers, prototyping and that sort of stuff.
Jon: Being really sort algorithmic. Precision making is, That to me is really, in inside the lines of business where I think that, that's one of the big tick points
Kulvinder: And based on your experience, have you have you had to go into a company where there's absolutely zero, they have nothing. So obviously they have data sets clearly, but they have the desire to do something with the data, but they don't know where to start.
Kulvinder: Is that something that you can, you've done and you can help with?
Jon: Yeah, absolutely. So I think one of the, one of the things I've, noticed over the last sort of 30 years getting the [00:05:00] business to really understand about algorithmic decision making and defining the requirements, based on what they want to do. It's actually really tricky. , we felt, we in data and tech, we're notorious for using Bud Buzzwords and coming up with methodologies and business requirements documents and stuff like that.
Jon: And half the time the business was left scratching their head going, actually, what does this actually mean? How do I actually orientate this? Or, the business are there, saying this is what I do day to day and I don't really wanna go outside of what I want to do. So that's another big thing.
Jon: And it's all sorts of spectrums, right? So you get large companies where you got pockets of that. You get small startups where they don't how to actually articulate the the business requirements, that sort of stuff. So yeah, absolutely this is this, these are the key things that we need to solve.
Jon: And for me it's basically helping, I Data and analytics, people helping, driving the business, driving those kind of outcomes. Value frameworks, basically helping to articulate the business requirements in data and analytics terms, which are easy and simple and deliverable for the business.
Jon: So it's, yeah, it's really trying to, I see, some being sort of data leaders being not Sherpas now, people carry kind of stuff, but actually being like being guides.
Kulvinder: That's a great that's a great example you've given there and I think, [00:06:00] would you say that's all equally as true where you've got let's say there's an established data strategy for whatever reason isn't quite working or isn't yielding the results that certainly the business would hope for.
Kulvinder: Do you take a different approach to that or is it broadly similar?
Jon: I would just look at a data strategy and, fundamentally has it got there's a couple of things that, that it needs. First of all, is there a, is it a is there a product mindset around that? Is there, were you building analytics? Actually as a product management process.
Jon: So fundamentally, are you solving the business problem? Do you have a customer? Do you actually, are you building the kind of the data and the analytics, the model and stuff, and in a product life cycle type stuff? So that's my first looking at it. And the second bit is really, is there a culture of, algorithmic decision making?
Jon: Scientific methods. Is that there already? So data strategies are great. I've seen lots. You will build a platform, we'll build infrastructure, we'll get all our data sets together, we'll classify it, we'll get taxonomies and great. It's hang on a minute. How does that actually drive the business strategy?
Jon: Does/is the last major initiative that happened that the data was supposed to support, was it successful because of it or was it a hindrance? Does it take three months to spin up a, an experiment [00:07:00] environment to actually allow people to start testing before they, they can actually Go go with the business strategy, to set it out if it's gonna work.
Jon: Is that all that stuff there? Is it frictionless? Is it like low friction? Can it, can people can operate and move in, in a kind of fluid way? So these are the things I look at the data strategy.
Kulvinder: Yeah, I was gonna ask, following up on that with how important would you say the culture of the company is in that journey of ensuring that their data strategy and their approach to it. There, there is that alignment there.
Kulvinder: I guess
Jon: Absolutely. Again it comes back to a couple of things for. Do they support scientific methods and algorithmic decision making? That's the first thing. Cause there's no point having a great, I can build you a great data analytics capibility unless it's capability system. But if you, they're gonna ignore all the results and not allow people to, to actually experiment and test, then it's it's not, you're not gonna get anywhere.
Jon: So that's the first thing. The second thing is really how receptive are they to business cases in terms of actually driving change? So if I say, actually you're not supporting this particular business initiative because x, y, and z I can change that and save you 5 million pounds a year or whatever, but you need to change this, and [00:08:00] this, and it's some structural changes.
Jon: Are they amenable to do that? Cause again, you can say, look, I can completely turn your business around, but if they're prepared to actually do it, to execute, to action, that those sorts of things, then that's another kind of big, a bit of a, thing we need to pull a thread on, and also of how do they think about data fundamentally? Do they treat it as something that's a, a competitive advantage? Is it or is it just plumbing that just needs to be fixed when it breaks, so again, to me is part of that cultural aspect that you need to reassess
Kulvinder: Excellent. I I, and I think I think the interesting thing for me I guess it's, it is like a lot of things, right?
Kulvinder: With technology you can easily go out and procure the software or the means to be able to fix a need. But until you can articulate that need and , the whole business around that, it's gonna be tricky to execute on. My, my go-to is always CRM in that regard. Where back in the day was we need a CRM system.
Kulvinder: What do you need it to do? can't quite articulate that. So feels quite akin to that, so in terms of that kind of alignment and prioritization of those data initiatives, sounds like it comes back to. What do you want to do with it? Like you say, do you wanna realize efficiencies, et [00:09:00] cetera?
Kulvinder: Do you have any kind of go-to strategies that you employ to help companies understand what those data initiatives or business needs are that would lead to those data initiatives?
Jon: Absolutely. That's a great, great question. So I think the first thing I look at and to see is that have they got a sense of kind of what we call a value framework? Have they got an idea of actually what's valuable to them in terms of actually they what they wanna do? And that's not a data thing.
Jon: That's a kind of a business initiative. Seen lots of business initiatives which have been kicked off, big Bang, massive programmes spinning up 30 people, but they actually haven't got a clear outcome or a clear business value outcome. So I, that's the first thing I look at, is basically can they, can you articulate a value.
Jon: Fundamentally, do they know what, if you've got two execs, they understand the same version of value, can they prioritize stuff based on that kind of value? The second bit is can they, do they have a a process for, testing and iteration without, again, spinning up big bang, do small incremental delivery.
Jon: So again, they wanna build a, a an analytics app for their customers. Can we prototype it? Can we test it out with a friendly set of customers? Can we measure it? Can we do that kind of iterative, stuff they do very well in the UX world and the agile world to a degree. [00:10:00] But D and A, it is more difficult.
Jon: It's more difficult, it's always let's start with the data and build a whole kind of infrastructure before we even start talking to our customers. Like, Actually we wanna do completely the other way around. And, this is where the whole data product thing comes in. And in my mind it's about product management and real product management is working with customers.
Jon: It's testing, it's experimenting, it's fast feedback, it's failing. It's, very much in the startup world. So that's the kind of the second bit. And the third bit is really, when people, they've decided what they wanna build. , can they actually build it? Can they actually, is it gonna take a 12 month program to do it before any kind of value gets realized, are there loads of friction blockers inside the organization? Again, waiting for tickets to open up firewall ports to get access to databases and all that type of stuff. All the way. Is there a way of actually doing iterative in incremental delivery of these pieces?
Jon: And then, you then iterating and orientation, observe, decide, and act, which is, you come from this sort of fighter pilot mentality. Can we implement that type of process? Because it's really around lean, efficient and getting back to getting to the market and reacting.
Jon: We're gonna make mistakes. That's the nature of the thing. There's, I see data missions being kind of expeditions into the unknown a little bit. We might change past, we actually might change the outcome, but [00:11:00] we get to the other side.
Jon: But it's, we need to iterate around it. Can we come across a rough terrain or bad weather or whatever, but you need to measure and kind of that sort of stuff, so for me that's the kind of key things to look at.
Kulvinder: And I think that actually brings me on nicely to my next question because I think, if you. Actually, let me take a step back with, when you said one of the things is the ability to test and, test your hypothesis and see how what might happen and then fix it or change it if you need to. I guess that's easier to do in a smaller kind of startup scale up company. Again, in your experience is is that something that takes a lot longer to do in a larger organization or what's your sense of that?.
Jon: Absolutely. The thing of it, the thing about larger organizations, it's not always true. There's some organizations are very good at them. You look at the big tech giants, they're, they're one side of the spectrum, but some more traditional companies are on the, on the other side.
Jon: So sort the process and organizational debt actually can actually get in the way. But it's, it's also, companies spinning up incubators and in labs and this type of stuff, which can help, but obviously you need the link back to the, pushing it out into the main organization, I think.
Jon: But for me it's around that commitment to actually be able to [00:12:00] get those blockers. If you are doing a data strategy or data initiative, having the empowerment to say, look, again, going back to my, opening a ticket to get database access for three months, there should be a data leader who's got the empowerment, who can override all that.
Jon: Fundamentally you can come back and say, Let's get the ticket sorted Or let's go and find the data, in, in some way in transactional system, and if the system owners, you're not touching my data cause and the whole thing stops, so it's, I think there's a lot of empowerment that needs to happen from the top. What we used to call the executive hammer in management consulting, the ability to actually get over these kind of organizational barriers, but also then, be able to win hearts and minds.
Jon: There's a sort of partnering aspect of it, saying actually you go to a line of business, say, we are going to make your business initiative work, or you're gonna save me this amount of money and this type of stuff. You partner with me and all the systems and processes in your organization, we need to change them, but you need to be a partner on that as well.
Jon: So organization. You see it, with the whole data mesh at the moment. This is this sort of friction. with large organizations, it's happening at the moment. People are trying to change the way people work and stuff, but it does take time.
Kulvinder: I completely agree with you. I think in terms of any initiative you need that the [00:13:00] leadership, you need that to be empowered to be able to make decisions and, change things as necessary.
Kulvinder: And it's I guess it comes onto, it's as much organizational as it is technical in some respects in terms of that delivery.
Jon: People process and technology, right? It's all free, right?
Kulvinder: Yeah. exactly. so let me ask you another question. And this is something that that has affected me quite recently in the organization I work at.
Kulvinder: Building the pipeline, putting in the, the groundwork and the technology and even getting the buy-in all straightforward. But in terms of data quality and data governance, Is there a way that, and I'm not suggesting there's a one size fits all, how would you propose to overcome challenges that are related to data quality and data governance
Jon: Yeah. This is a, again, a very two, two very great subjects and there's a lot of a lot of talk in the industry about this at the moment, and having seen, centralized data governance and data quality teams and staff struggle day in day out with the whole concept of it actually being able to get the rest of the organization to care, which is the biggest problem for me.
Jon: One of the kind of, I would say solutions, one way to [00:14:00] actually to solve this problem is, again, if you go to much more of a product management approach, data governance, data quality are features of each product we put out. Fundamentally. So you think about I want to do a recommendation engine for an e-commerce solution, right?
Jon: But I need to get, data from my, from the sales the sales system to get data from CRM, from the e-commerce system, from, the marketing system. And if some of those, the data quality of those isn't good enough for me able to do that, then I need to go back to those systems and those owners and stuff and actually use them to do, to actually to improve it.
Jon: But that's a conversation between a business. , i.e. The person who owns the use case for, the e-commerce, which could be sales or marketing, and then going back to the actual line of business, say, look, we need to improve. , it's a dollar amount on it, and it's feature of that particular, that that, that use case, that kind of product effectively, that recommendation engine.
Jon: And oh, also we need to improve these processes. We need to stream it out, streamline it and all that. So it to actually better do this, but it's a business conversation. The challenge is if it isn't a business conversation, it's like why do I care?
Jon: I've got my own lines of business, I've got my own metrics, this kind of business metrics. So I do think that we need to bake it into kind of that product man management men mentality with the. Features around [00:15:00] that, and actually that then becomes a much easier conversation. It's not easy, but it's easier than actually something separate, and it's also, then underpinned with capabilities around the technology and kind of, taxonomies and stuff like this as well.
Kulvinder: And with that in mind, would you say that, there's so many different paradigms and some of them newer and some of them older, but with something like data mesh or data as a product, do you think that gives the kind of the greater alignment with regards to processes and having the right people?
Kulvinder: I suppose enacting the right amount of governance and ensuring the quality of the data over maybe something like let's say data lakes or data warehouses?
Jon: I think so. I mean there's definitely the meshes are a good starting point. I think there's this whole concept of, and I'm not big fan of the data as product term. I wanna talk about data, products being genuine, business type facing products, but take that to one side, fundamentally getting the product discussion into it is actually, it's gotta be only be a good thing, right?
Jon: Fundamentally, product owners, proper product owners who actually drive features and drive business functionality, you understand the business, this sort of stuff is gotta be, has gotta be good. If you look at like data, dumps data lakes and warehouses, they're always [00:16:00] seen as something off the side.
Jon: If you talk to a business person, generally, unless it's something very close to the business, like you're driving a, P and L off a portfolio in a trading company or something like that. But other, most of the time it's like business think, oh, this lake and this warehouse it's over there somewhere.
Jon: and if it doesn't work, I'll jump up and down. But if it works, like it's fun, silent, so it's not, there's no ownership. Whereas if you go much more to a product view where you are, you are building, discrete. Business facing, components that actually solve business problems.
Jon: And it's all part of that. I, you can actually add product management kind of features and stuff into it and life cycle. It is a lot easier conversation to have because actually it's the product owner who owns the data policy rules, for instance, rather where it's on the lake and the warehouse, it's some data team, know, off to the side somewhere.
Kulvinder: So with that, I mean with that in mind, do you, would you say teams that are aligned closer?
Kulvinder: Both. Both from a technical point of view, but also a product. Business point of view are able to adapt to change and be more flexible?
Jon: I, I think so, yes. But there there's a sort of balance, right? You don't want basically complete chaos where every line of business is doing [00:17:00] completely their own thing and this kind of stuff. But, where I've seen it work quite a lot was, this embedding this, these business teams and enabling the business teams to actually build their own analytics and own, generally own they're owning them.
Jon: These are features and the product, the backlog and stuff like, , it's, it is a lot better, but you do need have a, and a coordinated effort. Like I said, if you wanna build a product that's across three different departments, for instance, you wanna make sure that you know that not sales, marketing, and maybe, logistics are all calling things the same things, and that you can actually go and get the data and you can actually merge 'em together.
Jon: So there's a little bit of a, a a balance to be struck there.
Kulvinder: We talked a little bit about, I suppose the people side of things and let's just cover the technology side of things. How, from your experience in the different environments you've been in, how important is the technology that either you procure, or build important in, in, a successful execution of that data strategy
Jon: Yeah, that's a great question again. A lot of, data strategies usually are very tech focused and it's, and people sort, lots of people are saying, oh, it's not just a technology problem. Mean technology's there to support your operating model. End of story, in my mind that's, fundamentally you've gotta get the operating model right.
Jon: First of all to [00:18:00] but then you've gotta get the technology to do a lot of the heavy lifting. Know, if you look at a lot of really great tech like Snowflake and Databricks and a lot of the really kind infrastructure stuff, they're very focused, on sort of the developer, technical person.
Jon: But actually not all D data analytics are really technical. So for instance, if I want to do a sales forecast, I want to get, the head of risk, head of finance, and the head of sales to agree to, I should get them in a room. So how are you gonna calculate the forecast, how you measure this kind of stuff, and then go away and build it and then it should be done really quickly and easily.
Jon: Without going two months of high, heavily technic heavy technical lifecycle. Building a simple sales match should be super simple. So one I, look at doing and something we're building at the moment is basically an infrastructure build, a really good UX kind of templating workflow, that type of stuff.
Jon: With and looking at different types of kind of analytics cause in most organizations there's only really four or five kind of major. What we call archetypes, types of analytics component. There's a metric which is, standard window function. There's a forecast, there's some sort of classification.
Jon: There is some sort of correlation piece, with other cohort analysis, that type of stuff. And there's and there's trying to measure behaviors and they really fall into the four or five different type of buckets. And then obviously look [00:19:00] at the fringe stuff like peak learning and what have you.
Jon: And obviously ChatGPT has changed that a little bit, but, fundamentally, so what the tech should be able to do is actually allow people to. Analyst really quickly and easily through UX to to deliver that. So I and the business, the finance team wanted say what's our cashflow forecasting?
Jon: We should be able to go through a wizard based approach. We're using big data as well large data sets, but actually then push out that forecast into a, something like a, what I'm calling the modern data ecosystem, of data product. But the tech should be able to facilitate all that sort of stuff.
Jon: You don't want people inside finance, For six months wrangling data. It's, being able to overlay things like, the business process and taxonomy, metadata and stuff, but all through a nice friendly ux. That to me is where the tech really comes in. But it's support, I say supporting the operating model,
Kulvinder: and that's a great point. For me I guess what I'm left wondering is, building a good usable UX is an art in itself. That's a craft, right? What I'm wondering is with regard to being able to access the data quickly versus efficiently, not necessarily both the same things, is there is, again, in your experience, is there something that's, that one is valued [00:20:00] over the other?
Jon: Yeah, it's, it 's the holy grail of, most organizations, large organizations, small organizations, a bit different cos there's people sitting next to each other, so it's less of an issue. But large organization, the holy grail is basically having almost like a marketplace of all my data.
Jon: I can go and have a look, see what it is, understand it, and just use it. I remember getting a A requirement from a risk officer in one of the big banks, and they said, I wanna be able to touch and feel the data. That was his requirement. I'm like, it's about 200 terabytes.
Jon: What does that actually mean? But it's, again, that's the holy grail. So what does that actually mean in, in practice? And, and what is lots of people trying to build it into centralized models and lots know, subject areas, all this type of stuff, which are divorced from the business practice.
Jon: And for me, it's quite simple. If you build a, an analytics model or a data use case as when you, as you go through and. Look at the data, bring it in, what have you, being able to curate it, annotate it, and then publish all those, almost like your homework around it. A bit like you Kaggle, you go into Kaggle again, Kaggle's not an enterprise company, but, and you go into Kaggle.
Jon: I can see what the data sets they're using for that model. I can see how they've actually come about it and this kind of stuff. It's more that sort of content management type [00:21:00] approach around to building data and building models, this type of stuff. We've got data sets attached to models and then reusing them.
Jon: To me, I think that's a much more powerful kind of paradigm than. Almost semantic layers.
Kulvinder: That brings me onto a kind of a follow up question, closely aligned with that one. ChatGPT is the poster child, I guess for AI in terms of the broader collective and the consciousness, and people now understanding, things that it could do.
Kulvinder: Do you foresee a situation whereby there is there is something similar to that where people perhaps can ask questions in plain text. And our, our AI has got so good that we're, we are able to, present. And answer those questions with very quickly. You go back to that touch and feel the data they wanna be able to touch and feel in the way that they want.
Kulvinder: Whether that's most pertinent to them as an individual or a department. So do you see that coming down the line?
Jon: I think so, but the interesting, I, I did a project on, the start of last year actually doing something very similar. It wasn't quite level the scale of ChatGPT, but was looking at, corpus of job adverts and able to actually re um, do text generation based on, so I know [00:22:00] how difficult that, that piece is.
Jon: And the thing that always interests me about, models and deep learning based nlp. And natural language processing kind of problems is the amount of data that's required. It's obviously ChatGPT and all kinds of its ilk are all trained on public data and this type of stuff, which is huge, right?
Jon: It's big corpus to train it and this kind of stuff. You see it numbers like 65 billion parameters. All enterprise data isn't that big. , unless you are talking about, a manufactured company, large amounts of IOT data or connected car or stuff. Most enterprise data isn't that big.
Jon: So being able to train deep learning, GPT, type models on enterprise is actually really difficult. Cause you, you don't have enough data. You take a loans company for instance, you know the loan. They're revaluing the loan monthly or even quarterly, it's like you've get the entire lifetime of a loan.
Jon: It's might be 40 data points, not a, you can't really use deep learning for that type of stuff. That said, natural language processing and that's language interfaces. It's been around for a long time. SQL Server and Power BI have got natural language interfaces , but no one uses them.
Jon: So I think what really is, the ChatGPT and the meta and the Google type, it is the breadth of kind of the capability be able to generate, [00:23:00] sequel Python and generate, good prose and the start of Donald Trump and that, that's something that's quite new.
Jon: But being able to actually ask questions of your enterprise data or enterprise analytics, there's still a little bit of a gap there unless you basically, execute the analytics and then just say, ChatGPT Can you describe this. Put the data in that way. So there's still a little bit of gap there.
Jon: But yeah, what it has is the, that kind of it's lit the fire on natural language interfaces fundamentally. for me that's the really big thing. We're not, dashboards and that kind of stuff anymore, be able to actually say, what, why am my orders falling off a cliff?
Jon: Great question to ask. Now ChatGPT wouldn't be able to answer that because fundamentally that's a forecast that's basically looking to the future, whereas ChatGPT is all about, fundamentally historical data, fundamentally. So you need a natural language interface to be able to actually execute a forecast model or a semantic model, something that can actually predict something so that linkage between the models and the natural language interface is something that needs to built and that isn't there yet at the moment and that's really where things get super interesting.
Kulvinder: Yeah. I think that brings me onto to, I was gonna ask you about ethics and privacy with regard to data.
Kulvinder: So naturally my mind goes towards, a model whereby [00:24:00] there is access to public data or, public data sets a lot, coupled with your own enterprise data sets that could lead to potentially some shady kind of processes and sha shady behavior. Is there, are other things that companies can do to ensure that they are taking that ethical and they are respecting the privacy of individuals?
Jon: Yeah, obviously there's a there, there's a couple. There's probably spent a whole session just on this piece, but I think there's a, me the first things is about transparency. fundamentally what's actually happening here. And, deep learning models are the notoriously black box, it's quite tricky to actually link stuff all the way back to where the data's come from.
Jon: So they go through lots of different convolutions what have you. But the transparency is a big thing. I if, a company's making decision about me or it's using my data, I wanna know about it. I wanna know actually what it was. And you look at the ChatGPT around hallucinations.
Jon: So we I saw that as. The solution I was building on, a year ago, these hallucinations but it was very confidently saying, this is what it is. It's actually, is that right? So fundamentally I say, can I question that? Can I actually go back and see where, how it's made that decision?
Jon: Where's it got the source of that data from? Especially with Enterprise, [00:25:00] it's if you end up with two kind of finance numbers and they're quite different. I'll ask the question three times in different ways. It's got lot of different numbers hang on a minute, that doesn't look right.
Jon: And there's no way of me to drill into that. I think that's, that that's definitely. But also having things like, commercially aware CISOs and stuff. People who are, who understand the privacy and the ethics, but also are commercially orientated. Not people are just gonna say I can't do anything cause I don't wanna go to jail.
Jon: This is what we need to think about as part, again, features of the product, features of the of the delivery. These are, these have gotta be in there. We want to build this new app that can predict, car breakdowns, but actually this is how we get, this is how we're gonna predict it and this is how we're gonna use it.
Jon: So your car's gonna break down. This is. We've arrived. That situation, that's a bunch of features and a bunch of product features that need to go into the product by someone who understands those things, understands, gdpr, understands the ethics, understands the new European rules, the rules are coming in and one to the US as well.
Kulvinder: I think that's a, that's. That's the next point. I It's an area that I suppose we'll hear more about. and, I think you're right. I think transparency is key there. And it depends on the nature of the company as well around what they're doing.
Kulvinder: So yeah. No, thank you for that. I think that probably brings me [00:26:00] onto my last question for you, which is in terms of you think about all the projects you've been involved with, and I'm sure they've be numerous. So is there one that sticks in your mind as a really successful, or what you might term successful data strategy, implementation?
Kulvinder: And if there is what lessons do you think can be learned from that particular experience? What would be useful for others as a takeaway, do you think?
Jon: Yeah, no, it's a good question. Yeah, I've just been thinking about this quite a bit actually. There's, there, there are a lot I can tell which haven't worked a lot I've stuff, it's and, but the ones that have fundamentally it, it comes back to kind again, two or three things. One is this business are basically, being out using, scientific decision making, fundamentally the execs buying into the fact that they're gonna, we're test it gonna build analytics and we're gonna, we're gonna act on fundamentally, and there have been a few companies have actually done that. And it's worked really well. Also, the, from the operating model perspective getting the lines of business to actually build the analytics, it's a lot of time.
Jon: It's very hard to do that cause without budgets and it's we used to hand this over to other people, but actually getting people sat by side by. With the business teams and actually iterating and this kind of stuff that's super successful. It's funny, I would [00:27:00] go back to my trading days where developer would sit by the trader, and the trader would go build this model, but you hack away and build it, and then he'd put it out and if it didn't work, they lost lots of money and the developer gets shouted at, but the point was they'll get it would really drive the business.
Jon: They're sitting side by side seeing that kind of partnership. And I think that's, that to me, from a data and analytics perspective is really where we need to get to. Whereas, DNA is seen as a partnership with the business to help 'em be successful, and getting people into the lines of business, which I've spoken to a colleague yesterday around a big manufacturer's just done this, and they've been very successful.
Jon: Getting the execs to buy into the kind of algorithmic and scientific decision making, seeing what the benefits are, rather than just saying, I'm gonna do it heuristically and I know my customers better than them, and also, that sort of testing that kind of, trying it out.
Jon: I You see it in the startup world all the time. Startups live and die by testing the market and prototyping their solutions . Some the companies I've seen have done that have actually, saved loads of money.
Kulvinder: Great. Look, John, thank you so much for your time. I hope everyone who's listening and watching got as much outta the discussions as I did. I could have carried on talking to you for a lot longer. Not least around the sort of the AI predictive analytics [00:28:00] and using external and internal data sets, but we're gonna have to kind a call it there.
Kulvinder: Thank you again. Thanks very much for your time. On today's episode.
Jon: Pleasure, I really enjoyed it.
Kulvinder: Thank you And yeah, thanks sir for being my first guest as well. I appreciate that. Folks, I hope you enjoyed the episode. The conversation. Went into some different directions. I hope you answered the questions you had.
Kulvinder: If you've got any other questions, please feel free to drop a comment or send us a note. Jon where can folks find you best? On, on the interweb?
Kulvinder: Yeah. So I'm on LinkedIn, Jon Cooke data. Just do
Kulvinder: Fantastic. So thanks. Look, if you enjoyed this episode and I hope you did be sure to follow along and leave us a review. Your support helps us reach more listeners and bring you even more exciting episodes in the future. Hopefully there'll be a new episode up soon.
Kulvinder: Until then, stay curious.
Kulvinder: And thank you very much.
===
Kulvinder: [00:00:00] Welcome to Tech Exec Talks, where we sit down with leaders of the tech industry to discuss their experiences, perspectives, and vision for the future of technology. This episode's talking all about data strategy. Every organization creates data, sometimes knowingly, sometimes unknowingly, but the desire to utilize that data or use it better impacts every organization at some point in their evolution.
Kulvinder: So where do you start? That's the subject for today. That's the question we're gonna be answering. So I'm Kulvinder, I'm your host. And this episode we'll be talking to Jon Cooke. So Jon is the owner and founder of Dataception, a data analytics and data product company, and the creator of the data product pyramid, which is an adaptive data product operating model. He's a data product storyteller and technology specialist, and he spent the last 30 years delivering data analytics and product capabilities and marketplaces for analytic use cases across many companies. So [00:01:00] Jon, thank you for joining me today. How are you?
Jon: Oh, thanks Kulvinder. Thanks a lot for having me on the podcast. Yeah I'm great. I'm certainly looking forward talking around data in this episode. So yeah, thanks for having me.
Kulvinder: Okay, let's get into it. Let's start let's talk about all things data. So I guess for our listeners and also for me I'm of curious as well how'd you get into data as an area to specialize in and why was it interesting?
Jon: Yeah, no it's been a very, quite an interesting, journey for me. It's like over the 30 years, I graduated in the mid nineties. Actually worked for a research company in inside British Aerospace doing virtual reality, believe it or not. But then Went into gaming and then sort really, started getting around getting involved in message orientated middleware, low latency messaging.
Jon: Did a lot of service orientated architecture on, in the early two thousands, low latency web services, that sort of stuff. And it was interesting again, very data orientated, but much more kind of around the software software space. And then I got into investment banking in mid two thousands.
Jon: And I was sort of front office trading and risk models and all that type of stuff, which is. Hugely data and analytics heavy, doing Monte Carlo simulations on a billion risk points and all that type of stuff, and low latency market data messaging [00:02:00] very much involved in around using data to actually drive analytics for investment banking type portfolios.
Jon: I then moved from there , got into a consultancy and also when start looking at big data, so Hadoop was like 2010 Hadoop and 11 that was around, and we're trying to build risk models and risk capability. On top of Hadoop and big data. So again, massive amounts of data, lots of low latency, that type of stuff.
Jon: Ended up building a practice in data practice in the consultancy and then moved on from there to work to build data and AI practice in pwc. So got much more around into the AI and analytics side of it, doing lots of data strategy. Data, governance, that type of stuff, driving business initiatives with data.
Jon: I left there and then went for Databricks, who everyone, I'm sure everyone knows who they are fantastic company. Was there for very early days of 250 people I think it was. So I was in, in EMEA and built out the solution architecture team there. And yeah, getting involved in AI and data, of course that's their bread and butter.
Jon: And I left there in sort of 2019 and formed my own company. Cause I, I thought there was. Databricks and these other companies are really good. But there was a space in the middle around linking business data and technology really driving [00:03:00] that kind of product mindset. And then I founded the company in 2019 Dataception to do that.
Jon: And that's exactly what I've been doing for the last sort of four years. It's very much that kind of data product, that product mindset with data and analytics in the business. Yeah, it’s been an underlying theme all the way through my career, but very much a focus in probably the last fifteen years on data itself.
Kulvinder: Fantastic. And that's probably a nice segue into, what for you constitutes a good and successful data strategy. What are the key elements of that, would you say?
Jon: Yeah, no, it's a great question. So I think I've been involved in many data strategies and many data initiative, and I think there's two or three key pieces. There's the specifics and general. And the specifics are can you can data and analytics now, and again I tried to use the terms of data and analytics, cause data itself is like the fuel, but the analytics is really the vehicle for driving this sort of stuff.
Jon: So I'd see the the data and analytics strategy really been around specifics, how the company does it and are there specific initiative that data needs, needs to drive. So I'm working with one customer, they're building a new analytics app. They're trying to put out the 4 million customers, have they got the data to, to drive that and actually make that successful.
Jon: [00:04:00] So there, there's some specifics that need to be addressed, but then there's the more sort of general capability, and split those into kind of two areas. There's the transactional plumbing, are your data processes. Fit for purpose, for doing things like, your reg reporting for updating your sub-ledger to doing your risk models, to doing your, your sales forecasting. good hygiene in those sort of things. And then there's a kind of the really interesting bit around embedding data and analytics into the lines of business and being what I call, data driven or scientific driven, scientific methods, experimentation, testing measuring with customers, prototyping and that sort of stuff.
Jon: Being really sort algorithmic. Precision making is, That to me is really, in inside the lines of business where I think that, that's one of the big tick points
Kulvinder: And based on your experience, have you have you had to go into a company where there's absolutely zero, they have nothing. So obviously they have data sets clearly, but they have the desire to do something with the data, but they don't know where to start.
Kulvinder: Is that something that you can, you've done and you can help with?
Jon: Yeah, absolutely. So I think one of the, one of the things I've, noticed over the last sort of 30 years getting the [00:05:00] business to really understand about algorithmic decision making and defining the requirements, based on what they want to do. It's actually really tricky. , we felt, we in data and tech, we're notorious for using Bud Buzzwords and coming up with methodologies and business requirements documents and stuff like that.
Jon: And half the time the business was left scratching their head going, actually, what does this actually mean? How do I actually orientate this? Or, the business are there, saying this is what I do day to day and I don't really wanna go outside of what I want to do. So that's another big thing.
Jon: And it's all sorts of spectrums, right? So you get large companies where you got pockets of that. You get small startups where they don't how to actually articulate the the business requirements, that sort of stuff. So yeah, absolutely this is this, these are the key things that we need to solve.
Jon: And for me it's basically helping, I Data and analytics, people helping, driving the business, driving those kind of outcomes. Value frameworks, basically helping to articulate the business requirements in data and analytics terms, which are easy and simple and deliverable for the business.
Jon: So it's, yeah, it's really trying to, I see, some being sort of data leaders being not Sherpas now, people carry kind of stuff, but actually being like being guides.
Kulvinder: That's a great that's a great example you've given there and I think, [00:06:00] would you say that's all equally as true where you've got let's say there's an established data strategy for whatever reason isn't quite working or isn't yielding the results that certainly the business would hope for.
Kulvinder: Do you take a different approach to that or is it broadly similar?
Jon: I would just look at a data strategy and, fundamentally has it got there's a couple of things that, that it needs. First of all, is there a, is it a is there a product mindset around that? Is there, were you building analytics? Actually as a product management process.
Jon: So fundamentally, are you solving the business problem? Do you have a customer? Do you actually, are you building the kind of the data and the analytics, the model and stuff, and in a product life cycle type stuff? So that's my first looking at it. And the second bit is really, is there a culture of, algorithmic decision making?
Jon: Scientific methods. Is that there already? So data strategies are great. I've seen lots. You will build a platform, we'll build infrastructure, we'll get all our data sets together, we'll classify it, we'll get taxonomies and great. It's hang on a minute. How does that actually drive the business strategy?
Jon: Does/is the last major initiative that happened that the data was supposed to support, was it successful because of it or was it a hindrance? Does it take three months to spin up a, an experiment [00:07:00] environment to actually allow people to start testing before they, they can actually Go go with the business strategy, to set it out if it's gonna work.
Jon: Is that all that stuff there? Is it frictionless? Is it like low friction? Can it, can people can operate and move in, in a kind of fluid way? So these are the things I look at the data strategy.
Kulvinder: Yeah, I was gonna ask, following up on that with how important would you say the culture of the company is in that journey of ensuring that their data strategy and their approach to it. There, there is that alignment there.
Kulvinder: I guess
Jon: Absolutely. Again it comes back to a couple of things for. Do they support scientific methods and algorithmic decision making? That's the first thing. Cause there's no point having a great, I can build you a great data analytics capibility unless it's capability system. But if you, they're gonna ignore all the results and not allow people to, to actually experiment and test, then it's it's not, you're not gonna get anywhere.
Jon: So that's the first thing. The second thing is really how receptive are they to business cases in terms of actually driving change? So if I say, actually you're not supporting this particular business initiative because x, y, and z I can change that and save you 5 million pounds a year or whatever, but you need to change this, and [00:08:00] this, and it's some structural changes.
Jon: Are they amenable to do that? Cause again, you can say, look, I can completely turn your business around, but if they're prepared to actually do it, to execute, to action, that those sorts of things, then that's another kind of big, a bit of a, thing we need to pull a thread on, and also of how do they think about data fundamentally? Do they treat it as something that's a, a competitive advantage? Is it or is it just plumbing that just needs to be fixed when it breaks, so again, to me is part of that cultural aspect that you need to reassess
Kulvinder: Excellent. I I, and I think I think the interesting thing for me I guess it's, it is like a lot of things, right?
Kulvinder: With technology you can easily go out and procure the software or the means to be able to fix a need. But until you can articulate that need and , the whole business around that, it's gonna be tricky to execute on. My, my go-to is always CRM in that regard. Where back in the day was we need a CRM system.
Kulvinder: What do you need it to do? can't quite articulate that. So feels quite akin to that, so in terms of that kind of alignment and prioritization of those data initiatives, sounds like it comes back to. What do you want to do with it? Like you say, do you wanna realize efficiencies, et [00:09:00] cetera?
Kulvinder: Do you have any kind of go-to strategies that you employ to help companies understand what those data initiatives or business needs are that would lead to those data initiatives?
Jon: Absolutely. That's a great, great question. So I think the first thing I look at and to see is that have they got a sense of kind of what we call a value framework? Have they got an idea of actually what's valuable to them in terms of actually they what they wanna do? And that's not a data thing.
Jon: That's a kind of a business initiative. Seen lots of business initiatives which have been kicked off, big Bang, massive programmes spinning up 30 people, but they actually haven't got a clear outcome or a clear business value outcome. So I, that's the first thing I look at, is basically can they, can you articulate a value.
Jon: Fundamentally, do they know what, if you've got two execs, they understand the same version of value, can they prioritize stuff based on that kind of value? The second bit is can they, do they have a a process for, testing and iteration without, again, spinning up big bang, do small incremental delivery.
Jon: So again, they wanna build a, a an analytics app for their customers. Can we prototype it? Can we test it out with a friendly set of customers? Can we measure it? Can we do that kind of iterative, stuff they do very well in the UX world and the agile world to a degree. [00:10:00] But D and A, it is more difficult.
Jon: It's more difficult, it's always let's start with the data and build a whole kind of infrastructure before we even start talking to our customers. Like, Actually we wanna do completely the other way around. And, this is where the whole data product thing comes in. And in my mind it's about product management and real product management is working with customers.
Jon: It's testing, it's experimenting, it's fast feedback, it's failing. It's, very much in the startup world. So that's the kind of the second bit. And the third bit is really, when people, they've decided what they wanna build. , can they actually build it? Can they actually, is it gonna take a 12 month program to do it before any kind of value gets realized, are there loads of friction blockers inside the organization? Again, waiting for tickets to open up firewall ports to get access to databases and all that type of stuff. All the way. Is there a way of actually doing iterative in incremental delivery of these pieces?
Jon: And then, you then iterating and orientation, observe, decide, and act, which is, you come from this sort of fighter pilot mentality. Can we implement that type of process? Because it's really around lean, efficient and getting back to getting to the market and reacting.
Jon: We're gonna make mistakes. That's the nature of the thing. There's, I see data missions being kind of expeditions into the unknown a little bit. We might change past, we actually might change the outcome, but [00:11:00] we get to the other side.
Jon: But it's, we need to iterate around it. Can we come across a rough terrain or bad weather or whatever, but you need to measure and kind of that sort of stuff, so for me that's the kind of key things to look at.
Kulvinder: And I think that actually brings me on nicely to my next question because I think, if you. Actually, let me take a step back with, when you said one of the things is the ability to test and, test your hypothesis and see how what might happen and then fix it or change it if you need to. I guess that's easier to do in a smaller kind of startup scale up company. Again, in your experience is is that something that takes a lot longer to do in a larger organization or what's your sense of that?.
Jon: Absolutely. The thing of it, the thing about larger organizations, it's not always true. There's some organizations are very good at them. You look at the big tech giants, they're, they're one side of the spectrum, but some more traditional companies are on the, on the other side.
Jon: So sort the process and organizational debt actually can actually get in the way. But it's, it's also, companies spinning up incubators and in labs and this type of stuff, which can help, but obviously you need the link back to the, pushing it out into the main organization, I think.
Jon: But for me it's around that commitment to actually be able to [00:12:00] get those blockers. If you are doing a data strategy or data initiative, having the empowerment to say, look, again, going back to my, opening a ticket to get database access for three months, there should be a data leader who's got the empowerment, who can override all that.
Jon: Fundamentally you can come back and say, Let's get the ticket sorted Or let's go and find the data, in, in some way in transactional system, and if the system owners, you're not touching my data cause and the whole thing stops, so it's, I think there's a lot of empowerment that needs to happen from the top. What we used to call the executive hammer in management consulting, the ability to actually get over these kind of organizational barriers, but also then, be able to win hearts and minds.
Jon: There's a sort of partnering aspect of it, saying actually you go to a line of business, say, we are going to make your business initiative work, or you're gonna save me this amount of money and this type of stuff. You partner with me and all the systems and processes in your organization, we need to change them, but you need to be a partner on that as well.
Jon: So organization. You see it, with the whole data mesh at the moment. This is this sort of friction. with large organizations, it's happening at the moment. People are trying to change the way people work and stuff, but it does take time.
Kulvinder: I completely agree with you. I think in terms of any initiative you need that the [00:13:00] leadership, you need that to be empowered to be able to make decisions and, change things as necessary.
Kulvinder: And it's I guess it comes onto, it's as much organizational as it is technical in some respects in terms of that delivery.
Jon: People process and technology, right? It's all free, right?
Kulvinder: Yeah. exactly. so let me ask you another question. And this is something that that has affected me quite recently in the organization I work at.
Kulvinder: Building the pipeline, putting in the, the groundwork and the technology and even getting the buy-in all straightforward. But in terms of data quality and data governance, Is there a way that, and I'm not suggesting there's a one size fits all, how would you propose to overcome challenges that are related to data quality and data governance
Jon: Yeah. This is a, again, a very two, two very great subjects and there's a lot of a lot of talk in the industry about this at the moment, and having seen, centralized data governance and data quality teams and staff struggle day in day out with the whole concept of it actually being able to get the rest of the organization to care, which is the biggest problem for me.
Jon: One of the kind of, I would say solutions, one way to [00:14:00] actually to solve this problem is, again, if you go to much more of a product management approach, data governance, data quality are features of each product we put out. Fundamentally. So you think about I want to do a recommendation engine for an e-commerce solution, right?
Jon: But I need to get, data from my, from the sales the sales system to get data from CRM, from the e-commerce system, from, the marketing system. And if some of those, the data quality of those isn't good enough for me able to do that, then I need to go back to those systems and those owners and stuff and actually use them to do, to actually to improve it.
Jon: But that's a conversation between a business. , i.e. The person who owns the use case for, the e-commerce, which could be sales or marketing, and then going back to the actual line of business, say, look, we need to improve. , it's a dollar amount on it, and it's feature of that particular, that that, that use case, that kind of product effectively, that recommendation engine.
Jon: And oh, also we need to improve these processes. We need to stream it out, streamline it and all that. So it to actually better do this, but it's a business conversation. The challenge is if it isn't a business conversation, it's like why do I care?
Jon: I've got my own lines of business, I've got my own metrics, this kind of business metrics. So I do think that we need to bake it into kind of that product man management men mentality with the. Features around [00:15:00] that, and actually that then becomes a much easier conversation. It's not easy, but it's easier than actually something separate, and it's also, then underpinned with capabilities around the technology and kind of, taxonomies and stuff like this as well.
Kulvinder: And with that in mind, would you say that, there's so many different paradigms and some of them newer and some of them older, but with something like data mesh or data as a product, do you think that gives the kind of the greater alignment with regards to processes and having the right people?
Kulvinder: I suppose enacting the right amount of governance and ensuring the quality of the data over maybe something like let's say data lakes or data warehouses?
Jon: I think so. I mean there's definitely the meshes are a good starting point. I think there's this whole concept of, and I'm not big fan of the data as product term. I wanna talk about data, products being genuine, business type facing products, but take that to one side, fundamentally getting the product discussion into it is actually, it's gotta be only be a good thing, right?
Jon: Fundamentally, product owners, proper product owners who actually drive features and drive business functionality, you understand the business, this sort of stuff is gotta be, has gotta be good. If you look at like data, dumps data lakes and warehouses, they're always [00:16:00] seen as something off the side.
Jon: If you talk to a business person, generally, unless it's something very close to the business, like you're driving a, P and L off a portfolio in a trading company or something like that. But other, most of the time it's like business think, oh, this lake and this warehouse it's over there somewhere.
Jon: and if it doesn't work, I'll jump up and down. But if it works, like it's fun, silent, so it's not, there's no ownership. Whereas if you go much more to a product view where you are, you are building, discrete. Business facing, components that actually solve business problems.
Jon: And it's all part of that. I, you can actually add product management kind of features and stuff into it and life cycle. It is a lot easier conversation to have because actually it's the product owner who owns the data policy rules, for instance, rather where it's on the lake and the warehouse, it's some data team, know, off to the side somewhere.
Kulvinder: So with that, I mean with that in mind, do you, would you say teams that are aligned closer?
Kulvinder: Both. Both from a technical point of view, but also a product. Business point of view are able to adapt to change and be more flexible?
Jon: I, I think so, yes. But there there's a sort of balance, right? You don't want basically complete chaos where every line of business is doing [00:17:00] completely their own thing and this kind of stuff. But, where I've seen it work quite a lot was, this embedding this, these business teams and enabling the business teams to actually build their own analytics and own, generally own they're owning them.
Jon: These are features and the product, the backlog and stuff like, , it's, it is a lot better, but you do need have a, and a coordinated effort. Like I said, if you wanna build a product that's across three different departments, for instance, you wanna make sure that you know that not sales, marketing, and maybe, logistics are all calling things the same things, and that you can actually go and get the data and you can actually merge 'em together.
Jon: So there's a little bit of a, a a balance to be struck there.
Kulvinder: We talked a little bit about, I suppose the people side of things and let's just cover the technology side of things. How, from your experience in the different environments you've been in, how important is the technology that either you procure, or build important in, in, a successful execution of that data strategy
Jon: Yeah, that's a great question again. A lot of, data strategies usually are very tech focused and it's, and people sort, lots of people are saying, oh, it's not just a technology problem. Mean technology's there to support your operating model. End of story, in my mind that's, fundamentally you've gotta get the operating model right.
Jon: First of all to [00:18:00] but then you've gotta get the technology to do a lot of the heavy lifting. Know, if you look at a lot of really great tech like Snowflake and Databricks and a lot of the really kind infrastructure stuff, they're very focused, on sort of the developer, technical person.
Jon: But actually not all D data analytics are really technical. So for instance, if I want to do a sales forecast, I want to get, the head of risk, head of finance, and the head of sales to agree to, I should get them in a room. So how are you gonna calculate the forecast, how you measure this kind of stuff, and then go away and build it and then it should be done really quickly and easily.
Jon: Without going two months of high, heavily technic heavy technical lifecycle. Building a simple sales match should be super simple. So one I, look at doing and something we're building at the moment is basically an infrastructure build, a really good UX kind of templating workflow, that type of stuff.
Jon: With and looking at different types of kind of analytics cause in most organizations there's only really four or five kind of major. What we call archetypes, types of analytics component. There's a metric which is, standard window function. There's a forecast, there's some sort of classification.
Jon: There is some sort of correlation piece, with other cohort analysis, that type of stuff. And there's and there's trying to measure behaviors and they really fall into the four or five different type of buckets. And then obviously look [00:19:00] at the fringe stuff like peak learning and what have you.
Jon: And obviously ChatGPT has changed that a little bit, but, fundamentally, so what the tech should be able to do is actually allow people to. Analyst really quickly and easily through UX to to deliver that. So I and the business, the finance team wanted say what's our cashflow forecasting?
Jon: We should be able to go through a wizard based approach. We're using big data as well large data sets, but actually then push out that forecast into a, something like a, what I'm calling the modern data ecosystem, of data product. But the tech should be able to facilitate all that sort of stuff.
Jon: You don't want people inside finance, For six months wrangling data. It's, being able to overlay things like, the business process and taxonomy, metadata and stuff, but all through a nice friendly ux. That to me is where the tech really comes in. But it's support, I say supporting the operating model,
Kulvinder: and that's a great point. For me I guess what I'm left wondering is, building a good usable UX is an art in itself. That's a craft, right? What I'm wondering is with regard to being able to access the data quickly versus efficiently, not necessarily both the same things, is there is, again, in your experience, is there something that's, that one is valued [00:20:00] over the other?
Jon: Yeah, it's, it 's the holy grail of, most organizations, large organizations, small organizations, a bit different cos there's people sitting next to each other, so it's less of an issue. But large organization, the holy grail is basically having almost like a marketplace of all my data.
Jon: I can go and have a look, see what it is, understand it, and just use it. I remember getting a A requirement from a risk officer in one of the big banks, and they said, I wanna be able to touch and feel the data. That was his requirement. I'm like, it's about 200 terabytes.
Jon: What does that actually mean? But it's, again, that's the holy grail. So what does that actually mean in, in practice? And, and what is lots of people trying to build it into centralized models and lots know, subject areas, all this type of stuff, which are divorced from the business practice.
Jon: And for me, it's quite simple. If you build a, an analytics model or a data use case as when you, as you go through and. Look at the data, bring it in, what have you, being able to curate it, annotate it, and then publish all those, almost like your homework around it. A bit like you Kaggle, you go into Kaggle again, Kaggle's not an enterprise company, but, and you go into Kaggle.
Jon: I can see what the data sets they're using for that model. I can see how they've actually come about it and this kind of stuff. It's more that sort of content management type [00:21:00] approach around to building data and building models, this type of stuff. We've got data sets attached to models and then reusing them.
Jon: To me, I think that's a much more powerful kind of paradigm than. Almost semantic layers.
Kulvinder: That brings me onto a kind of a follow up question, closely aligned with that one. ChatGPT is the poster child, I guess for AI in terms of the broader collective and the consciousness, and people now understanding, things that it could do.
Kulvinder: Do you foresee a situation whereby there is there is something similar to that where people perhaps can ask questions in plain text. And our, our AI has got so good that we're, we are able to, present. And answer those questions with very quickly. You go back to that touch and feel the data they wanna be able to touch and feel in the way that they want.
Kulvinder: Whether that's most pertinent to them as an individual or a department. So do you see that coming down the line?
Jon: I think so, but the interesting, I, I did a project on, the start of last year actually doing something very similar. It wasn't quite level the scale of ChatGPT, but was looking at, corpus of job adverts and able to actually re um, do text generation based on, so I know [00:22:00] how difficult that, that piece is.
Jon: And the thing that always interests me about, models and deep learning based nlp. And natural language processing kind of problems is the amount of data that's required. It's obviously ChatGPT and all kinds of its ilk are all trained on public data and this type of stuff, which is huge, right?
Jon: It's big corpus to train it and this kind of stuff. You see it numbers like 65 billion parameters. All enterprise data isn't that big. , unless you are talking about, a manufactured company, large amounts of IOT data or connected car or stuff. Most enterprise data isn't that big.
Jon: So being able to train deep learning, GPT, type models on enterprise is actually really difficult. Cause you, you don't have enough data. You take a loans company for instance, you know the loan. They're revaluing the loan monthly or even quarterly, it's like you've get the entire lifetime of a loan.
Jon: It's might be 40 data points, not a, you can't really use deep learning for that type of stuff. That said, natural language processing and that's language interfaces. It's been around for a long time. SQL Server and Power BI have got natural language interfaces , but no one uses them.
Jon: So I think what really is, the ChatGPT and the meta and the Google type, it is the breadth of kind of the capability be able to generate, [00:23:00] sequel Python and generate, good prose and the start of Donald Trump and that, that's something that's quite new.
Jon: But being able to actually ask questions of your enterprise data or enterprise analytics, there's still a little bit of a gap there unless you basically, execute the analytics and then just say, ChatGPT Can you describe this. Put the data in that way. So there's still a little bit of gap there.
Jon: But yeah, what it has is the, that kind of it's lit the fire on natural language interfaces fundamentally. for me that's the really big thing. We're not, dashboards and that kind of stuff anymore, be able to actually say, what, why am my orders falling off a cliff?
Jon: Great question to ask. Now ChatGPT wouldn't be able to answer that because fundamentally that's a forecast that's basically looking to the future, whereas ChatGPT is all about, fundamentally historical data, fundamentally. So you need a natural language interface to be able to actually execute a forecast model or a semantic model, something that can actually predict something so that linkage between the models and the natural language interface is something that needs to built and that isn't there yet at the moment and that's really where things get super interesting.
Kulvinder: Yeah. I think that brings me onto to, I was gonna ask you about ethics and privacy with regard to data.
Kulvinder: So naturally my mind goes towards, a model whereby [00:24:00] there is access to public data or, public data sets a lot, coupled with your own enterprise data sets that could lead to potentially some shady kind of processes and sha shady behavior. Is there, are other things that companies can do to ensure that they are taking that ethical and they are respecting the privacy of individuals?
Jon: Yeah, obviously there's a there, there's a couple. There's probably spent a whole session just on this piece, but I think there's a, me the first things is about transparency. fundamentally what's actually happening here. And, deep learning models are the notoriously black box, it's quite tricky to actually link stuff all the way back to where the data's come from.
Jon: So they go through lots of different convolutions what have you. But the transparency is a big thing. I if, a company's making decision about me or it's using my data, I wanna know about it. I wanna know actually what it was. And you look at the ChatGPT around hallucinations.
Jon: So we I saw that as. The solution I was building on, a year ago, these hallucinations but it was very confidently saying, this is what it is. It's actually, is that right? So fundamentally I say, can I question that? Can I actually go back and see where, how it's made that decision?
Jon: Where's it got the source of that data from? Especially with Enterprise, [00:25:00] it's if you end up with two kind of finance numbers and they're quite different. I'll ask the question three times in different ways. It's got lot of different numbers hang on a minute, that doesn't look right.
Jon: And there's no way of me to drill into that. I think that's, that that's definitely. But also having things like, commercially aware CISOs and stuff. People who are, who understand the privacy and the ethics, but also are commercially orientated. Not people are just gonna say I can't do anything cause I don't wanna go to jail.
Jon: This is what we need to think about as part, again, features of the product, features of the of the delivery. These are, these have gotta be in there. We want to build this new app that can predict, car breakdowns, but actually this is how we get, this is how we're gonna predict it and this is how we're gonna use it.
Jon: So your car's gonna break down. This is. We've arrived. That situation, that's a bunch of features and a bunch of product features that need to go into the product by someone who understands those things, understands, gdpr, understands the ethics, understands the new European rules, the rules are coming in and one to the US as well.
Kulvinder: I think that's a, that's. That's the next point. I It's an area that I suppose we'll hear more about. and, I think you're right. I think transparency is key there. And it depends on the nature of the company as well around what they're doing.
Kulvinder: So yeah. No, thank you for that. I think that probably brings me [00:26:00] onto my last question for you, which is in terms of you think about all the projects you've been involved with, and I'm sure they've be numerous. So is there one that sticks in your mind as a really successful, or what you might term successful data strategy, implementation?
Kulvinder: And if there is what lessons do you think can be learned from that particular experience? What would be useful for others as a takeaway, do you think?
Jon: Yeah, no, it's a good question. Yeah, I've just been thinking about this quite a bit actually. There's, there, there are a lot I can tell which haven't worked a lot I've stuff, it's and, but the ones that have fundamentally it, it comes back to kind again, two or three things. One is this business are basically, being out using, scientific decision making, fundamentally the execs buying into the fact that they're gonna, we're test it gonna build analytics and we're gonna, we're gonna act on fundamentally, and there have been a few companies have actually done that. And it's worked really well. Also, the, from the operating model perspective getting the lines of business to actually build the analytics, it's a lot of time.
Jon: It's very hard to do that cause without budgets and it's we used to hand this over to other people, but actually getting people sat by side by. With the business teams and actually iterating and this kind of stuff that's super successful. It's funny, I would [00:27:00] go back to my trading days where developer would sit by the trader, and the trader would go build this model, but you hack away and build it, and then he'd put it out and if it didn't work, they lost lots of money and the developer gets shouted at, but the point was they'll get it would really drive the business.
Jon: They're sitting side by side seeing that kind of partnership. And I think that's, that to me, from a data and analytics perspective is really where we need to get to. Whereas, DNA is seen as a partnership with the business to help 'em be successful, and getting people into the lines of business, which I've spoken to a colleague yesterday around a big manufacturer's just done this, and they've been very successful.
Jon: Getting the execs to buy into the kind of algorithmic and scientific decision making, seeing what the benefits are, rather than just saying, I'm gonna do it heuristically and I know my customers better than them, and also, that sort of testing that kind of, trying it out.
Jon: I You see it in the startup world all the time. Startups live and die by testing the market and prototyping their solutions . Some the companies I've seen have done that have actually, saved loads of money.
Kulvinder: Great. Look, John, thank you so much for your time. I hope everyone who's listening and watching got as much outta the discussions as I did. I could have carried on talking to you for a lot longer. Not least around the sort of the AI predictive analytics [00:28:00] and using external and internal data sets, but we're gonna have to kind a call it there.
Kulvinder: Thank you again. Thanks very much for your time. On today's episode.
Jon: Pleasure, I really enjoyed it.
Kulvinder: Thank you And yeah, thanks sir for being my first guest as well. I appreciate that. Folks, I hope you enjoyed the episode. The conversation. Went into some different directions. I hope you answered the questions you had.
Kulvinder: If you've got any other questions, please feel free to drop a comment or send us a note. Jon where can folks find you best? On, on the interweb?
Kulvinder: Yeah. So I'm on LinkedIn, Jon Cooke data. Just do
Kulvinder: Fantastic. So thanks. Look, if you enjoyed this episode and I hope you did be sure to follow along and leave us a review. Your support helps us reach more listeners and bring you even more exciting episodes in the future. Hopefully there'll be a new episode up soon.
Kulvinder: Until then, stay curious.
Kulvinder: And thank you very much.
