Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

Data Analytics Chat

Data Analytics Chat artwork

Are Companies Wasting Money on Data & AI?

Nov 3, 202648 minAudio

In this episode of Data Analytics Chat, we welcome Lior Barak, Author & Founder of Impact Operations. After 15+ years in data organisations, Lior now helps data teams and their leadership treat data and AI as a capital allocation decision, so teams can show what their work returns, not just what it costs. Lior is the author of What Data Really Costs, a business fable about the hidden cost of data: the "zombie taxes" and "FTE debt" that organisations pay every year when business teams spend their time checking, rebuilding and second-guessing data. Listeners can read the first two chapters of the book here: https://liorbarak.com/what-data-really-costs We discuss how leaders should decide what is worth funding, when to stop a project, how to escape the POC trap, and how technical teams can stay focused on work that creates real business value. 00:00 Cold Open 01:24 Introduction to Data Analytics Chat 02:15 Introducing Lior Barak and His Background 03:29 Should Technical Teams See the Commercial Value and Cost of Their Work? 09:19 Where the Most Money and Time Is Wasted Across Data & AI 12:15 Why Companies Invest in Projects That Don't Create Business Impact 15:51 Questions Leadership Should Ask Before Funding a Data or AI Initiative 18:50 Building the Business Case Before the Team Starts Building 24:00 Communicating Value Effectively to Senior Stakeholders 27:29 What Are We Actually Getting Back for All This Data & AI Spend? 32:09 Deciding Which Projects to Continue, Deprioritise or Stop 34:19 The POC Trap: What Makes a Project Production-Worthy? 38:29 How Technical Teams Can Work on Problems That Matter, Not Just Tickets 39:52 What a High-Performing Data & AI Team Looks Like 45:22 What Leaders Should Audit First on Their Roadmap Connect with our guests: Lior Barak, Author & Founder, Impact Operations: https://www.linkedin.com/in/liorbarak/ Thank you for listening!

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show. Tap a time to play from there.

Lior BarakToday I'm doing a lot of freelancing job, but also just released a book that called"What Data Really Costs." And it's talking about exactly the problem that as a fable rather than a manual regarding basically what we're paying on data and the hidden cost of it, that all of the zombie taxes that we're paying in a yearly basis. And this is what I'm called the, the hidden cost of data, or in my term, it's the FTE debt. And the idea behind it is it's a full-time employee debt, that the time that was supposed to be focused on generating campaigns or creating tickets for the tech- technology team, for the tech team, sorry, is actually investing, invested in data work that was supposed to go under the data team.

Lior BarakAnd we always say data is the new gold. Data is we are data-driven. But what we are doing today in the industry, in many cases, is not being data-driven, is not treating data as gold. It's just treating data as a technical thing. I need to move a box from point A to point B, rather than I need to actually now figure out if it's actually making it profitable for me to do it

ben parkerHello, everybody. I'm your host, Ben Parker, and welcome back to this week's episode of Data Analytics Chat. I guess sometimes we just need to reflect and acknowledge that this is an amazing period across data and AI. There's obviously a huge amount of money being invested into the industry. Obviously, it's created many opportunities, but also challenges. And I guess with more investment, it doesn't always mean teams are working on the right things. So today, we're gonna discuss how leaders should decide what's actually worth investing in, how to communicate that value to stakeholders, when to stop a project, and how technical teams can make sure they're focusing on what-- the work that really matters.

ben parkerAnd joining me this week is Lior Barak. So welcome, Lior.

Lior BarakHi, Ben. Really excited to be here. Thank you for inviting me

ben parkerThanks for joining. So I guess before we dive in, Leo, do you wanna just give the, the listeners a quick introduction to who you are and your background?

Lior BarakSure. So I am a 15 years plus veteran in the data industry. Worked in companies here in Berlin such as Zalando and Idealo. Today I'm doing a lot of freelancing job, but also just released a book that called"What Data Really Costs." And it's talking about exactly the problem that as a fable rather than a manual regarding basically what we're paying on data and the hidden cost of it, that all of the zombie taxes that we're paying in a yearly basis.

ben parkerWicked. And yeah, no, obviously this is a huge topic, isn't it? 'Cause there's so much investment going in, and businesses need to be focused on the right thing'cause you can easily get sidetracked. There's-- every business now has multiple use cases. Every business unit has use cases. It's... You can't do it all, can you?

Lior BarakNo, and I think that one of the most important thing that we should look at is that we are not overspending our resources and becoming too thin in what we're trying to achieve

ben parkerOkay. So let's dive into the... Oh, Sorry

Lior BarakOh no. Go ahead. Sorry

ben parkercool. Let's dive into the first question. Should technical teams have more visibility into the commercial value and the cost of the work they are doing?

Lior BarakYes, they do. And I think, so maybe we go back a little bit to the beginning of the data industry when we had big data as the big thing that everybody went through it, right? And back then, the only thing we cared about was collecting as much data as possible. This was a very technical process, which required us to be able to connect to different sources, bring it into one place, creating some pipelines that we can consume the data. Fast-forward to the days of visualizations and machine learning getting the comeback somewhere in the 2015 area, and our view changed that now we need to process this data and make the best decisions based on it.

Lior BarakAnd we continued creating more machine learning product. Everybody were building something with machine learning. And then we moved into an era where we realized machine learning is not really returning the investment maybe or maybe yes, and we went into the dashboards and data, so we need as much data p- as possible again to make decisions, so we build more dashboards. We created more and more content for the business to consume. And all of this time, it was a very technical conversations how we're actually enabling all of this consumption and all this movement.

Lior BarakBut I think that what happened in the past two years or three years, there is a big shift that realizing that having more is not technically the right amount. It's not technically correct, that what we need, and that technical teams cannot anymore be a cost center or enablers only. Because when this happens, what is cr- what it's creating for the technical team is a very technical solution. Okay, I need to buy a new tool. I need to add more storage.

Lior BarakI need to have more budget. And the conversation is always about what they need and not enough about what they return. And when we're looking at today reality, investors cannot anymore accept the fact that there is a unit in the company that there is no clear return for it. They cannot continue accepting the fact that you're paying for five-- a, a team of five or a team of 10 or even bigger teams. And when somebody's asking,"Yes, but what are they returning to us?" And everybody answers like,"Yeah, they're enabling other teams to consume data." And this is when data teams cannot any longer be technical, and they need to be much more business-oriented, which means much more driven by the impact that they're gonna create.

ben parkerDefinitely. So should they, individuals now, should they know exactly what projects cost the organization, or is that just too much information?

Lior BarakThey need to know not only what the project costs the organization, they need also to know what it's gonna return to the organization. And again, here, when we thinking and reflecting about the way that we working in the industry in many cases is somebody's asking a dashboard, we checking if it's feasible based on the existing data. If not, we need to create new pipelines, and we producing it, and we ship it, and then start the movement of actually prove that there is a return for the expense that we just had.

Lior BarakWhile the pyramid needed to be upside down, and actually the value should have come first, and only then the cost and the execution.

ben parkerYeah. And that's something if, to be honest with you, the, the hiring we've been doing over the last couple of years, people want that more ownership now and more impact. So if you're getting-- understanding that the impact to the business, obviously, that's gonna make a-- it's one that's gonna attract your s-people to your organization'cause they want that more scope, the more ownership, and it's just they're gonna feel part of more involved, aren't they?

Lior BarakYeah, but so when we're talking about ownership, this is, I think, the key component, right? So in many cases, what we exposed ourselves to, ownership means somebody needs to own the outcome, somebody needs to own the request for it, right? And when we're talking about business units, for example, marketing department, when they're asking for a, a marketing dashboard that shows them all the campaigns that are running right now. From the data perspective, the marketing re- department ask for a dashboard, they're gonna build a dashboard, and they, for them, their perspective, they are not owning properly the dashboard.

Lior BarakThey're just, they're gonna fix it. But the ownership move to the marketing, but the marketing still looking at the data as the ones that owning the dashboard, and whenever something is broken, they are attracting or they're going to the data team and asking them to fix it. And what happened here, and this is what I'm calling the non-linear culture of data, is that we ship the product, we thought we are done with it, and then it's coming back from the business each time that something is broken. Incidents, support ad hoc request on top of it, and it's causing even more noise for the data team.

Lior BarakAnd the ownership of it now is unclear. Who is actually deciding what we stop and we don't do? Who is the one that defining the value that we're investing now in here? Because an incident costs money. Engineers putting their time to fix the product when the incident happens. Same thing goes to the support ticket, same thing goes to the ad hoc. And in many cases, we having maybe ownership over the data, so we know where it's coming from and who needs to make sure it's working, but we don't have the ownership on making the right decisions that steer value rather than just an enablement arm.

ben parkerOkay, interesting thoughts. So when you look inside organizations, where do you see the most money and time being wasted across data and AI projects?

Lior BarakI see it mostly at the business side. And so they-- when, we're looking at the data team, it's a fixed cost more or less of the people and what they doing, right? So they-- their main focus is to produce data and make it available and make it useful. When we're looking at the business, what we don't see is actually the loss of time of people working with the data. This half an hour a day that they need to pull out a CSV file and quality check the data that they received in the dashboard.

Lior BarakThis half an hour that they trying to figure out if the number they're looking at is the correct one or not, and going now to the source to compare it. The time that they're realizing that what they have in the dashboard does not fit what they need, so they need to build a new dashboard on top of it to get the data they need. And this is what I'm called the, the hidden cost of data, or in my term, it's the FTE debt. And the idea behind it is it's a full-time employee debt, that the time that was supposed to be focused on generating campaigns or creating tickets for the tech- technology team, for the tech team, sorry, is actually investing, invested in data work that was supposed to go under the data team.

Lior BarakAnd here in this, I-- and this important to say it, the data team are not doing anything wrong, and they are not to be blamed for that. It's just a culture that was created, the system that failing all of us in the cy- in, in the cycle to actually cause us to spend more and more money

ben parkerSo is there like common early warning signs that spend is getting out of control?

Lior BarakThere are many signs for that, and especially, when I talk to teams, what I see, there are cer-certain common patterns. So the first one is the data teams are very reactive, meaning they're always constantly putting down fires, incident here, support there, ad hoc there. More than 50% of the time is actually invested in, in, in fixing things rather than working on innovation. Then there is the second part that you can see as a sign is the data team burnout, because they are constantly in a run, in a maze to try to supply the business what they need.

Lior BarakAnd then the third one is the business unable to actually make decisions. And I always say it, you always see it in the management room when they're sitting there with a dashboard and everybody looking at the KPIs, and then the CFO pulling out a CSV file and say,"Okay, so this is the amount of orders we had last month. This is the sales that we generated, and this is the net income." And none of it is in the dashboard because nobody trust-- touch or trusts, sorry the numbers in the dashboard.

ben parkerOkay. So why do companies end up investing in projects that sound valuable but don't create much business impact?

Lior BarakIt's not that they-- Again, this i-I think this is a fail of system that we got used to it that we need to invest. It's the same thing as we need more computing power. We need to have a visualization tool. We need to have now a monitoring tool because it's the trend and everybody in the industry do it. The same thing happens now with AI, for example. Everybody wants to do initiatives with AI, but what is actually the value of it in compared to the cost is a question that nobody asked when we went into this project And I think that a lot of it is because we never build a system that, A, evaluate what is the value of what we are doing, B, creating the clear ownership, so who decides when and how to kill it.

Lior BarakAnd then the third part is the managing tradit-- m-mana-management tr-translation. So the management had an idea and concept, and they let the professional or the technical teams to do it, but nobody actually translate it into business goals or business KPIs. And this when things are getting lost and unfortunately losing a lot of money on it as well.

ben parkerYeah. Look, every-- businesses, everyone's investing now in like-- They, they wanna build models, but again there's always-- I think there should be more emphasis on the data foundations'cause I think that's just-- that's the ugly work, but it's obviously important. I think especially now, which is coming to trend, is that it's m- so important for a lot, especially for LLMs and agents, is you need to have good data now, don't you?

Lior BarakYes, you do. And I will add here something very interesting that I had in recent conversations, and this is my observation on where we're going. The code of the platform of the website have lost its value, is no longer as valued as it used to be in the past. The new IP of the company is actually the data. It's shifting, and this is why there is a high pressure right now on data teams to show the capabilities to actually deliver value and impact in the organization that never been there before.

Lior BarakIt's why we see a lot of teams moving under the CFO. That is because people in the organization trying to figure out how do we actually making this unit profitable, that's A, and B, how do we actually proving that we're working on the right things? And if we looking at a year from now, AI gonna be a very present thing or supreme. How d- how they wanna call it now? Supreme intelligence, right? Gonna be very present in whatever we're doing, and it's gonna reduce the friction of building code.

Lior BarakIt will reduce the friction of getting information. But the value of the company will be measured based on two main components. What is the unique value they give to the users if it's basically the product that they're selling. And the second part is the data that they're collecting and how they're actually leveraging it to create profit out of it in a better way. And I think this is an interesting direction that we are-- we're gonna basically see more and more often now.

Lior BarakAnd this is why data teams cannot stay in this enablement mode of,"Okay, we just enable teams to consume their data." It's not gonna work out, it's not gonna hold up because to enable, you can just take an agent and do that. To make an impact, you need a human

ben parkerOkay. So then, I guess before funding a new data or AI initiative, what questions should leadership be asking?

Lior BarakThere are a few questions that you need to ask. So first of all, what is it for? What will it change in the business? And what happen if we don't have it? So this is the main three questions that we should ask ourselves. And the reason behind it is to make it very clear for us what is the value of it. But at the same time, and here I'm gonna say something that I think is also important, when you answer these three questions, it doesn't mean that you just run and execute the project. Here, what you need to do is now is stop and figure out what will be the best approach in the next two weeks or three weeks to prove that what you're say-- what you're answering in these three questions that passed your test are actually returning the investment.

Lior BarakAnd I think this is an important thing to remember before we starting. And only after we've done the proof of concept and we have the return that says it's working out perfectly fine as we expected, then we can dive deep into the project itself.

ben parkerYeah. And also I think it's important to link it to business initiatives, isn't it? Because look, when we had the LLM boom, obviously everyone jumped on and built a chatbot, but then obviously it's not got the ROI, isn't it? So I think it's important to, have your business basics. Like obviously businesses still need to make money or make efficiencies, whether it's saving time, et cetera. So you still need to have that business mindset as well, don't you?

Lior BarakCorrect. Th- this is the value, this is the proof of value, right? So if it's not valuable for the organization, if it doesn't create this IP for the organization and actually change something, then why are we doing it? So if we spend 100 euros on tokens or 100 euros on storage of data, and it returned zero. And here, I ca- I can tell you a funny story. So I was talking to a CFO a few weeks ago, and he was telling me that they were reviewing the queries that people were using for their LLM models, and somebody by mistake was running a, a very expensive model from Claude, the fable.

Lior BarakAnd what they ran it for was to find the best cheesecake recipe The query or the basically the prompt cost to the organization 22 euros,$22, sorry. And they were all laughing about it and everything, and the, the person was in the room as well, and he was like,"Ah, yeah, it was me." And everybody asked him,"Did you actually-- So did you bake the cake? Was it actually worth this$22 of research?" And he said,"No, I actually forgot it, and I don't know now where the chat is." So when we're running something and we're expecting at least for it to return something, right?

Lior BarakSo if he would have at least baked the cake and it was good, they would have asked him to bring it to the office. But he didn't do anything with it. He actually lost it, so we wasted, basically we sinked$20 on nothing.

ben parkerYeah. No it's fascinating, yeah. That's a good story. So then Sorry. How should a sort of data or AI leader build the business case for investment before the team starts building?

Lior BarakSo when they need to build the data case, the first few things is actually to understand what it gonna change in the business itself, I think. So it's asking what is the value of it to the business? If we gonna do it, what we gonna see that changes? And in changing, I don't mean now the marketing department gonna stop coming and complaining that they don't have data. I mean that now the data team will be able to actually steer better the 1.2 million budget a year that they have, and most likely it's gonna create an incrementality in either the conversion of users, so we're gonna be much more efficient on conversion, and then specify what is the percentage expected for that.

Lior BarakOr we're gonna save costs that are now sinking on bad campaigns, for example, that nobody knows. And once we discuss and understand how it's gonna change the business and what they, what it does, only then we can start evaluating in this document how we're gonna execute it, what is the minimal MVP that we can build, basically. So the, the MVP, the smallest one that's gonna prove that we are in the right direction, and also the mechanisms of stop.

Lior BarakAnd when I'm saying stop, it means that in each step of the way, what will cause us to stop what we are building, and what will cause us to stop it even after it started running because we're realizing it's no longer serving the business as we expected. And this is-- I think this is a very important part that we keep forgetting, because in many cases, and also I did it many times, I build a product, I released it, and I forgot about it, and I was like,"Okay, now go and do it." But when we're talking about rollout, a rollout is not just,"Okay, that's it.

Lior BarakI introduced it in the team. They know how to use it." It's to be there present on a weekly basis and see if the use case has changed, if they're still using it the same thing the same way it's intended, if it's still returning what it was supposed to return, and if not, is it actually still covered by covering the cost that we have to maintain it? And I think that this is something that we need to start thinking about a little bit more, about capital allocation, about what is actually when we have a bank of euros, how we're actually investing it for the best returns.

ben parkerSo do you mean like defining conditions under which way you should stop a project? 'Cause, look, look, doing traditional ML projects is always getting that into production has been hard, and it's the same with AI now, isn't it? It's-- They both got challenges. So is it-- Do you think from the start of the project there should be like, like defined conditions? 'Cause once you're involved, sometimes it's easier to carry on, isn't it? Do you think there should be like hard conditions where you go,"No, this is not the right project"?

Lior BarakYep. Yeah. So when we're talking about it, it needs to be conditions for stop. So usually what we're doing in the traditional way is like conditions for continue, for build. We never do a condition for stop, and this is why we finding ourselves with 1,000 dashboards sitting in our visualization tool layer. This is how we finding ourselves with 400 different pipelines. This is how we finding ourselves sitting on 60 or 70 more sources of data that are streamed inside our data warehouse, because we never actually defined when we stop, when it doesn't make sense anymore to bring this data, what it means actually to organization.

Lior BarakAnd I will give here for example, a very simple example. We in a company I worked with, we ingested a lot of data from third party. And if it was advertisement platforms and attributions and so on that we were working with, and we ran on top of it a monitoring tool, and we stored it, and we were processing it on a daily basis. And then I asked them,"Can you actually check what is the percentage of this data that was accessed in the past, let's say, six months?" And what I discovered was that almost 60% of the data that was ingested was not actually consumed.

Lior BarakBecause while we were ingesting it into the data warehouse, we brought it inside, we monitored it we controlled it, we processed it, the business didn't see the value of it because they needed specifically only on a period of a Black Friday. And when we computed the cost of these sources, realized it's a quite expensive operation, and is it actually making sense to keep running it or not? And the finance decision was very simple, kill it, because we don't see the reason.

Lior BarakFrom the engineering perspective, it was a very simple engineering. We just were, we were asked to actually process it and bring it on board

ben parkerSo then, like a, a lot of teams can explain I guess what they're building technically, but struggle to explain why the business should care. How can you communicate the value effectively to s-senior stakeholders? 'Cause obviously sometimes our ROI is difficult to determine, isn't it?

Lior BarakI will disagree about that. I'm gonna say that the costs are clear for the data team. They know exactly what they invested. The revenue, it's up to them now to go and ask what is actually changed based on the data that we have. And even if it's an estimation, it's better than saying we need the perfect number for getting an ROI. Because if I know that we're steering 100K a month of marketing budget with this data, and due to that, we managed to optimize 20% of the campaigns and cut costs, it's a good base to already start talking about it and say,"Okay, there is a value here," right?

Lior BarakI generated something. And now the conversation is moving from it's hard to do a ROI to the marketing team to prove actually that there is higher revenue that what we managed to understand because they want the product to continue working. And it's creating a culture of supply and demand. I supply the d- the da- the data because there is a demand for it, but at the same time, I'm not gonna supply it if it-- if I'm losing on it money. It's the same thing as in being in the market, right?

Lior BarakSo nobody gonna offer you tomatoes for euro 50 when he knows that they bought it for 2.50 So why are we doing the same thing with the data?

ben parkerInteresting. So obviously I speak to obviously a lot of leaders on here and obviously through work, and it's, yeah. Obviously, co-companies-- interesting pers-perspective. Some say that ROI is difficult to determine, and also it's good to hear your thoughts against it. So it's interesting to hear. It's nice, good to hear ev-people's opinions. So do you think it's just more that it's more for a difficult conversation with stakeholders then? Or do you think they just need to, Are they struggling to just understand I guess translate the business, the technical projects into business terms, do you think?

Lior BarakYou just touched the right exact point. We treating project as technical thing and not as a business. Whenever we're getting a project that asking for a dashboard, we're looking it from the perspective of the technical aspect. Can we actually make it available? Yes or no? Is it working for us? Yes or no? Can-- Do we have all the data we need? No. What do we need to do now to bring the data? We never start with the question of what is the value of it. We never asking ourselves what it's gonna change in the business.

Lior BarakIt's a, it's changing the language and the perspective of how we actually treating data. This is not a call center that answers calls. It's not about closing tickets. And I know a lot of people now gonna get pissed off about what I'm saying. But data, I-- And we always say data is the new gold. Data is we are data-driven. But what we are doing today in the industry, in many cases, is not being data-driven, is not treating data as gold.

Lior BarakIt's just treating data as a technical thing. I need to move a box from point A to point B, rather than I need to actually now figure out if it's actually making it profitable for me to do it

ben parkerYep. Interesting take there. So I guess what would happen when a CA, CIO or CFO asks what actually are we getting back for all this data and AI spend?

Lior BarakThen we need to start auditing. So I was talking to a company a few weeks ago, and what happened there, the CFO the ownership of the data moved from the CTO to the CFO. And the CFO asked to audit all the dashboards that they had, and they had around I think 460, 470 dashboards. And after the auditing, what left from the CF- CFO perspective was only 10 dashboards. The-- All the rest of them got deleted, then the visualization tool contract got canceled because there was no value to it whatsoever, because nobody could prove the value.

Lior BarakAnd I think that we try to prove the value after we gave the data, but we never try to prove it before it. And then it's even harder to chase down the value once you already release something out there. And I believe that the CFO today is a much closer mindset to where data is going as an IP than, for example, a CTO that's looking at it as a technical way. But at the same time, I think that the only one that can literally understand and connect all the dot is the CEO, because they own the business vision.

Lior BarakThey know where the business needs to go, and they needs to be able to evaluate how the data that they having is actually changing the direction that we are going towards. Is it too late to say the opinions are only mine, and if anybody has complaining can come to me?

ben parkerThat's cool. So I, it's... I think it's just, it's a lot of it is more communication skills, isn't it? Like speaking across the business,'cause everyone's got a different perspective, haven't they? Like obviously CTO is gonna be more, more technically driven. Obviously CFO more business-focused, finance. It's, I think it's more, it's getting that communication aspect I think. 'Cause it, like any data, the problem with data projects, as soon as you start them, you're constantly chasing your tail, right? It keeps changing. So s- you need just to understand where you're at. And obviously the... I guess it's just getting, focusing on like the key project, isn't it, at that time.

ben parker'Cause you can't keep chopping and changing or... But what's gonna, what's gonna get, bring you the back, the most value now? What's gonna make the biggest impact to your business? And then go from there. 'Cause you can't do too much, or you like... You've heard stories, people trying to do 10, 20 projects and then, yeah, it just goes tits up.

Lior BarakYeah. I think that we need to live in a reality that says and we got used to it, that we have long cycles of build. The only long cycles of build that we can allow ourselves is when it's coming to a platform tool. So when we're introducing a mon- a, a monitoring tool or when we're introducing a new processing engine, and so on. When it's coming to products like AI, like dashboard, and this is something I'm saying already for the past 10 years, it needs to hit the market as soon as possible.

Lior BarakNot yes, we try to do the best," and then come with it in three months or in six months or in a year. It needs to hit the market in a week or two. So in one sprint time, it already needs to have something that somebody can play with, even a, an Excel sheet, and understand already what is it worth for them and how they-- how it actually changing how they're doing things. Because this is how we're starting to evaluate basically value, right? This is the core principle of understanding what is the value of what we are doing to someone.

Lior BarakAnd only after we acknowledge that and the PCU actually prove... The PCU. The proof of concept, sorry. Prove itself, only then we are enabled to continue this next step. And also the next step needs to be another small iteration, another sprint. But here there is something that I need always to say, and we go back to this nonlinear culture of data. While we're trying to do innovation, there is always things that holding us back that are the ad hoc, the supports, the maintenance that we are doing, that is always blocking us from running to innovation 100%, because we still need to maintain whatever we carry with us in our portfolio from past products.

Lior BarakAnd this is again, why we're going to this when or what we stop question always. This is why it's so important to ask it, what do we stop? Because at some point, carrying so many products is actually causing us not to be able to actually produce the POCs that we need to move as fast as possible.

ben parkerSo then how should leaders decide which projects to continue, which to deprioritize, and then which ones to stop completely?

Lior BarakIt's a, a simple and complicated system at the same time, I think, because there is a lot of politics here, and what we need to come with, and this is the tough part, is to come with hard numbers that actually showing the value. They're showing the cost. It's easier to compute what is the effort that is existing there. So I worked with many data teams, and I always had the same conversations. We cannot we are unable to estimate how long it's gonna take us to develop before we're gonna start developing it.

Lior BarakWe are unable to know what it's gonna cost us before we having the first data there, and then we'll be able to know what we can do with it. We are too-- We are unable to know what is the return of it because we don't know. It's part of the business. The business needs to explain it to us. But if we want to decide what we continue building, what we are stopping, what we're putting on parking lot, what we need to do is actually, is to have these hard conversations because this is business conversation, and it's not a technical one.

Lior BarakIt's not,"Can I execute it with my Athena or do-- or Snowflake or whatever, or if DBT will support it?" It is a question that asking, what is it actually making s- is it actually making sense to, to continue doing it, or is it conti- is it making sense to even build it? Then we can talk about the technology. The technology needs to come at the end, not at the beginning, because many products that we've built, and I built many of them as well in the past, are products that if I needed to evaluate them from my perspective today of what I'm doing, I would've canceled them because they don't return what we invested.

Lior BarakThat's A. And the cost of it for the team is too high. This is why they're becoming so reactive because of the amount of carriage that they need to carry.

ben parkerDo you think too many teams spend too long building POCs without being clear on what would actually make them production worthy?

Lior BarakSo usually what I'm and what I'm preaching for here is do three to four POCs. So you're spending let's say between a, a month and a half to two. And then between these three and four, you decide what you exploiting, so what you keep building on top of it and expanding it. Only one of them. And in this system, what it means is that you have slots that you can insert inside it certain projects.

Lior BarakYou need to test them, and only then you can keep building them forward. And I think this is an important concept that we forget constantly, because what we do is we sitting in the data strategy meeting at, let's say, the end of the year or the beginning of the year, it depends on how you're doing it in your company, and we're creating a shopping list. This is what we need to build this year. This is what is super important for us to have. And by month two, we arrive and we're realizing there is no chance we're going to achieve any of the shopping list requests.

Lior BarakAnd then we're moving to a reactive mode. We're just moving to who screams the loudest And if we are not structuring it in a way that we're saying,"Okay, we're exploring now only-- we have only three spots on the table for Q1 to decide what we want to explore, and we have only one spot to continue building towards Q1, Q2," our entire system is collapsing because we're constantly reacting to whatever happens around us.

Lior BarakWe didn't clear the board in a clean way. Plus, the advantage for this, for the business is that they know what you're gonna test, and they know that the decision is no longer an emotion or a feeling. It is baked with numbers. You were unable to prove that it's gonna change anything in our cost or it's not gonna change anything in our revenue, so this is why we didn't continue with this specific product. If you can prove it, we'll be happy, more than happy to experiment with it again and see if it's actually it's being driven forward.

Lior BarakBut this is not gonna happen in this quarter anymore because we already committed to three and four tests and one thing that we know for sure that we will need to build

ben parkerSo do you think when organizations, like, struggle to kill projects, is it just'cause they've already committed too much money, or is there... What's our guess? Is there a particular reason that organizations just can't end projects sooner?

Lior BarakNo, it's because there is no ownership. Nobody ever defined who owns to kill it. So we keep building, we keep maintaining, and the building is cheap. Building a dashboard is a cheap thing. To collect data is becoming a cheap thing, right? S3 is very cheap thing to use to contain data. The compute costs, in some cases, when you're optimizing and there is data teams that always optimizing queries, can be also feel very cheap. The visualization tool that you are using, you don't look at the cost of license in many cases.

Lior BarakYou look at the cost of computing the dashboard there. It seems to be also very cheap. But in reality, each new product you're adding to your portfolio is new commitment from the data team that blocking them from doing innovation. But nobody defined what and how we decide to kill things. And then there is this culture of spring cleaning that once a year, the data team say,"Okay, we cannot do it anymore. We have too many dashboards.

Lior BarakWe need to remove them." And everybody fights to keep their dashboards, and then you manage to clean, I don't know, 20%, and everybody happy and everybody cheering. But the 80% that's left behind, you're gonna still build on top of it new products, which means it's gonna do a comeback, and the cost of maintaining it, nobody actually seeing it in many cases

ben parkerSo then how, okay, so how can data scientists, ML engineers, data engin- data engineers make sure they're working on the problems that matter rather than just completing tickets?

Lior BarakTheir question needs to be closer to the business. And I think this is one of the most important thing today, is sitting with the business and understanding what they're trying to achieve, and then figure out how they build it for them in a way that they are able to use it based on their knowledge and based on their needs. So it could be that it's something that they need only once a quarter, or it could be something that they need daily. And then this is where the machine learning or the data scientist needs to literally sit with a business unit and understand how they're using data.

Lior BarakAnd you know what? In many cases, I will tell you that I discovered that the business were saying they need to solve X, but when I was sitting with them, I realized that actually Y was the problem, and X was just a symptom that they thought that if they're gonna solve, if they're gonna deal with it, they're gonna solve the problem. And I think this is one of the, the main components that I am keep encouraging more and more engineers and more and more analysts and even product managers to sit down with the business and see how they're doing their day-to-day. Become a marketeer for a week.

Lior BarakAnd trust me, you're gonna figure out much more than what you do during the discovery discussions with them and all the meetings that you're sitting trying to understand what they need.

ben parkerThen what does a high-performing data and AI team look like when everyone understands where their time should be spent?

Lior BarakA high performing team means they are not reactive. There is always a reactive percentage, but it's very small. It's much more controlled. There is a clear stop decision, so they know when they stopping s- to do things, and they knowing also what is the value. So even if you're gonna wake them up in the middle of the night and you ask them,"Where are the keys for the submarine?" No, I'm joking. Not the keys for the submarine, but where did the product value sits today? They will be able to point you at two to three products that actually having already a proven value and tell you this is the main drivers of the business.

Lior BarakThis is for me high performance team. And I know this going against a lot of the industry standards, and I'm really sorry about that, but I truly believe that we can change and we need to change

ben parkerI guess it's a leader's role to create this clarity, isn't it? What needs to be delivered and, yeah, where a team should be spending their time

Lior BarakI think that the first thing for data leaders is to stop looking at shopping lists and start looking at business value. And this is the starting point. The second part is to take their team and embed them. And I know everybody's saying,"Oh, but then it's creating silos, and it's very bad." So don't do a silo. Don't just take one engineer and leave him with the marketing team for the rest of his career. Take them once a month for two to three days to sit with a business team and learn how they're working with the data.

Lior BarakLet them teach you what they are doing and what is their tasks, and then bring it back to the organization and think how are we actually improving what we existing. And as a data leader, you also need to ask the question, am I having the right people right now to answer the needs of the business? And I think this is becoming a much more critical question for all of us

ben parkerWell, in terms of... 'Cause on, and obviously the last couple of years, the sk- the s- skill set has completely, it's drastically changed, isn't it? It used to be, even three years ago, it used to be, hiring we s- it's... A lot of the focus was on technical skills. Now it's definitely on business, product mindset like unders- getting that understanding. And that's the complete culture change, isn't it?

Lior BarakCorrect. Correct. Correct. Correct. It's a lot about the product. It's a lot about are they able to actually, reflect how they changing the existing products and move them into a value. And a lot of the good engineers that I'm meeting, this is exactly what they do. They sit there, they see how the usage is happening, and then they reflecting on what needs to change to make it better. So if you know that there is a dashboard, and here I'm going again to collect data for the data team itself, understand how many support tickets you have on product, understand how how many incidents you have for products.

Lior BarakBecause when you know that, you know also where you need to focus and where you need to send the right people to investigate and understand better why the incidents happen, why the collapsing are happening, and are we actually reducing it. And if we know what it actually costing us, and we know the return of it, the value for the business, and how it's actually changing something and what it's changing, it's easier to actually then know where we need to prioritize things. So the s- the I think that the skills are now really much about how in, from leadership perspective, how am I managing my portfolio to be a healthy portfolio?

Lior BarakAnd we gonna take here I will bring this great example of do you wanna run a buffet or do you wanna run a Michelin star restaurant? You have these two options. When we're talking about a buffet, it means that you're just serving many dishes to match many different taste buds, and hopefully the guest will find at least one or two dishes they're gonna enjoy eating. In the Michelin star perspective, when you're going into a restaurant, and Ben, I, I hope you have been to a Michelin star.

Lior BarakDid you have the chance? It's a story, and each dish that is served to you is curated specifically for this event. So when you're getting the menu, it's not a buffet that you go and eating your potatoes and you're enjoying it. You're eating different dishes, and each of them is telling a story and building up your experience in the restaurant. When you're coming out, you're not saying,"Oh, the potatoes were the best thing in the buffet." Each dish created something in you and left you a memory, and this is exactly what data leaders needs to think about whenever they now creating these products.

Lior BarakWhen we're talking about engineers and data scientists and analysts, their perspective now needs to be, how do I bring the best dishes to our clients that is actually creating the right impact in the right costs? So yes, you can go and source a, a langoustine from a, I don't know, a Alaska to y- to Berlin. The cost of it gonna be crazy. Is it actually making sense to do it? Yes or no? And most likely not.

Lior BarakSo find something that is local here and you can cook with it that gonna make the right dish for the people. And I think this is when we're talking about, like, where are we going towards with the seats-- with the set of skills.

ben parkerI like your example there. I'll use that. So then j- I guess to end on then I think you did ta- tap into this a minute ago obviously discussing about l- what leaders building cl- clarity. But I guess say for leaders looking at their entire roadmap tomorrow, what would you tell them to audit first to make this, to make impact?

Lior BarakFirst of all, their gut feeling. Trust it, what they think that doesn't make sense to have on the menu, and then clean it out. Then the second part that I will say here, and I think this is, I think something that is important to say look at it as a story that you wanna tell, not as a shopping list or a buffet dishes that you need to maintain. What is connected to what, how it's telling the story, and how it's making sense, and why it's making sense to produce it. So when they're auditing their, the list tomorrow, this is what they need to reflect on.

Lior BarakAnd then they need to reflect on does it make sense in the cost to produce it, and will it actually bring us the right return, and the user's gonna enjoy it? I remember that I was in a Michelin star restaurant where they served something that was very strange in my taste buds, and I didn't like it. And the chef then came and said"This was an experiment. I just wanted to see if people actually gonna like the dish or not." And he asked for the feedback. he was there, and then when I said it, he was sitting down with me, and he tried to explain-- he tried to understand from me why I didn't like it, what didn't fit, what didn't work out.

Lior BarakAnd I think that this is also what they need to sit down and do on their lists. Sit down with the people, understand what is working and what not, and why people don't like what they're being served today

ben parkerBrilliant. Lior, it's been a pleasure having you on and some, yeah, great views and examples shared. So thank you for your time

Lior BarakI appreciate it. Thank you for your time I hope there is a value here, and I will be more than happy to discuss it even more if anybody's interested

Open the transcript file

Hosts and Guests

Nobody is credited on this episode yet.

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. What It Takes to Become a CIO Today

    Nov 10, 202638 minAudio

    In this episode of Data Analytics Chat, we welcome Saadia Khilji, Global Chief Information Officer at Coverys, a medical professional liability insurer that has served the industry for more than 40 years. Based in...

    Credit the people on this episode

  2. Your AI Is Live. But Is It Actually Working?

    Oct 27, 202639 minAudio

    In this episode of Data Analytics Chat, we welcome Ryan Stevens, Senior Director of Applied Science at Ramp, the financial technology company helping businesses save time and money across cards, bill pay, expense...

    Credit the people on this episode

  3. Nestlé’s Chief Data & AI Officer on Scaling AI

    Oct 20, 202649 minAudio

    In this episode of Data Analytics Chat, we welcome Jessica Matheron, Chief Data & AI Officer at Nestlé. Jessica brings a rare mix of experience to the role, from a French engineering and mathematics background and a...

    Credit the people on this episode

  4. AI Is Moving Fast. Is Your Data Ready?

    Oct 13, 202646 minAudio

    In this episode of Data Analytics Chat, we welcome Erica Howard, Head of Consumer Data Ownership at TD. Erica has spent 18 years at the bank, starting in the call centre before working on TD's early data governance...

    Credit the people on this episode

  5. They Tried 3 Times. Then AI Changed the Enterprise Data Build

    Oct 6, 202646 minAudio

    In this episode of Data Analytics Chat, we welcome Mary Alfheim, VP of Data and Analytics, and Joseph Kroon, Data Architect, both from Headspace. Mary previously led analytics teams at Amazon Prime Video and HBO, while...

    Credit the people on this episode