Managing Marketing: Data transformation within your organisation

Kshira_Saagar

Kshira Saagar (Pronounced Shee-Raa Sa-Ga) is the Head of Analytics and Data Science at The Iconic and a speaker at AdTech Sydney 2019. Here he talks with Darren about the business intelligence and strategy role that data plays in driving business performance across the whole organisation and why an organisation must transform their culture and processes to fully deliver the results data can provide.

You can listen to the podcast here:

Follow Managing Marketing on Soundcloud or iTunes

Transcription:

Darren:

Welcome to Managing Marketing. Today I have the opportunity to sit down and have a conversation with Kshira Saagar who’s the head of analytics and data science at The Iconic, and we are actually here at The Iconic. Welcome Kshira.

Kshira:

Thanks Darren, thanks for coming all the way here.

Darren:

It’s my pleasure because I’m actually looking forward to hearing you speak at AdTech in Sydney which is next March.

Kshira:

Yeah that’s right.

Darren:

I don’t think you’ve probably started putting your presentation together.

Kshira:

No but I’ve pitched the idea and they really liked it so it’s kind of already in my head.

Darren:

I can understand that because I was told it’s called transforming your whole company into one big data team and I think that’s an incredible thought; I’m just wondering about the practicalities of it. Has this come from experience or is this something you see as the opportunity?

Kshira:

It comes partially from experience. It comes from the way I see data being used and the philosophy of how we look at data. The way I look at it is like a supply chain not because of any particular affinity to it, but it just makes a lot of sense. When I mean supply chain it’s called a data supply chain. Data is raw material, think of oil, think of coal, think of anything, it’s the raw material. Data is the new oil, it’s crude oil.

In this day and age so many systems are producing data. Everybody who buys a system has data now. It’s no longer a paucity of data. Nobody has no data, everybody has a lot of data and we are drowning in it so data is the raw material. Now the end product in terms of supply chain is decision. At the end of the day what you do with this data is you need to make a better decision with that.

Darren:

I am so glad you said that because people think the output of data is insights and it’s one of those terrible words that you go what do they mean by insights? But it’s actually the most important role, to inform a decision.

Kshira:

At the end of the day it’s to inform a decision and if you believe that way then you believe data teams or data organisations in a company are basically a servant organisation to the business to help them make better decisions. So a raw material called data becomes a finished product called decisions and there are some processes in between. Think of a chemistry lab set up like a sulphuric acid set up and there are three big blocks in my head.

The first block is data engineering. Just like crude oil needs to be brought all the way from the ground up so you can access it and actually touch it and start seeing what’s there rather than talking about this mythical Loch Ness monster called data. Data engineering for me is literally pipelines and building the architecture, putting in a data base, cleaning it up so that everybody else can use it in the business. So getting data from your source systems to a place that people can use it; that’s data engineering at a very high level.

Then, once data is available then the next part is the data science part. With data science I mean somebody can apply intelligence on that data at scale. It doesn’t just mean AI or Ml or that sort of thing. It could be a simple dash board, it could be anything.

There are three levels of data science. One is the algorithmic part, which is the AI/ML part, the analyse part, which is basically deep diving into it and figuring out what is happening, and the third part is the interpreting part, which is why did it happen and trying to put some reason behind it, some storytelling.

Darren:

So to go back to your metaphor of crude oil. I see that part as a bit like fractionation where they pour the crude oil, which comes out as this black stuff of average consistency—they put it into a fractionation plant and it just separates. So, you’ve got at the top the most volatile, then around the middle your petrols, your fuels, and then you’ve got your oils and lubricants, and then down the bottom (this is the bit I love) the sludge that becomes the bitumen on our roads.

And each one has an important role but until it’s actually fractionated there’s not a lot you can do with crude oil. I like that metaphor.

Kshira:

We break it down and then we apply intelligence at scale and the outputs of that could be multiple things. It could be an algorithm, an analysis, a dashboard, and there’s still a final part. People stop at that part. They apply data science, data engineering but the final part is the data translation—what I call translating maths to English.

So, when you ‘re translating maths to English you’re basically translating data into an understandable language. This could be through memos, Saturday documents, dashboards, models, visualisations. It could be anything the business understands and makes sense.

So those are the three parts. And when you do that translation the business is going to say, ‘I understand all of this; I’m going to make a decision’. So the end point is the decision.

Darren:

And then that feeds back into impacting the data you’re collecting in the first place.

Kshira:

I like to call it the ‘virtuous cycle’. The more decisions you make the more data you want to collect, the more data you collect, the more decisions and so on. That’s how we think about the data supply chain. Thinking about it that way what tends to happen is then how do you get this done?

There are a lot of systems for data and a lot of people want to make decisions out of this data translation. People tend to hire a lot of data scientists side by side with analysts and it’s so difficult to hire people who can do this at scale. When you start doing the cycle and it becomes faster and faster you cannot keep hiring more and more people to do data.

What if we could move data engineering to the people who get the data out, which are basically the tech teams and the developers who build these systems? What if you provide a way that they can directly get the data from the ground up to a place where we can all see without having to do much data engineering?

Darren:

So streamline your supply chain?

Kshira:

Yeah. So they can get it to the market directly and what if we could make data translation much easier through simulations and optimisation dashboards and all of that? The business does not ask you to keep building that simulation so they can make decisions much faster.

They can translate the data maths to English by themselves and if they can insource both the data engineering and data science to the business suddenly you have the whole business being a data team and you can focus on the parts where you are playing intelligence at scale.

And that’s what we mean by getting the whole company to be a data team rather than just one data team.

Darren:

So everyone plays their part, their role in actually moving that process forward from collecting it to preparing it to interpreting it and then making a decision and then going around again?

Kshira:

So, when everyone is involved everyone feels accountable and proud of what they’ve done. It’s no longer ‘oh it’s a data request—just do it’ and then it comes back. It’s not a transaction thing; it’s more of a harmonious thing that happens.

Darren:

One of the things I like about that model and it overcomes the single biggest thing I see in organisations is that as soon as you use the word data there are those who feel numerically literate (and this is not two groups—it’s a continuum) and then there are those who absolutely abhor anything to do with numbers.

I think it goes back to secondary or even primary school when people were taught mathematics badly; they suddenly went it’s all too confusing, I don’t understand it, it’s not for me. Clearly you were a STEM (science, technology, engineering and maths)—that was your focus at school.

Kshira:

Yes, That happens a lot. Either there is this prior feeling of data being too intellectually or technically complicated—it’s one of those two things. One thing we continuously try to do is demystify it and tell everyone ‘this is just applying logic at scale’ It’s a macro scale; it’s a matter of applied logic.

And any human being can apply logic. By logic we mean; someone who has done this, this and this could have done this. That’s logic. And we apply that through programming and we try to break it down into simple logical steps and people start embracing it. It’s less numbers and more logic and that’s how we try to sell data as a solution within the Iconic group.

Darren:

There’s a TV series called ‘House’ about a doctor who diagnoses. He has this saying ‘people lie and results don’t, numbers don’t; people do’. One of the big challenges people have, especially those people who don’t like and are not comfortable with data is their concern over the impact on their right to have a ‘gut feel’ or an intuition but you’re saying that the two go side by side.

Kshira:

I’ll give you an example of what I mean by that. Take somebody at a warehouse, an operational team trying to work on an assortment algorithm that we’ve built. That algorithm will tell you if you scan an item where you put that item so that it can be pegged and delivered efficiently. So it helps you assort items in a more efficient fashion.

Now the standard operating procedure is here’s the algorithm, here’s the gun, scan the item, it tells you to put it in X but nobody is going to buy into it because people feel they know how to run their warehouses better.

So we opened up that algorithm to say these are all the components that are considered to make the decision, the logic we use to make the decision. And we can see people asking why we considered that particular fact or feature—maybe shoes should not be treated this way. So it had some input from them.

It makes a lot of sense because they know their area much better and we know our data much better and when those two things marry they no longer feel it’s our algorithm, it’s their algorithm too. And they call it their algorithm not the data team’s algorithm, which is exactly what we want.

We want people to feel it’s something built for them, it’s theirs and they can understand everything that’s happening and can start questioning us on specific factors of the algorithm and not the algorithm itself or why we should use it.

Darren:

That’s a really clever way of approaching it because the term algorithm gets thrown around a lot, especially Facebook and the like talk about algorithms. But they’re not particularly disclosing at to what those algorithms are and they almost come with a stamp of infallibility.

But in actual fact an algorithm is purely like an equation that is trying to mirror or create a model of what’s happening in real life.

Kshira:

That’s true. The root word for algorithm is Alger which means pain so I like to sell algorithm as it will solve your pain. That’s how we like to think about it.

Darren:

It’s an algorithmic analgesic. Well it can work two ways can’t it? One is it starts to make it easier to replicate but the other is that if there are any issues or problems you’ve at least got a base model that you can either see there is something wrong in the reality or something wrong in the algorithm and we need to adjust it to reflect reality.

Kshira:

And the other motto we have is ‘let human beings do what human beings do best and let machines do what machines do best’, which means there is a lot of groundwork involved in making a decision when you consider there are 25 other factors and an algorithm can do all of that for you so you can just make a decision.

Human beings can spend that time doing something that’s intellectually smarter and better. That’s how we have always progressively done stuff. That’s what our algorithms do. It will take away all the decision-making process and it will help you just press yes or no.

Darren:

It allows you to make the final decision. It’s interesting because the human brain is a phenomenal data processor that doesn’t necessarily follow a logical process. In engineering they call it fuzzy logic because it’s as close as you can get to a human decision making process.

We get huge amounts of data. We’ve produced more data in the past two months than since the start of time. It does mean that things like artificial intelligence are important because it will allow huge amounts to be processed and brought down to a human brain level of comprehension.

I want to ask you a personal question; how many languages do you speak?

Kshira:

Spoken human languages? I speak around 5 fairly confidently.

Darren:

I would say 6. The reason is I believe that mathematics is a language.

Kshira:

Fair enough.

Darren:

Every mathematician I know sees mathematics purely as a language and it was an interesting insight for me because when you think about it as a language in its own right, not numbers on a page it suddenly becomes like when I listen to other people conversing in other languages, it’s really fascinating.

Rather than thinking of mathematics as something to be concerned about that it is just another language.

Kshira:

That’s true. People who are into music can see music in maths.

Darren:

They are completely linked. All music is based on mathematical principles.

Kshira:

The golden ratios.

Darren:

All of science has mathematics. That’s why it’s seen as a pure science because every scientific theory at some point has to have a mathematical proof. If you can define the theory in a math proof it’s seen as a proven theory.

Kshira:

That’s true. Interestingly, a small detour, have you read Homo Deus by Yuval Noah Harari? There is one section in the book where he talks about how knowledge in the Middle Ages was about scriptures multiplied by logic—people interpreting scriptures through logic. So if you don’t know scriptures you have no knowledge and if you have no logic you can’t interpret scriptures.

But then in the scientific age it’s data multiplied by math. If you don’t have data you can’t apply maths to it; if you can’t do maths you can’t apply data to it and that’s knowledge. That is what has brought us a lot of knowledge about the world—applying maths to that data at scale. Now it’s all about sensitivity multiplied by experience, which is taking that maths to one more level—experiencing that maths for yourself.

Darren:

And bringing the human condition because we are emotionally driven beings. It’s so important. I think it’s one of the reasons why behavioural economics has grown so much. I love that term ‘predictably irrational’. Science is starting to define the human condition and acknowledge that we are who we are but all of these things are tools that we’ve created.

It’s interesting you started off in the Middle Ages because you have to remember it was also a time when those who had the knowledge were desperately trying to keep it to themselves. The religious leaders and the kings and emperors wanted to keep the knowledge to themselves.

The Enlightenment and the Renaissance was still giving that mainly to what became the middle classes; the affluent, wealthy, and the educated. But one of the great things about the last half of the 20th century and now is the way that knowledge has become so much more democratised.

The internet for all its flaws was designed to make information available to everyone and it has done that. I think the only wrong decision was allowing anonymity because it’s led to crime and all sorts of activities –it facilitates that as well.

Kshira:

That’s a fair thing to say. With so much access to data and democratisation of everything that’s exactly the principle we’re trying to mirror with how we approach data.

The Medieval Ages for data was probably 15 or 20 years ago—basically a few people in the company could get that data and keep it to themselves—data fiefdoms where nobody could get access—only people who knew the password. It’s come through a process exactly mirroring the same knowledge approach.

And what we’re trying to do with data is everybody should have access to data. Everybody should be able to understand and estimate and make decisions based on the data.

Darren:

So in my team of consultants (I have a really eclectic group of people) and a couple of them like to go straight to an excel spreadsheet because they can see the patterns in the numbers. Then there is the group that can’t really follow that—please turn it into a bar chart, make it visual for me, and then there are others who like storytelling—tell me what this is actually saying.

When you get to that step (going back to your supply chain) of you’ve processed it and made it available then you use the word translate. Using my metaphor of maths as a language; it needs to be translated into lots of formats; visual people, numbers people, and then the storytelling, sense making people. Have you found you need to accommodate all those?

Kshira:

Yeah, so we try to accommodate all the way down to people who lip read as a story. At Iconic we have this principle that everything needs to be in the form of a memo. That means it describes how did the problem start, what we have done, the numbers, the tables and then if people are interested there are graphs and dashboards. It’s a whole document in itself. That’s how we try to translate that data back. People can make a decision based on I’ve read all of this, it makes logical sense, let’s do it.

We try to cater to all kinds of people. The people who want data they can go to a dashboard and get that data. There are some people who don’t want to see graphs, who hate point and click.

Darren:

They want a piece of paper or something they can sit and read.

Kshira:

That’s one extreme and at the other extreme is the people who if there is anything other than a black terminal and green text they will not do it. They will not touch it. I’ve worked with people who if it is point and click, dual item, I’m not using it.

They strongly believe in a terminal base and we cater to those people too. We cater to people who want point and click and to those who want to get their hands dirty, to all levels.

Darren:

So when you used the term translation you literally meant translating that information. I’m surprised you haven’t ended up with one of those graphic novels—almost like a comic book.

Kshira:

We try to add all these philosophic quotes to our memos so they read almost like a storybook. We try to keep what is the problem, the situation, the questions asked, the answers so anybody who comes and joins us and after two years wants to know how we made that decision they can read it and understand how that decision was made and the data points that supported it. It’s no longer about the data and the numbers; it’s about what is the business problem.

Darren:

It’s what is the story, the insight, the information that is going to inform the person who has to make that decision?

Kshira:

And you don’t need a narrator narrating that or somebody clicking on a PowerPoint explaining it. It is a whole compendium in itself; it explains everything including the footnote that says ‘ABS means average basket size’ so people don’t get confused by what the term means.

Darren:

All the 3 letter acronyms. Probably the finance industry is the worst at that. They create them almost as if it’s gold.

Kshira:

That’s their version of keeping knowledge to themselves.

Darren:

A lot of organisations really do struggle with that don’t they? That hierarchy: data, information, knowledge, and wisdom is at the top of that pinnacle. Someone argued recently that it’s not a linear process; that you can go from information to knowledge, that you don’t have to go through those steps.

But there are organisations where you see the data team is almost impenetrable and over here is the research and insights team. As part of building that supply chain how do you get those silos to break down?

Kshira:

One answer is I was lucky. The other answer is I got the opportunity at Iconic to build it all the way from scratch again. When I joined we had two people in the analysts’ team working with me. Now we have almost 20 people added working in this team.

What that has done for us is we have set it in a way that is completely cross-functional, i.e. there is no data science team, engineering team, data analytics team sitting separately. It’s all broken down in terms of the mission we want to serve; are we solving a customer problem, an internal problem or a marketing problem?

Basically we have three big divisions like that and then within each team there is an eclectic mix of data scientists, engineers, and all of them work towards solving a mission rather than solving a particular problem. Therefore it no longer says I’m a data engineer, I won’t give you data or a I’m a data scientist I don’t do data engineering—none of that stuff happens.

Everybody is expected to do all sides of the spectrum. Everyone is expected to get the oil from the ground and sell it in the market and get more money for that oil they’re trying to sell. That’s the expectation we’ve set up. Because we have set it up this way all the team members have X and a Y we like to call it—a X axis is the data analytics and insights and the Y is their exclusive home where they are working. It could be marketing, operations, finance.

They always have two homes and they’re always welcome to live in this home or the other one but they just solve one problem at scale.

Darren:

I’m going to be a little bit De Bono’s black hat here; with all of these concerns around data security. When you have opened up the supply chain across the whole organisation is there one point of responsibility for the security of the individual’s data or is that something you have to build across the whole supply chain?

Kshira:

We’ve made data governance the bedrock on which we actually created these teams. Govern comes from the Greek kybernan, a word which means to pilot people in. In big organisations like banks governance is all about filling out forms saying I want access to this data—that’s not what we mean.

The bedrock of data governance is who has access to it, are they looking at the right data, the integrity of the data, and is it accessible in a way that they can rationally be confident that it is the right thing.

We have a horizontal team that supports these vertical teams and that horizontal team is tasked with the job of making sure we exactly know who is accessing what at what level of access and everything is controlled from the back.

Darren:

By the platform.

Kshira:

By the platform. So if you’re a level 5 data accessor you can see everything, level 3 you can see some things but you can’t see everything. So it’s like if you’re a level 5 general you have access to the nuclear codes.

Darren:

The football.

Kshira:

You have access to the football and can just do drop table.

Darren:

And that probably aligns to the level people need any way to do the job. If you give people all the data it suddenly becomes overwhelming. Each of those levels would be consolidating, refining or packaging it so that’s it’s easier to process it for the task at hand. It’s not a control thing beyond protecting the integrity of the data base in the first place.

Kshira:

Protecting them from themselves. If I give access to everyone on every system they don’t know whether to look at this avenue or this one or this one. If they have only one avenue they are sane enough to know that there is only one.

Darren:

Opening up everything is almost counter intuitive to what you’re trying to do, which is to make the raw resource as valuable as possible and giving them as much raw resource as they want is not going to make that process simple.

Do you think that the language the industry (not just marketing but all business) uses around data is also incredibly confusing? There’s big data, 1st and 2nd and 3rd party data, native data. These are all terms that just befuddle most people.

Kshira:

That’s true. People try to create a sense of artificial hype around it, try to make it something more complicated or mystical than what it is. We keep trying to break that mystical barrier. We’ve gone to meetings where someone has wanted to know this thing about a customer. You could take 6 months to come back and tell us but we answered it in 15 minutes—it completely breaks the myth about how complicated it is.

You could see the instant change in how they think about data and then they start asking questions that we have no way to answer but at least now they’re asking. We tell people ask us questions we can’t answer and don’t worry about the data. That’s not your worry. Just ask complicated questions and it makes our jobs interesting too.

Then we get to work with interesting problems rather than the same daily thing again and again.

Darren:

Having the challenge is what drives progress. If you’re just doing the same thing over and over again it just becomes a bit prosaic, a bit boring. I was doing a podcast and having a chat with Martin Cass who’s been running a media agency in New York for the last four years, which is all data-informed.

They’ve built series of algorithms and he said that one of the things that amazed him is that the traditional media process–media agencies talk about being data informed but are only working with 4 or 5 sources of data—research, proprietary research—the teams he’s working with they’ve got 100s of sources of data. And the more data they get the more accurate they’re able to predict behaviours.

Knowing the theory of large numbers do you think that there is actually a point where there is too much data? And what’s the role in working out what you need and what you should be using?

Kshira:

That’s a really interesting question. Typically there are two kinds of problems to solve. One problem is I have no idea what is happening and can you give me an answer? And the other one is I know exactly what is happening—can you make this much more optimal?

I think there are a lot of problems in this day and age where I have no idea what’s happening to the marketing dollar, to my customer and you don’t need 200 data sources to solve that problem. You know the rule of 30 where you just need 30 touch points to answer a question? Apparently when you talk to a live person you don’t even need 30 data points. Someone has proven you just need one.

You don’t need to do an overkill of that in a greenfield space but if you’re in a space like high frequency buying in ad exchanges or stock markets where you already know that everything is at optimal level and you’re trying to optimise that 93.4 into 93.7 maybe you need a few more features and data points. But again that’s not really going to help.

The Netflix price is a really good example. Their algorithms were recommending a 90% or something and they wanted an algorithm that predicted much better so they created their own algorithm and it increased the price by 3 or 4% but it took them almost a day and a half to get that recommendation.

At what cost do you want to get an accurate recommendation? Do you want to wait for 2 years to make a recommendation on a customer who is so whimsical and might move on? Or do you want something that’s smart and quick and can get an answer out? That’s how we make the decision on more features.

Darren:

It comes down to probability and outcome. All of these models are really just predicting probability and they’re trying to get more accurate in their prediction. It still amazes me when you talk to people about the coin tossing exercise where I’ve got 3 heads in a row—what’s the next one going to be? How people don’t actually understand that the probability of it being another head is always 50/50 every toss.

They believe that the previous outcomes have an impact on the next one. It really worries me that there isn’t even a base knowledge on very simple mathematical principles like probability and outcome and statistics.

Kshira:

That’s true.

Darren:

They are very basic principles. They are principles that are alive in everyday life. We’re not even talking about high-end business or the stock market—just in everyday life, they’re quite useful mathematical principles.

Kshira:

It completely boils down to the education system and how maths is taught and how people understand it. The physical value of learning something is quite different to learning it on a graph. I’ll give you an example. The log normal distribution is quite a famous solution but people are bored with it, they don’t want to memorise it but long number distribution happens to everyone.

The growth of your hair, your nails—everything is a long number distribution. And when you teach people how long your nail is going to grow as a long number distribution nobody is going to forget it.

When people are standing in a queue it’s a simple poison process you stand in queues all the time so if you understand the poison process you’re going to pick the best queue to stand in. If maths is taught that way people don’t forget it.

Darren:

You think it’s the lack of practical application in the way that it’s taught is why people reject it?

Kshira:

If you learn the poison process and it’s just some weird property density function nobody understands it. Teach poison process by standing in a queue in the classroom and figure out how people come into the counter you’ve learnt something valuable and something about the world you can use in your everyday life.

Darren:

It’s a good point. There is a move to encourage students to follow STEM. You’ve got quite a sizable data team—it is difficult to recruit isn’t it?

Kshira:

That’s true. We have this policy of trying to go back to Uni, as fresh as we can because knowledge is not something we’re really after. We’re not really about do you know this technology because something you know today in two years will become irrelevant in the data analytics space. It’s more about are you willing to learn and more importantly unlearn what you’ve learned previously?

That happens a lot. You learn something and you’ve done it for 5 or 6 years and you think that’s the only way to do it and you come into an area with that idea and that’s not fair to either the area or yourself. That is the only thing we look for.

When we start looking with a different lens they’re more willing to unlearn and reinvent themselves and they’re not worried about what they know. That’s the kind of people we look for.

Darren:

It’s something I’ve heard before. I have a colleague in South East Asia—he trolls through the universities, especially China and India for those graduates or undergrads that are coming through because he’s said that as soon as someone has had even a year in the industry they’ve been taught or exposed to all of these flaws in the application.

He wants people as pure as possible. And the other thing he looks for is their ability to explain complex in very simple terms. It goes to your translation again—the ability to be able to do the analysis but then bring it down to simpler so that other people can process it.

Kshira:

Apparently the motto of the city of Naples is avea povea—which literally translates to making the complicated look simple. It’s what we try to do here. It is demystifying data, making it look as if it’s nothing. Instead of saying it’s very difficult, you need some special glasses that’s not how we treat it.

Darren:

I’ve just noticed the time. Thank you so much, Kshira, for sitting down and having this conversation. I’m really looking forward to seeing your presentation at AdTech in March but before we go do you think it’s possible to build me an algorithm to give me the tax lotto numbers for next Saturday night?

 

Are you struggling with the complexity that digital and data offer to business? Let TrinityP3 make sense of the new digital ecosystem for you