Myths and Realities of Data and Machine Learning in Marketing

Machine generated transcript…

there’s no doubt of course that we can automate many and many of the tasks that underlie the marketing advertising business today and that could be very valuable but there’s always going to be room for the human perspective Im gonna talk to you today my topic is machine learning and data and Im not really gonna get into the weeds of what machine learning is beyond just saying this its a means by which computers can effectively solve problems or derive insights without being explicitly.

Programmed or having rules to do so, that’s all Im gon na say about, it, instead Im gon na turn to talking about the myths, and realities of how data and machine learning apply to business. From specifically my vantage point within advertising industry, which is on the technology side, so a little bit about me, you heard a little bit previously, but I started off my professional career as an academic received. My PhD in mathematics specifically said her studying geometry and general relativity and black holes that’s the short shield black hole which

I was very familiar with in the olden days I did a couple of postdocs first at Stanford and the second one here at Columbia just a couple buildings over and partway through MIT realized that I was miserable and I as much as I loved the math and I loved the teaching I felt like I was disconnected from the world and not really able to use my whole self as I said Im not really going to get into the weeds of what machine learning.

is rather Im going to present to you a sort of a series of metaphors about the content but before I do that let me first ask you all a question how many of you have seen saw the movie Black Panther raise your hand Oh awesome and those who hadnt seen it you’re at least familiar with it raise a hand okay so I want to start by talking about a vibranium now you may recall that vibranium is this fictional metal that.

sits in a deposit and solely within the borders of the hidden nation of Wakanda that black panther comes from and it has these uncanny third thermodynamic and kinetic energy properties that he uses in his suit and it has boots to give rise all kinds of superhuman actions that hes able to take and it causes the local flora and fauna to mutate giving rise to an herb that when ingested gives one speed and endurance and in agility and extrasensory perception and healing powers and the ability to see in.

the dark and to hear heartbeats and on and on and on so I dont in any way mean to diminish the really cool and intricate lore underlying the storytelling of how exactly vibranium gives rise to the power of the wakandan people but you gotta at the end of the day you gotta admit its a comic book universe and vibranium is sort of this all-purpose source of power right which is kind of akin to how we here machine learning and AI being used in the.

industry its this all-purpose source of power that’s gonna solve all your problems now we know that the reality is a bit different than that and Im going to my goal for today is to try to separate the fact from the fiction around machine learning and AI in the marketing industry so if you take one thing away from my talk today let it be this machine learning is not vibranium okay well the specifics here that I want to dive into are focused on three.

specific areas this as data and machine learning apply to strategy of your business the organization of your business and people so lets start myth number one creating a data and machine learning strategy is a one-and-done you gather together some experts you do your research you sit in a room you create your strategy and there you have it your data and machine learning strategy once and for all well the reality is much different data and machine learning is not a one-and-done its a new ingredient for.

strategy in every part of the business the metaphor I want to give you for this is think of it like our favorite thirteenth element aluminum now aluminum has thousands of uses today in automotive industry and aircraft household and building you name it aluminum shows up its an amazing gradient by the way my metaphor extends a little bit here because when it was first discovered aluminum was actually.

thought to be very rare and the French crown jewels actually feature bars of aluminum because it was thought to be this newly discovered rarity but not long after they discovered that actually its one of the most common metals on the face of the planet and its now everywhere so maybe some day machine learning wont be like that to the severity that we all hold up but everywhere but the point here is aluminum is not something that you sit down and think to yourself what is my.

aluminum strategy know you have your business strategy and you look to see how aluminum can help solve the problems that each part of your business whether that’s to make a better product or make it more efficient or safer or cheaper so I submit to you then that machine learning and data science and data are like the aluminum for marketing and other businesses as well let me turn to the marketing our marketplace.

advertising marketplace rather and talk about how it fits in there now I say marketplace but the advertising marketplace isnt just one marketplace there’s a whole spectrum of these ranging from television up fronts linear television on the one in two hand sold digital advertising in the middle all the way out to the RTB or real-time bidding marketplace on the far end for those who may not be familiar real-time bidding is that means by which advertising is bought and sold via.

instantaneous programmatic auction every time a browser loads or an app comes up a call a sent an auction is held and the winning bid the winning advertiser gets the privilege of serving a creative in real-time in milliseconds and that’s the business that AppNexus was in so much of the work in my experience has been deploying machine learning into that real-time bidding marketplace let me.

give you a few examples first of all using our proprietary technology for our real-time bidding marketplace we deploy machine learning algorithms to help determine that bid price that has to be calculated in milliseconds on behalf of an advertiser to decide on the basis of what website its coming from what user it is what time of day what the advertisers goals are how much to bid which then in turn determines whether or not theyre likely.

to win great use of machine learning on the sell side on this of our marketplace supporting publishers with their monetization of their content we have machine learning algorithms that help to set reserve prices or floors per action to help protect their bottom line their monetization or to help determine whether to allocate a given impression to a guaranteed deal that they have already sold versus something coming from the spot market RTB market helping.

publishers monetize we have algorithms that support the marketplace in itself to create more efficient more liquid marketplaces so just as an example here providing the means by which buyers whether theyre using our proprietary technology or not can buy not just impressions but actually video completes or views we arbitrage on the back end so that they only pay if they get the outcome that theyre looking for and.

finally my team focused a lot on marketplace safety so that little four is intentionally placed between the supply and the consumers because were doing both were cleaning the the traffic that comes in to the platform ensuring that its human and not bot traffic and also making sure that the supply meets the criteria for our marketplace for instance not piracy or hate speech or pornography so these are all different ways that weve deployed machine learning in the RTB marketplace but when AT&T acquired app Nexus I think.

of it as me acquiring a new arm to the data science team and that team is looking at interesting ways of using the AT&T data to enrich this marketplace the future xander marketplace even further so how are they doing that well theyre taking the pterri AT&T identity data in an anomaly in an anonymized privacy safe way and looking to see how we can push that into the marketplace to allow advertisers to connect users across screens their different viewing experiences were using our proprietary media assets and.

understanding how consumers engage with media how that ties to the metadata around the content how they then engage with brands whom they see through that content and get deriving insights from that feeding back into the marketplace and finally perhaps most apropos for this audience looking to see how we can understand that consumer data to provide insight about the audiences that marketers are hoping to reach and where.

they are in their customer journey and eventually tying that all back to attribution around their brands so I hope youve seen how that there’s not one overarching data and machine learning strategy for zander there are many places in which it ties in all in service of the broader vision of the converged xander marketplace and while you all may not be running marketplace.

companies I submit that you probably have many different parts of your strategies as well and should think of machine learning and data as a raw ingredient to be used in those areas as well so next myth machine learning is best left to the experts in a data science organization now at the face of it this has to be true you probably want to have your machine learning experts actually designing and deploying your machine learning algorithms fair enough.

but the reality is slightly more complicated than that machine learning and data science must directly interface and engage with the business in order to derive to drive results you do not want them sitting off in a little silo to the side so in order to explain this further Ill give you another metaphor for machine learning this time electricity this metaphor was made famous by Andrew eing a few years ago who said.

that he believed that machine learning and artificial intelligence would be to the 21st century what electricity was to the 20th meaning that it would one-by-one transform every industry I actually think that’s true and I think that the metaphor is powerful because when you think about what it means to have electricity transform an industry that gives you some sense of what it means to have data and machine learning transform an industry so for instance suppose you’re running some sort of a manufacturing operation and you want to electrify your factory you dont hire a.

bunch of electricity experts and put them over in some corner no theyre gonna come look at all the different parts of your business from your tooling to your platform to your infrastructure and address those one by one according to the business need and by the way once youve meant once youve electrified her your factory you’re gonna have to make sure that you keep an eye on your supply chain making sure that you have a full supply of electricity coming in at all.

times and its reliable so similarly as we bring machine learning and data into our organizations we have to think one-by-one about what that means in terms of platform and tooling and priorities and by the way making sure that we look after our data supply chain let me give you some sense of what that’s looked like app Nexus this slide was is adapted from a talk I gave at the very end of 2014 when we were just on the cusp of the modern age which is why.

it it ends at 2015 just when Id been made chief data scientist and I was talking about the sea change that Id already seen then in 2015 just a tap Nexus now Dark Ages is back before I started when the company was founded but in the years prior to when I started in 2012 the company was producing a large amount of data so I would say was plentiful but it was essentially a byproduct it was all reporting data on.

behalf of the fundamental service being offered by the platform nobody was really using that data very effectively or when they were they had to use incredibly sophisticated mathematics on top of the data because it was so rigid that in order to drive any insights you had to be very very subtle now as the data science team was created we we then found a new home.

in engineering and started being able to customize some of the data we started being able to pick some of the variables we wanted to work with and create our new aggregations and get better tooling and start actually using some machine learning techniques so this was the beginning of the electrification if you will and then around the time that I was made chief data scientists we turned data science into its own organization and built out that started building out.

in 2015 and that it took took place over the course of a few years a true platform by which the team could go and access all the raw data that they wanted to extract insights and drive machine learning algorithms to create the value for the market place that you saw in a previous slide now you may say well you just started this section by saying that you didnt want to have your own machine learning experts sitting off in some.

organization that they needed to be integrated into the business but here I am saying that data science became his own organization and that’s when we started achieving some success but the point here is that from the time that we came became our own organization we started having to make an explicit intentional decision to integrate with the business rather than already being integrated into a product or engineering team we started having to operate in a trifecta manner this slide is adapted from one that I presented to the company.

right when we we became an organization of top-level organization and I was saying where does data science fit in right here we have to operate in lockstep with product and engineering organizations to make sure that were solving the right problems and then appropriately executing them for the business now again Im speaking from an advertising technology perspective and I know many in the room are not but this.

generalizes I think pretty directly to probably your own businesses as well when you have machine learning in the mix you need to ensure that it is operating in lockstep with the people who know what the business needs are what part of the factory needs to be electrified where you’re going to derive that benefit and in lockstep with whomever is going to executing that vision whether its a.

services organization or a sales organization or whatever the insights are that you’re gonna plug back into your business you need to make sure that theyre operating always in lockstep not a siloed organization okay and finally myth number three this one is very common its almost in the water it feels like machines will replace human intervention in marketing there’s a lot of fear underlying this myth but I think the reality is machines can do a lot but they cannot do everything in marketing which seems yeah.

let me go in thinking about what metaphor to use for machine learning and data for this myth I thought long and hard and realized that its the simplest thing possible at the end of the day machine learning his machines so the image on the slide here shows go you may be aware that deepmind recently beat the worlds best go player at go which was thought to be sort of an insurmountable hurdle in artificial intelligence some time ago and I have to confess that I.

was a little bit sad when that happened but at the end of the day even as smart as that particular machine was it was a machine it was solving a problem that was confined to a specific task with specific parameters in a specific end goal and that is what machine learning does today so there’s no doubt of course that we can automate many and many of the tasks that underlie the marketing and advertising business today and that could be very valuable but there’s always going to be room for the human.

perspective humans to understand what patterns are emerging from the data look at all of that understand the humans behind that data behind those patterns and bring that to novels storytelling formats originality creativity because that’s how our culture evolves is through this creative evolution that’s not confined a specific task after a specific task at the end of the day humans have something that machines will never have at least not in any kind of future eyes Percy other than a.

science fiction one and that is empathy and that makes all the difference thank you.


Comments

Leave a Reply

Your email address will not be published. Required fields are marked *