Life Beyond Cookies: Increase Ad Personalization with Cannabis Consumer Identity Graphs

{}Introduction

Host: All right, good afternoon everyone, welcome to the MediaJel podcast, where we cover the latest marketing trends and strategies that are most effective in growing your cannabis dispensary, delivery service, or brand. MediaJel connects brands and retailers with cannabis consumers through our ad network of mainstream publishers, mobile apps, games, and TV, we help companies advertise through paid search, SEO, and programmatic display advertising, to drive e-commerce sales. I'm your host, and today we'll be discussing life beyond cookies, how to increase ad personalization with cannabis consumer identity graphs, and I'm super excited to introduce Dana Silva, who is the CTO here at MediaJel, bringing a breadth of experience, over 15 years of engineering experience, to today's podcast. Welcome to the show, Dana.

Dana Silva: Thanks for having me on the show, I'm glad to be here.

Host: Yeah, happy to have you.

Why Are Apple and Google Moving Away From Cookies?

Host: Well, let's kick it off with the big tech companies — why aren't Apple and Google going away with cookies?

Dana Silva: Well, there's a lot of reasons for that, and some of it is choice, but it's the way of the world, people don't want their privacy to be invaded, and most of the reasons have to do with that — deterministic identifiers can cause people to violate privacy practices, you can build really rich profiles, and that's not allowed. Also, the big companies are in a position where they have so much first-party data, that doing away with cookie-based tracking will affect them not nearly as much as small companies like ourselves, or other companies out there marketing, because they have so much first-party data under the hood, they don't necessarily need the cookie as much as somebody like us, who operates on the open web, without as much PII data under the hood. I probably can't speak to all the reasons the different companies are doing away with cookies, but I think mostly it has to do with privacy concerns, being compliant, and all the things we need to do as good marketing companies, to be ethical.

Host: Yeah, exactly, and with all the things that occurred with the last election, and all the things that went on with Facebook, there was definitely a crackdown after that, on how people can use data — Facebook took a multi-billion dollar hit from Apple limiting their ability to track across devices and apps, so, yeah, data and privacy is top of mind for everyone. And, you know, there's alternatives, and we're going to go a little deeper into that — what are the limitations as far as a cookie-based tracking platform?

The Limitations of Cookie-Based Tracking

Dana Silva: And that's something a lot of people don't understand in general, cookies are not the end-all-be-all of tracking, and haven't been for a long time. One of the things I could say is, in the beginning, cookies are very old — the idea of cookies was invented in the mid-90s, in that really "web one" sort of landscape, and, at that time, the use case was, a person used a computer, probably on a desktop, used the same browser for almost everything, that was also during the time of the browser wars, and things like that. In that ecosystem, the cookie was amazing, because it's a deterministic identifier given to you and your browser, and then you can be tracked, third-party, as you move across the internet, with these third-party cookies. Well, even without it going away, or without the compliance issues, there's a lot wrong with that in today's landscape — nobody really uses the internet that way anymore, people have many devices, whether it's a bunch of phones, connected TVs, or whatever. A cookie is kind of like saying "browser," in many ways — you've made an assumption they're going to use the same browser to see an impression and make a purchase, and that's not necessarily how people do things anymore, people probably see ads in apps, where there's no browser — if you're using Candy Crush on your phone and see an advertisement, there's no browser and no cookie to even be accepted. So, in that situation, cookies already don't work, they don't get you anywhere — it's still effective for web-to-web, when they're not blocked, for that specific use case, but any time cross-device happens, anything in-app happens, and we're also seeing a world where people are starting to use different clients for different situations, not everybody uses the same browser for everything, there are other browsers coming online, like Brave, that also block things like that. So, yes, they're going away, yes it will have impact, but there are already many situations and use cases that have arisen in the last decade, 15 years, where they already weren't adequate to track everything you'd want to track, to show attribution, show value with advertising, and all the challenges marketers face.

Host: Exactly, and it's really this thing that changed — you know, phones are an extension of ourselves at this point, we keep it in our pocket all the time, we have our location services on all the time, we're actively searching on platforms, playing games, always on the move, ordering food delivered to our place, going out finding restaurants on Google — it's endless, and it's always evolving, it's really like the emergence of smartphones, and then social media, that completely changed the landscape, and especially Google too, these are some of the most valuable companies in the world because data is king, data is more valuable than oil — I mean, maybe not right now, because oil prices are so high, but yeah, it's pretty hard to fill up the tank these days.

Dana Silva: Exactly, exactly.

How MediaJel Solves Attribution Without Cookies

Host: So how do we move forward without cookies, and how do we overcome the limitations we've discussed previously?

Dana Silva: Well, that's a loaded question, and there are many different techniques to solve that — what I can do is speak to how we do it at MediaJel, specifically some of the ways we tackle that problem. From the beginning, for display advertising — not necessarily for paid search and SEO, but for our display advertising product — we were always faced with the challenge I just mentioned, how do you make the connection between app traffic and purchase behavior. Just to elaborate, when we target campaigns as MediaJel, we target a lot of mobile users, and a lot of the ways we build audiences enables us to target by device ID with our DSP partners, and because of that we have all these impressions showing up in apps, but there's not a deterministic way to get from the app to the cart, traditionally, or at this stage anyway — there are very few cannabis companies that will allow you to complete the purchase in-app, a few are coming online, but in general things are still done over the web for carts, like Dutchie and Jane, and all the big ones out there, they're web apps, so you're receiving web traffic. So we collect data on both ends of these experiences, but how do we tie the data together — we use probabilistic techniques, similar to the way many identity graph companies work, where we'll use things that come natively from the TCP and HTTP protocol that are very difficult to block or stop, they don't have the deterministic advantages, but if you do things correctly you can get very highly probabilistic results. So we're able to use identifiers like combinations of user agent and IP address — for those unfamiliar with what a user agent is, it's really a string that contains as much information about your connection over the HTTP protocol as can be recorded, so it'll record things like the device you're using, the device type, browser information, quite a lot of rich data, and, when combined with another identifier like IP, it's quite accurate, reasonably highly probabilistic, and something we can use to make that jump from web traffic to mobile traffic. So that's one of the ways, using probabilistic techniques like that, and then, after you have some matches, and an identifier making the match, you can get to something that exists in an identity graph — and, before we get into exactly what an identity graph is, it's limited in the amount of identifiers you can use to match inside of it. We use a partner that we primarily match with device IDs and MAIDs in the graph, so we somehow need to make a jump to get to either a device ID, a MAID, or a cookie that's one of our partner cookies, because all the DSPs have a cookie sync that goes with these identity graphs. So, maybe we got a little too into the weeds there, but really the two answers to your question are, we use a variety of probabilistic techniques, with our own data and proprietary data sets, and, in addition, we use an identity graph to get better reach for attribution and targeting.

Host: Yeah, I mean, it's definitely above my pay grade, I'm not an engineer, my friend, but I know the data points that come in, and I've learned how we connect these impressions to devices, to transactions, and it's quite impressive, all the things you've really led on the innovation front for MediaJel, so kudos to you and everything you've built and managed. What are some of the advantages and disadvantages of using this probabilistic model?

Advantages and Disadvantages of Probabilistic Matching

Dana Silva: Well, the word "probabilistic" is the disadvantage — the fact is it's not a hundred percent accurate, there's always going to be some amount of false positives that come from these techniques, and those are really, I'd say, the main disadvantages. Now, we do lots of things under the hood to mitigate that and increase accuracy — one of the things we do is filter out IPs that are known hotspots, by monitoring things like frequency, the amount of signal we're getting, it's unlikely to see more than a certain amount of times, over a certain amount of time, coming off an IP address, and we discard all of that to rule out false positives as best we can — but there are situations where we just cannot eliminate them 100 percent. Maybe you're sitting at a Starbucks, we send you an ad, and the person beside you buys because they're behind the same router — probably a bad example, since a Starbucks would have been filtered out, but some sort of situation, maybe an office, where different people aren't connected in any way, but are behind closed doors right next to each other, going through the same router, leads to things like false positives. So we're always trying to improve those techniques, and we've done quite a good job — we always try to match VPI with something else, lots of times we'll pull the user agent apart, pull out the device information so it's not client specific, so we can match device to IP, and if we can match the entire user agent, it's similar to the cookie match, we've matched the client, browser, device, and IP. But the fact is it'll never be completely free of false positives, so those are the disadvantages, in my opinion — the advantage is, using these probabilistic techniques, and I don't want to say they're for sure here to stay, but most of the techniques being used right now, to either block cookies or increase privacy protection, they're all done through software, higher up on the application level, than things that happen right on the TCP/IP or HTTP protocol layer, which is how the internet works — so to get rid of all those identifiers would be very challenging, if not impossible. I don't like to say words like that, of course anything's possible with software, and we'll see more movement in the future towards blocking some of those identifiers, but, in the meantime, it's pretty solid, and probably something you can stand on for a long time — it doesn't mean we aren't still pursuing more deterministic solutions and results, but we'll probably use this to some capacity, as long as it's working for us.

Defining Key Terms: IDFA, MAID, DSP

Host: Let's thank you for the clarification on that, and, before we continue, can you define some of these keywords you've used, for our audience — IDFAs, MAIDs, DSPs, what do these mean?

Dana Silva: Sure, we'll start with IDFAs and MAIDs, they're basically the same word — it's the hardware identifier ID that identifies a phone or mobile device. I mean, they're not exactly the same word, IDFA is Apple, MAID is Google, and they've also taken some other steps to make other hardware-style identifiers that aren't actually the hardware ID, it's more like the MAID is an example of an advertiser ID that masks the actual ID — but essentially we're talking about hardware IDs, so, for the purpose of this conversation, we can interchange them and consider them the same thing, the ID of the device itself. Then the other one you asked about was DSP, that's a demand-side platform, a really common term in the advertising space, it's a platform that has a bidder that can serve ads on the programmatic exchange — we're partnered with many DSP partners, some we like better than others, but we use our partners' technology to actually execute the purchasing and bidding of all the apps we serve programmatically.

Host: I think you had some other ones too, but those are the ones, is there any other terms you'd like to define for our audience, so they're up to speed on all the latest marketing lingo as it relates to programmatic and data and identity graphs?

First-Party vs. Third-Party Cookies

Dana Silva: I'm not sure if there's anything else, but one thing I would like to be clear about is what a cookie is — it's a really popular topic, we all hear about these cookies, but I want to be clear about one thing, a cookie is called a UUID, but it's really just a unique ID, and the cookie we're talking about specifically is a third-party cookie. There's first-party cookies and third-party cookies — maybe there's some action going on, but, in general, the big news is not about doing away with first-party cookies, it's not about doing away with data cookies, things with identity partners — it really has to do with the third-party cookie. The big difference is, you can have software that sits on-site, like we have a tracker that sits on our clients' websites, and that gives a first-party cookie, and that's probably not going away soon — but then our collector itself, that sits on the cloud, that's where we'd give a third-party cookie, and that's what's going to be impacted.

Host: Great to know, great to know. Yeah, I mean, first-party data is something — we're here at the Cannabis Marketing Summit in Denver, and it was definitely a hot topic of discussion, how to collect, curate, and activate that data on the first-party side, whether it's point of sale, e-commerce, CRM, there's always different places you can gather data, and you can activate it through a lot of these platforms, including programmatic advertising, email, SMS messages, and retargeting. So there's a lot of ways you can leverage this data, and there's also third-party data sources, like New Frontier Data, or Headset, or some of these other platforms — it was definitely a big topic of discussion, all marketers talking about it, and privacy as well, and how to reach these different audiences, it's definitely a major priority for marketers, especially when we're trying to maximize our marketing budget, especially with the looming recession and everything going on right now. Thank you for clarifying all that — well, let's dive deep into the identity graph, what is an identity graph, Dana?

What Is an Identity Graph?

Dana Silva: Sure, and, once again, there's a few different versions of these things out there, but what we use is something called a snowflake graph, which is engineering terminology, but it really means it's a graph that has spokes — the way the data is structured is, if you visualize it, like snowflake-looking spokes, so it's the idea that one ID will connect to many different IDs. The idea is cross-device — the goal is to take somebody like yourself, and, if we know the device ID of your phone, we can use that to know the device ID of, say, your other phone, maybe your tablet, maybe we'll know about your wife's or child's phone, through a household match — but any devices we can associate, and, in general, device identity graphs have two ways of clustering data, by individual or by household, and they both have their use cases. That's the idea, the idea is to have a graph where, if you have one ID that's a match on a user, you can get the other IDs associated with them individually, as well as with their household.

Host: Great, and, greater than that, what are the pros and cons of building this identity graph database that you're talking about?

Pros and Cons of Building an Identity Graph

Dana Silva: Well, the cons all have to do with software and engineering complexity — these identity graphs, MediaJel has kind of two approaches to it, we have our own data that we've graphed to achieve as much results as we can, however we're not the largest company in the world, we're not quite as big as Google today, so it's limited, however it is useful, and we're able to do some interesting things with it. But then we have a partner that we purchase an identity graph from, and that's where some of the complexity lies — the identity graph we purchase from our partner is a few billion records, they generate it every six weeks, to build it fresh, that's part of doing probabilistic results, that over time IPs change, different probabilistic identifiers fall apart, so it's important to keep these data sets fresh. So, unfortunately, as an engineering problem, there's no way to incrementally, efficiently load this data into your data warehouse, we can't load it all up until today, then get tomorrow's data — so it's always this massive processing effort that happens every week when we get the new graph, to put it into our data warehouse. We built the job, it runs quite smoothly today, but there were many hours logged to get it to that point — it costs money to process data, and then, after you get it into a data warehouse, there's yet more code that needs to be written to analyze it. So I'd say most of the cons have to do with normal big-data problems, when you're dealing with really large data sets, there's overhead, there's talent that needs to work for you to do the work. And then, of course, you don't necessarily have to house it yourself, or do your own thing like we do, you could use a partner's graph out of the box, and some partners have APIs and things like that associated with it, but what you miss out on is the ability to augment or add your own data to the graph — and that's part of the reason we ingest our partner's graph, bring it into our data warehouse, and augment it against our own data and our own identity graph. And, in addition, what we've been able to achieve is, just like the demand-side platforms maintain a sync with these identity companies, so do we with the MediaJel data, and we maintain two things — we of course use our third-party cookie to make a sync with the data, it's not gone yet, it does still work to some capacity, however it becomes less effective all the time, and its match rate does drop as cookies continue to be blocked. But we've also tried syncing, in real time, some other probabilistic identifiers that we'll probably keep experimenting with, to have the best sync with our identity partner possible — right now we sync something we call the MJ hash, which is a hashed identifier, but, under the hood, it's the user agent and IP address, like we spoke of earlier, for probabilistic techniques — we'll probably be introducing a device-plus-IP version of that identifier in the near future, but likely we'll stick with the idea of syncing in real time with the identity partner, keeping our data in check, and although there's great advantages to that, there's once again overhead that goes with an effort like that. So I'd say all the pros are all the things you'd expect, the ability to increase scale on audiences, improve attribution, get better results, have deeper insight and analytics into the customer journey, and I'd say all the cons have to do with the engineering overhead it takes to achieve such results.

Host: Yeah, we need a top-tier engineer like yourself to manage something like this and create your own identity graph, it could be quite an undertaking, and an investment in a lot of time.

Dana Silva: Well, we've got a great team, that's the other thing, it would be difficult to maintain this technology without that.

Identity Stitching Explained

Host: And, as far as a component of building identity graphs, what is identity stitching?

Dana Silva: Identity stitching is really just what we were talking about before, bringing together different data sets, in a way that's in line with how your identity graph works, to connect the different spokes — like when I was talking about how we connect our data to the identity graph, that would be an example of identity stitching. We have other partners we're talking to about bringing in yet more data to augment the data further, we can bring in e-commerce data, but, at the end of the day, to target and show value with your advertising, you need that identity component on all of your data, to show who we think it was, or which device we think it was, that made the purchase, repeated the purchase, bought gummies from a cart as opposed to something else, and all those things — without identity stitching, there isn't really anything actionable that can come out of that analysis.

Individual vs. Household Device Clusters

Host: What about device clusters?

Dana Silva: And we touched on this before as well, but the idea of device clusters is, there's really two ways, household clustering and individual clustering, and, for MediaJel, because of the space we operate in, we're able to use different techniques for different things — for example, the household clustering, we can't use for targeting, because we market products that are age-restricted, and restricted for other reasons, and we could never take the chance on accidentally sending an ad to somebody in a household who was, say, under 18, or somebody's child. So we're not able to use the household cluster for targeting, for that spoke of the identity graph — however, we can use it for attribution, if we see somebody purchase something, the ID verification's already been handled with the credit card and the site, everything's compliant, somebody of age did make the purchase, and then, if we use the household spoke, and can't get a match on the individual, we can find something in the graph — maybe it's still you, maybe it's your wife or husband or somebody else who made the purchase, but it's still a reasonably good match, we matched somebody in the household, and it still could be you. But that's one of the ways we're limited by the household spoke — the individual spoke, we can use that to enhance audiences, expand reach, scale our targeting, more than we're already doing through other techniques, and, of course, we can use that for attribution as well. So those are the two ways people cluster IDs.

How MediaJel Approaches Attribution Windows

Host: And, touching on attribution a few times, can you dive in a little deeper on that, and what retailers and brands are expecting on the advertising side, as far as attribution and KPIs, the numbers people are actually expecting?

Dana Silva: I know it does vary, and, truthfully, somebody from our ops team would be a much better person to answer questions like that through the day-to-day, but I could elaborate on the attribution technique we use, and some of the things we do. At MediaJel we've taken the approach that we want to make our attribution as transparent as possible — so what we do is open up a reasonably large window, 30 days of attribution, with the idea that, if we've sent you an impression and can correlate it with a purchase, we aren't saying we 100% made the sale, we're saying we assisted in the sales funnel — at some point you saw the ad, and it was because of our advertisement. We all know there are many use cases where you'll see a bunch of ads while you're too busy doing something, on an app, playing games, you're probably not going to click right now, but if you remember the name, you might search Google later and go ahead and make the purchase. So we use the techniques we talked about before, to make the match, and we try to be as transparent about that in our dashboard as possible — when you look at the transactions, you can actually hover over some of the identifiers, see what techniques were used, and we give the advertiser the ability to change and manipulate their own attribution window — maybe you don't believe in 30 days, maybe you only believe we assisted with a purchase if they saw an impression and bought within five days, and we put that control into the advertiser's hands, they've got a little toggle, they can adjust it, change the attribution model to the one that makes sense for their business. And, as you certainly know, in the space of marketing, depending on what you're selling, the sales funnel is so different — if you're trying to sell somebody a car or a home, you can advertise to them for years, and it probably all had meaning, especially if they wind up doing business with you on other things they purchase — different windows of time that people believe in. So that's what we've decided to do, just give you all the information we can, the deepest insight into the user journey possible, let you make your own decisions, and I think our customer retention rate has proven that people are very happy with the results they're experiencing through their ads. I think most of our clients find the transparent approach to attribution refreshing — ad tech has got a really long history of bad actors, from click fraud to ad fraud to this and that, and that's, I'd say, the biggest way we're trying to be different with our technology, to just give you the full picture.

Host: Yeah, and then you can decide.

Dana Silva: Exactly, exactly.

Why One Attribution Model Doesn't Fit All

Host: Transparency is key, and, because some of the basic marketing 101 methodologies, like the marketing rule of seven, which is really old, means someone needs to see your ad or brand seven times before they'll consider making a purchase — it's something well known in the marketing space, and it's probably far higher than seven now, since we're in this attention economy, all these different companies competing over your time. So it's really, when I think of, let's say, programmatic display advertising, it's really like a mobile billboard — you're driving, you see this brand, you see that other brand, but you can't just get on your phone and look them up right there, maybe you can ask Siri if it works that well, but, in general, you're not going to stop what you're doing and take action right away. So we've seen that, when you're running a display campaign, we'll see an uptick in brand searches, in organic search through Google — "hey, we saw MediaJel on this, I was playing Sudoku, or looking at memes online, saw MediaJel, and then a couple days later I did a search on Google and found them" — it's really supporting that sale, supporting that interest, and staying top of mind. And this is something we really emphasize with our partners — even with existing clients, even though you've won that client and they've made a purchase with you once, if you're in, say, San Francisco or Boston, there's 30 other dispensaries or brands around, and you're always going to be competing to keep that business, so remarketing, and following that user around, ensuring you're top of mind, is a huge priority — it's one of the key components, you have a marketing funnel that can be used for customer acquisition, but it also can be used for customer retention, so you really need to balance that with what those goals are for your business.

Dana Silva: And that's one of the things you touched on, that I think sometimes marketers miss, is that, depending on where you sit in the funnel, people are going to pitch you a different attribution model, and one-size-fits-all doesn't necessarily make sense — truthfully, all of the channels help, and you may not buy all the channels from us, or you may not buy all the channels from Facebook, but that's kind of where we're trying to go with our future, to give you real insight into how they're all working together, and give you the most complete picture we can possibly give — I wouldn't say that's something we have today, but that's absolutely our mission, what we're in pursuit of, the richest cross-channel insight we can give. And there's little things you can just think about and make up your own mind on — is the last-click model very successful? It's very successful, especially if you own a search engine.

Host: Yeah, exactly, which we don't.

Dana Silva: Right, and there's — in my opinion, I'm not trying to kick off a complete philosophical debate, that's part of the approach, that our customers can decide these things for themselves — but if you have some strange name in your brand, like Kiva Chocolate, or something, and nobody's ever heard of Kiva before, and all of a sudden a bunch of ads go out that we've served for our customer, and Kiva starts being typed into Google and Bing and all the search engines all day long, well, those ads helped facilitate those searches, and there's a good chance they'll get last click, because that'll happen, you saw the ad, you typed it in, and you get last click. But it doesn't mean we didn't help, it doesn't mean we weren't part of the journey. And then, of course, the converse of that is, what if you typed in "cannabis chocolate near me," and the search engine directed you to a store or brand like Kiva, well, that's a little different, I'd argue they contributed greatly to the conversion, they had much more influence than the ads in that case, because they actually guided you through the funnel, through the search engine. But there are so many situations like that that are all so different, and that's, I think, sometimes what we forget, to look at the big picture, and not just commit to one model — I don't think a one-size-fits-all model really exists, different products and situations have the journey happen quite a lot differently, depending on what it is.

Host: Yeah, "a rising tide lifts all boats," that's one of the common phrases we use, and there's definitely a lot to it. What's the platform — you can always look at Google Trends, if, for example, Jeeter, the pre-rolls, they got a lot of popularity, and ran a ton of campaigns, or Cann, those cannabis drinks, they've made some massive pushes to really get their brand out there, and they do a great job really getting this exposure — if you go on Google Trends and search for that brand name, you can see the progression and demand for searches on Google. So it's all public information, ready for you to leverage, and there's also third-party data platforms like Headset, where it shows you product sale trends by different brands and categories — there's a lot of information out there you can use to evaluate the success of a campaign and really gauge market penetration with your brand.

E-Commerce KPIs and Tracking Walk-Ins

Host: Just to touch on some of the KPIs our advertisers are looking at — the way I like to talk about cannabis companies in general, especially retailers and delivery services, is that they're e-commerce companies, upwards of 50 to 80 percent of their sales are coming from online pre-orders, and COVID really forced consumers to use e-commerce as their primary method of engaging with the retailer, because they had to, they were doing delivery or curbside pickup, everyone wanted to be safe — it really pushed this industry forward as far as e-commerce, and some advertisers I know in metropolitan areas are seeing upwards of 80 percent of their revenue come from e-commerce orders, it's incredible to see that progress, and you really have to take that into account when running this business — you need to ensure your product catalog, your pictures, your descriptions, the checkout process, everything is as streamlined as possible, because we're investing all this money to get you that interest, and get someone to visit your website, but if you don't make the checkout process easy, people aren't going to buy. So e-commerce sales, revenue, transactions, new customers, those are all important KPIs our advertisers are looking at, and also walk-ins — because we can show these ads, and not all the transactions are going to go through the e-commerce platform, so that other 20 to 50 percent, depending on the store, those are people who still want that experience when they come to a dispensary, they want to talk to the bud tenders, maybe participate in an event — so tracking those walk-ins, if they've seen ABC Dispensary keep showing up on these ads, and they decided they want to come in Thursday night and re-up on their cannabis, so we're also tracking walk-ins and data there.

Host: If you want to share a little information on geofencing, and how that's used in this attribution model, that would be helpful.

Geofencing and Deterministic Walk-In Tracking

Dana Silva: Yeah, and that's a completely different technique than what we're doing with the event-driven data we're collecting, and that is deterministic by nature — once again it has pros and cons, there's a reason we do the event-driven stuff and don't just rely on walk-ins, but it is where we started. One of the earlier attribution solutions MediaJel was able to provide customers is the idea of tracking walk-ins, as you mentioned — so how do you track walk-ins? The idea is that all our DSP partners will forward, if the user has allowed it, device IDs to us, so we know the device ID associated with the impression, and we have another partner whose job is to house geospatial data, SDK-based geospatial data, so they have a really large, rich data set of device IDs combined with the latitude and longitude they were observed at. So, to make this deterministic match, for us, it's as simple as taking the device ID, MAID, or IDFA, whatever we got from the impression, and taking a window of time from our partner that has the SDK data, looking at all the observations we're seeing, and seeing if they wound up in the polygon, or geofence, that we drew around the retail location — if we see them in there, absolutely, they visited, or they almost visited, at least came to the parking lot, or walked through the door. And this is cool, but there's once again limitations and drawbacks to it.

Maintaining Consumer Data Security

Host: I know we've talked a lot about this already, but is there anything else you'd like to add on how MediaJel uses identity graphs in our programmatic advertising?

Dana Silva: I mean, we could get into the weeds if you want, but really the two use cases are attribution and targeting, and we use it for both.

Host: Awesome, awesome. Then, we've talked a lot, this whole session's been on data and identity graphs — now, how do we look on the technology side, how do we maintain consumer security with all this data we're gathering?

Dana Silva: Well, to maintain consumer security, you have to deal with the data in a very compliant and specific way. One of the things MediaJel doesn't do is house any PII —

Host: Good question, what is PII, just to make sure, for our audience?

Dana Silva: PII is personal identification information — it's your personal information, anything that could be tied back to you personally, your name, your address, your email, your phone number, all of those things. You have to be a good steward of PII, that whole topic is all about protecting that data, not letting it get out — so we don't house it, we don't really have first-party data, we'll likely move into that in the future, but we'll do it in a compliant way, of course, but right now we don't really have that concern about PII getting leaked, or hacked, there's not really that data to get. We do have some identifiers that could be argued as PII, things like device IDs, but we would never give those IDs to customers, you'd never be allowed to pull those identifiers to use on your own — how that works is, when we find them, they go into the DSP directly to be targeted in their ecosystem, and there's not really anything personal tied together at that level, it's just an ID, and you don't know anything about the person — cookie's the same thing, it's just an ID, there's not really anything personal about it.

Dana Silva: Our future — we're moving into a world where we're trying to have better results, more insight, and the integrations we're in pursuit of are to become closer to the actual POS experience itself, which is one of the larger blind spots our company has — we can track data off the online part, we can track walk-ins, but those people who actually walk in and buy with a different POS system, we really have no insight into at all, and we know we're helping with those sorts of sales. So it's something we've been aggressively trying to solve, working with different companies to see if we can get to a place where we can have that data, and likely what we'll use is the email, which is PII, but we won't ever bring in the actual email, we will have the email hashed, in either MD5 or SHA-256 — if anybody knows a little about hashing, the key is that it's not cryptography, like a cipher where you have a key and can decode the information, it's a one-way hashing trip, so if you've never seen the real email, you can never get it back from the hash. But the customer that does have the real email, like the retail client itself, that has the PII, they're able to verify that, if we say, "we got this hashed email, we think it's a match to something," they'll be able to verify that on their end, with the PII they've gotten in an ethical way, people signed up willingly, made a purchase, accepted the agreements.

Dana Silva: And there's other compliant stuff too, probably run out of time to get into all of it, but there's a ton of stuff you have to do with data — when you're collecting data like this, you need to respect something called the "do not track" header, if somebody turned on "do not track" on their browser, we discard those logs as they come in. We have an opt-outs pipeline, so if somebody wants to opt out of our ability to advertise, we collect a certain amount of identifiers to opt out, and there are several other things we do to maintain compliance, but probably the more advanced things are coming in the future for us, because I think we do want to get closer to the POS data, we want deeper insight, and that will involve housing some amount of PII at some point, and we'll have to go through the challenges of being good stewards of that information, like we have been with the information we already keep.

Closing Remarks

Host: Well, thank you, Dana, there's a breadth of information, I learned a lot on this session, thank you for all your insights and acknowledgment into everything on the programmatic display side, and identity graphs, and data, and really everything you do — any last words you'd like to share with our audience before we log off today?

Dana Silva: I don't know if I have anything to add, thank you so much for having me, I enjoyed being on the call, and I look forward to maybe the next one.

Host: All right, sounds good. Well, thank you everyone again for joining us today, once again, MediaJel, we are a cannabis marketing platform, we leverage paid search, SEO, and display advertising to really support e-commerce sales for our clients, our retailers, our brands, and beyond — so thank you for joining us today, and you can always check us out at mediajel.com, log on there and attend some of our other webinars and podcasts, and if you'd like to set up a call, just go to the "contact us" in the top right of the website, and we'll be happy to chat. All right, have a wonderful day, cheers.

‍