The unpublished Cambridge Analytica Interview
I am a Swiss technology reporter, working for Switzerland’s leading weekly print magazine “Das Magazin”. In the summer of 2016 I started investigating a then little known digital marketing company named “Cambridge Analytica”. They had like 150 views on their Youtube promo videos. They were pretty niche. But there was this incredible guy. He looked really well. And spoke so eloquently. A gentleman named Alexander Nix, the CEO of Cambridge Analytica.
After 5 months of research I reached out to him. Here is the transcript of our first conversation. A call, catching him in his car after work. Just the two of us talking, after many failed attempts of setting up a conference call through landlines and logins. This call would confirm many things I had unearthed together with my research companion Paul-Olivier Dehaye. Questions such as whether the company had been using Facebook data.
This interview would would become fundamental for my feature story that came out Saturday Dec. 3, 2016, and went through the roof within hours. The hashtag #DieBombe (german for The Bomb) would start filling up newsfeeds. It become THAT story. Like, I turned on the radio in my kitchen that night and they were talking about it.
Within days unauthorized translations popped up in many languages. Whilst almost the entire German language media was attacking my story — titled “I have only shown that the bomb exists” (original: “Ich habe nur gezeigt, dass es die Bombe gibt“) — I secretely went to work on an official translation for VICE News Motherboard, titled “The Data That Turned the World Upside Down”. I wanted to save the story before I would get wrongly “debunked” by those strange experts that suddenly seemed to pop up from every corner. (That’s when I acquired a taste for Information Warfare, btw). So, then, in January, when my story came out in English (again featuring my dear editor and colleague Mikael Krogerus as a co-author), it immediately went berserk, again.
Now that you have all watched the Cambridge Analytica Netflix documentary, read Carole Cadwalladrs amazing stories in the Guardian, seen the Channel 4 clips of Alexander Nix, here is a collector’s item. Mr. Nix, speaking freely, long before those nasty lawsuits, trials and commissions.
What I find remarkable in retrospect is that Nix seems to argue Trumps election campaign was actually profitable. Imagine.
Important reminder: this conversation happened on November 11th, 2016. Way before the world caught on to what it would mean that CA had used Facebook data. Way before everyone talked about microtargeting.
Hannes Grassegger: 00:06 Cool. Yes.
Automated voice: 00:27 Well to international access, please enter the conference. You want to join followed by the pound sign. Hello to the phone number. There was a conference you want to join followed by the pound sign.
Hannes Grassegger: 03:07 [inaudible]
Hannes Grassegger: 03:19 okay. I’m sorry. I shouldn’t [inaudible]
Hannes Grassegger: 04:02 [inaudible] [inaudible] [inaudible] [inaudible] [inaudible] [inaudible]
Hannes Grassegger: 05:23 that’s great. Thank you so much. I very much appreciate this. I’ve tried several times. So, , how much time do we have?
Alexander Nix: 05:32 Well, I’m in the car xxxxxxxxxxxxxxxxx say you have until they start screaming.
Hannes Grassegger: 05:38 Okay. Congratulations. , on two recent major successes for your company. And , I’m preparing a report that for Swiss Das Magazin. And so just to start out right away, what exactly, , makes Cambridge different from other microtargeting companies in your words?
Alexander Nix: 06:03 Well, micro-targeting is, is, , the use of largely demographic, , and geographic data to identify segments of the, , audience of interest in order to, , set them, , messages and communications. What we’re doing is, is predictive analytics using big data. So, rather than just using demographic and geographic, , data points, we’re synthesizing, synthesizing those with, , consumer and lifestyle and social media and transactional data, , in total, somewhere close to four or 5,000 data points on some 230 million adults in the USA. Okay. In addition to that, we have also profiled the personality of every adult in the USA. By having close to 2 million people undertake a quantitative survey to probe, , the five core personality traits that, , drive our behavior.
Hannes Grassegger: 07:09 How did you, how did you make people, , enter that survey?
Alexander Nix: 07:14 Well, we’ve been rolling out this survey for about four or five years now. And, , we use, , online digital surveys and we also use telephone surveys and we probably have about eight or nine different surveys running through social media. , we’ve done some very large scale telephone engagement programswhere we, I think we’ve contacted here, , hundreds, thousands of people at a time. And, , then we also use email.
Hannes Grassegger: 07:41 So this is why you also get offline people?
Alexander Nix: 07:44 That’s correct. So we got a, a Fab,[inaudible] , population sample.
Hannes Grassegger: 07:48 Okay. So what would be your best use cases? You know, we have all these elections referenda all year long in Switzerland. What would be your best examples where my readers would see, oh, this is kind of a proof of effectiveness, like,
Alexander Nix: 08:03 well, you didn’t get bigger than the US Presidential Donald Trump’s campaign. We’ve done Ted Cruz’s campaign. We’ve got Ben Carson campaign, had probably another 20 senate congressional and gubernatorial races this year. I mean that, that’s a lot.
Hannes Grassegger: 08:18 Is there a specific example like a state where you would say, okay, we’ve done this and that in Ohio or just to, to, to give an idea of the specific outcome, how it would look like the,
Alexander Nix: 08:33 well, I think the most, I mean, the first time we’ve really showcased our technologies in America in a very public way was Iowa and during the presidential primaries working for Ted Cruz where I had enough time to be able to use data to segment and target down to clusters of, you know, dozens of people. I mean, literally we were creating messages for the smallest possible, , groups of constituents. , a village of people would get very targeted specific messaging and that would be different to the village next door or, or the group. , next, you know, the, the next group.
Hannes Grassegger: 09:14 [inaudible]. Okay. So, , if I’m not on Facebook, could you analyze me too?
Alexander Nix: 09:20 Yeah, I mean, Facebook is, is one, , source of data, but a lot of people in America aren’t on Facebook due to their demographics, their age. , and, and obviously Facebook is only, , provides any limited amount of information in an equally valid is , lifestyle data, what you read, what church you attend, what golf clubs you belong to, what magazines. , you buy what , what you buy in supermarkets, what you put in your, in your shopping basket here. That can be very revealing.
Hannes Grassegger: 09:55 And that’s data you get from companies such as axiom, Experian.
Alexander Nix: 10:00 Exactly. These are large data aggregators. They go out and they collect these, these data from the, , the actual end user companies and then they aggregate it and we can then buy it from these companies and then augment it together and Hygiene it, , to build our own data set.
Hannes Grassegger: 10:17 So, , would it actually would the whole thing work without Facebook? You’re saying if Facebook wouldn’t exist? Okay.
Alexander Nix: 10:25 Yeah.
Hannes Grassegger: 10:25 And, , you, you’re building a hash I guess for each individual then?
Alexander Nix: 10:31 Yeah. So I mean, what we’re trying to do is we’re trying to find commonalities between individuals such that we can cluster them together and start to serve them, you know, engagements that have been tailored for a specific cluster of people.
Hannes Grassegger: 10:46 Okay, perfect. And, , how, how do you communicate with people and you, I mean, you are actually only making the analyzis and then there’s advertisers who create something out of your recommendations. Is that correct?
Alexander Nix: 11:01 Well, that can be the case. , you know, especially in the commercial and brand space where we’re working with large companies who already have existing relationships with advertising agents.
Hannes Grassegger: 11:12 So in the Trump case?
Alexander Nix: 11:15 but in the Trump case, you know, we were running all Trump’s digital. So, , we would take offline data segments and we’d matched those two cookies in order to drive his digital engagement and social media engagement. And what this meant. I mean, bearing in mind, Trump spent, you know, 60, $70 million on, on digital and only 15 or 20 on, on television. , this is very significant. So pretty much every single message that Trump put out was, was data-driven and whether that was asking for a donation or whether that was, , you know, a message of persuasion and so forth. But equally for television, we also help the Trump campaign. We are able to match our offline data segments to set top box viewing data. So this is a record of, , of what cable channels you’re watching in your home, such that we can begin to identify what programs the audiences that we’re interested in are, , are watching. And wecan then buy advertising in the programs that have the highest density of our key target audience.
Hannes Grassegger: 12:22 So for my non-tech readers, this means you’re creating this profile, which is the five ocean treat, psychological model. , clustering, , means making a group of like minded people and then people and then sending, , and then sending a specific message to those people over a variety of channels.
Alexander Nix: 12:49 Exactly. And you can, you can get into such granularity that, that you can, you can start to look at a group of people plus a million people who will care about the same issue. You can then start to subsegment them, sub segment them, and then serve them different messages on that same issue, , in order to increase the efficacy of your messaging.
Hannes Grassegger: 13:13 Can you specifically hit individual [inaudible] opinion makers?
Alexander Nix: 13:19 Well, hypothetically, , the technology today can be implemented such that you could make it at a one to one level. , but, , from a purely economic standpoint, the return on investment wouldn’t be worth it. So it’s better to cluster people into, into groups of thousands of people or tens of thousands.
Hannes Grassegger: 13:42 How do you see how these people are reacting? How do you measure your, how do you get feedback?
Alexander Nix: 13:48 Well in elections, there’s no clearer feedback than, than, than how people vote. ([inaudible],…. , but in Lexical, I’m sorry,) in the commercial and brand campaigns, there’s normally some sort of call to action. , so whether you driving people to a website, , or you know, asking them to click on something cool, so forth. Ultimately, , we just look at, , a consumer, , spending data, , and we have a look at how that’s changes after we’ve targeted people with an advertising campaign. So it’s possible that we can target a million people with our messaging and we then have a look at consumer spending in the geographies where those million people are located. And we compare that to the people who didn’t receive our messaging or got traditional ads or mass messaging and we can look at the uplift.
Hannes Grassegger: 14:40 So I guess in the Trump case, you did that with a donations for the campaign, right?
Alexander Nix: 14:46 Yeah, and we were extremely successful. I mean, most people didn’t understand that Trump got 280 $290 million of small dollar donations. , almost all of this was driven through digital.not, not all of it, but almost all of it. And, , and it was all data-driven and highly targeted him. And when we’re talking about people donating anywhere from $5 to $20, so you could imagine how, how large that can be.
Hannes Grassegger: 15:14 So hiring you actually was great business for him because you cost him, you said around 15 to 20 million, but you raised how much around?
Alexander Nix: 15:25 Well, I mean we, I mean the small dollar donations, the Trump campaign raised about 280 million the last time I looked. Maybe it was a bit more, something like that.
Hannes Grassegger: 15:32 And that was through your, , mostly through you, like your actions, Cambridge Analytica?
Alexander Nix: 15:38 Well, yeah, we were the only company that was running digital.
Hannes Grassegger: 15:42 Perfect. And do you have customers in Switzerland already?
Alexander Nix: 15:47 We’ve had some inquiries, , in the, in the last six months as a result of the work we’ve done in America. , and we’re in discussions, but we haven’t started working in Switzerland. , it, it’s obviously it’s a hugely important market, but just given the fact we’ve, we’ve been involved in the, the presidential that has been our focus.
Hannes Grassegger: 16:06 So how bis is your company? how many people currently overall work at Cambridge Analytica?
Alexander Nix: 16:12 , about 150.
Hannes Grassegger: 16:14 150. And do you have customers in Germany? Like I’m talking about the political space only also for Switzerland.
Alexander Nix: 16:21 I’m sorry, I thought you were doing commercial space. No. So our current focus is on the US and the UK and, , , Asia and South America. , Europe, , political market is not so exciting for us at the moment. , for two reasons. One, the legislative framework for data is much more prohibitive, , in Europe than it is elsewhere in the world. And secondly, , you know, what, what the technologies we’re doing are, are expensive. And, , you know, a lot of European election campaigns don’t have the requisite funding to be able to invest in these kinds of technologies and methodologies.
Hannes Grassegger: 17:05 Okay. So what would you be the minimum entry barrier for a party to work with you? For a certain, just to give them a rough idea.
Alexander Nix: 17:17 Oh, that’s such a, , an open ended question because if, you know how big the electorate, , you know, if you’re modeling 200 million people or 10 million…
Hannes Grassegger: 17:27 …right? Switzerland is 8 million people…
Alexander Nix: 17:32 … specifically sentiments. , well, I mean, and, and, and what is it you’re trying to achieve? I mean, what resources are we working with if we’re unable to, you know, we would need to undertake a lot of proprietary research in Switzerland in order to get, , the data sets that we need because it’s unlikely that we’ll be able to acquire a license or purchase data because it probably doesn’t exist legally.
Hannes Grassegger: 17:57 Okay. So that is the basic problem in the European Union and probably in Switzerland too, that public Facebook data does not suffice?
Alexander Nix: 18:07 Well, I mean, we could be, broadly speaking in the United States, it’s a, it’s an opt out data policy as opposed to Europe, which is more of an opt in data policy, right. That you can still get the data, but it just takes more time.
Hannes Grassegger: 18:24 Oh, quasi on request. Okay. So, , last question is, , what is the minim amount of information, , that you need? For a certain person to build a profile, like, , psychological idea of him?
Alexander Nix: 18:42 I mean, the point about the big data, I mean as if the name would suggest is that as you know, the more data you have, the, the more insights you can generate. , it’s a bit like baking a cake is, , no one ingredient is, is , you know, more than, than the other. You need to have eggs and flour and sugar and chocolate and whatever else it is. , but I mean, I suppose if you have it like in America, if you have things like voting history, , that’s obviously going to be a very clear indicator as to like if a debate in the, in the future. , but equally, you know, demographic data could be quite, quite indicative of, of people’s say choice. , I think, I think that question’s a little bit limiting.
Hannes Grassegger: 19:35 I understand that if it simplifies the problem. Okay. , thank you so much, , for talking to me and a pleasure. Great evening. With your kids. , Thank you. Okay. You take care. Bye. Thank you, Mr. Nixon.
Alexander Nix: Bye Bye.
Hannes Grassegger: 19:57 Ooh, Ooh.
