Our brief history of AI and mass-surveillance

The recent revelations regarding the horrific privacy breaches by Facebook and Cambridge Analytica might have only now made it to the mainstream press – but the problem is far from new, neither is the technology.

Tracking users activity across the web and customising their content dates back 15 years. In fact, we were right at the centre of this in 2004 when we created a web-platform that was designed to serve bespoke content based on learning – we called it VITES and spent six years developing it from a simple tracking application into a sophisticated content delivery platform.

In 2004 a client had come to us with a problem; they were spending upwards of £500k per month on web traffic across hundreds of channels, yet didn’t fully understand which channels converted and which ones didn’t. Often, the conversions didn’t occur on the first visit; sometimes it would take weeks or months.

VITES was based on the simple concept of mixing data acquisition, automated learning and testing to increase website conversions – and gave our clients a key competitive advantage.

To do this, we developed a platform that uniquely tracked every user across their entire web-journey, across multiple visits and multiple channels storing conversion data in a database. We learned that the more data we stored, the more insight we gained and the bigger the competitive advantage our clients achieved.

We spent 100s of thousands of pounds developing the platform and quickly realised we were onto a killer application, but we lacked the multi-million-pound capital to develop it into an off-the-shelf platform. And it was complex and fiddly to use, written in an older web language (Perl) that was falling out of fashion.

At about the same time we were approached by Advertising.com (Ad.com), who was, at the time, one of the world’s largest advertising networks. We pitched our idea as “The Universal Cookie” and pitch went something like this:

  • Every visitor/user to any of Ad.com’s clients would be “stamped” with a unique ID. At the time, Ad.com had thousands of clients across globe managing 100s of millions of visitors.
  • A data store as Ad.com would create a unique record of every user’s action, every page they visited, and every piece of data they left on every website
  • Users would then be shown a range of conversion approaches automatically until the system was confident of predicting the conversion rate.
  • Ad.com would then “categorise” different users groups and show tailored content to them based on the propensity to convert.

In hindsight, it was brilliantly simple (even if we say so ourselves). It would gather vast amounts of information from 100s of millions of users and then leverage the learning using automated testing to influence the outcome of every conversion across thousands of websites.

However, Ad.com had recently been acquired by AOL/Time Warner who, as they had a record of doing, was valiantly missing every opportunity on the ‘net. They additionally had (valid) privacy and data protection concerns, and as a result, the deal went nowhere.

We had some other interested parties, including some international agencies but mostly what we were talking about was either too technical, restrictive, or had privacy issues – but, mostly, folks didn’t get it.

We eventually canned further development on the platform in 2010 after the project lead left the business (to go on and develop a modestly successful “clone”) – we just couldn’t continue funding the expensive development for a few clients.

The big-picture sell of creating a publisher-level platform was dead, for us, and the explosive rise of new information-based advertising networks such as Facebook and Google sealed its fate.

But, in all fairness, we hadn’t thought it through properly. We never identified the risk of abuse by data gatherers, nor did we consider amalgamating data from third-parties. We were close, but no “banana”.

Looking back nearly 15 years, we had created the forerunner of the monster data gathering platforms that dominate the web today. Interestingly, our close brush with mass-surveillance and privacy instigated our adoption of privacy-first, open-source and transparency initiatives in 2013.