Deep learning beats templated bidding. Business outcomes beat vanity clicks. And you cannot build a brand inside a chatbot.

My journey has been nonlinear, but it covered all of marketing and advertising, which gave me a full view of the space. My first marketing job was at Digitas in 2000, helping blue-chip companies understand the internet, building things like the first online American Express rewards platform and an HBO email newsletter. After business school I moved to media and spent the standards years at the IAB, from about 2005 to 2009, when it was still the Wild West, learning how to run a media company. I went into ad tech on the publisher side at CBS Interactive, where we launched the first premium exchange, then moved to Rubicon Project, which became Magnite after we went public. Seeing the programmatic guts of the industry, and a new technology called deep learning, I co-founded Cognitiv in 2015. I am an engineer by training, and I have always wanted to work on the new interesting things.

At Rubicon Project we saw all the bids the DSPs were making, and they were formulated and templated, ten dollars, one dollar, ten cents, with little intelligence campaign to campaign beyond context or the number of cookies. Then my co-founder Aaron told us about this new thing from his Stanford postdoc called deep learning, a powerful predictive technology that needed a lot of data. I said I know where there is a lot of data, programmatic advertising. So we started the company on the idea that programmatic data can inform a deep learning algorithm to predict consumer behaviour far better and drive performance, beginning at the bottom of the funnel.

We were first movers, and old investor decks show a vision that is now realised: a full funnel algorithmic solution for a marketer. Our tagline from the beginning has been the brain behind the brands. Ad tech is built on one tactic, context, audience, DSP, DMP, so we had to start somewhere, but over the years we filled out a deep learning advertising platform that works across all channels and all of the funnel. Marketers say they have all this data in their CDPs, but putting it to work has largely not been practical. Deep learning, now synonymous with AI, has made that possible.

Everything we do is based on the marketer's goal and the campaign KPI. The most advanced marketers want a business KPI, and that is the next evolution. Before, we were held back to digital KPIs, a click, a view-through, a site visit. Most businesses do not drive revenue through those. Insurance companies want lifetime value and incremental lift, consumer brands want net new customers. Those do not come naturally from the infrastructure of the internet, but they are measurable, and our deep learning platform built the event systems to run marketing toward them.

It is just that real outcomes have not been an option before. That has always been the promise of digital marketing since I started, it is so measurable. But how do you use the data, what does it mean, how do you connect offline and online? We can now do that effectively, and marketers have the option. The question is whether they know, or whether they are defeated enough that they are still using clicks.

In 2015 AI was not living up to its potential, so saying AI was not good in our pitch, and we talked about deep learning, and that DNA has stuck. It is a litmus test. A company can use machine learning and say it has AI, because AI is an umbrella term, but if you are not saying you use deep learning, it could be anything, because deep learning is a subset of machine learning.

We have stuck to our message and not pivoted, always building this full funnel, omni-channel algorithmic solution. That has given us good brand recognition and trust. The three founders come from STEM backgrounds, and two of my co-founders and I have known each other since we were ten. Our position is smart, trusted, honest and transparent, because the algorithm is doing it, so we have nothing to hide. We break it open and show the insights, because that is the power.

The cookie apocalypse was coming, and as a performance technology we needed something that performed without cookies. Advertising is meant to be the right message to the right person at the right place at the right time, and ad tech has invested in only one piece of that per company. No company answers all four questions at once, you are an audience company, a context company, a DSP, or a dynamic creative company. We are one of the few that treats the auction and impression on all four at once, scientifically. We were already doing contextual understanding, so we pushed it out as ContextGPT, and we were the first to use prompts to target context. True contextuality does not scale, so we integrate LLMs on top to trade exactness against reach transparently, with full visibility into every URL.

We integrate with made for advertising definitions and similar tools, but brand safety is sort of solved from the advertiser's perspective. They have their own tools and lists of acceptable and non-acceptable sites. It is not much of a concern for us, because we can read the language on every page, run sentiment and inclusivity analysis, and filter everything out.

The standards organisations across Europe and the United States brought order when there was chaos. When I started at the IAB there were more than three hundred ad sizes, AOL, Yahoo and MSN each had their own, and advertisers could not function or create three hundred ads, so they would not spend. We fixed that down to ten or twenty. One of the most successful standards I worked on, with Ari Paparo, was VAST, which standardised the communication between the player and the ad server. Suddenly you could buy and scale video advertising no matter whose player you had, and it was adopted like wildfire. The IAB is very good at opening the floodgates for advertising.

Look at the class of 2014 to 2016 in ad tech and there is almost nobody left. We started when AI was a buzzword with venture capitalists but advertising was not good, and we have no VC backing, so we are really independent. Great angel investors let us be very capital efficient, which slowed our growth and also let us build at the right time, because deep learning was just beginning and we had to wait for things to catch up. We grew slowly, then after COVID it took off.

I am interested to see how agents evolve, but sales in our industry is still very relationship-based, and delivering your value proposition takes time, it is not instant. It will be interesting to see whether agentic selling fully replaces a sales team. And premium publishers have strong brands and custom value, so I cannot see that being misplaced by AI in the near future.

Chatbots are becoming a large part of life, but malls are doing better and younger people like to touch and try clothes, so we have two competing pieces of the journey. If brand did not matter, On and Hoka could not have built huge brands against Nike and Adidas, so brand is always key, and you are not going to build it inside a chatbot. As a student of history, I do not see those human and genetic needs for interaction, experience and touch changing. You can buy something online, but then you have to return it, and getting them to keep it is the key. I stay sharp by going to South by Southwest alone every year and sitting in non-advertising panels, getting cross-exposure to the world and to humanity.

There is a debate that institutions should specialise more and hand out vocational degrees, and I could not disagree more. With AI, specialisation could be taken away from you tomorrow. The key, especially in coding, is seeing a problem, designing a solution and understanding its effectiveness. So critical thinking, flexibility and problem-solving are the key to any career's future. You have to stay flexible, because you never know what is going to happen to the job you have taken on.