20 years of AdTech and one conclusion: clean data changes everything that comes after it.

I had a life before advertising. About ten years before advertising found me or I found it. Software engineering at Apple (typography), database administration at Oracle, then a startup called iHarvest that was Pinterest before the world needed Pinterest during the dot-com boom. We were business-minded enough to develop our own contextual-advertising solution to fund the build, a hacky version of Google AdSense before AdSense existed. That gave me the first inkling there was something here.

After we exited, I joined Ink2Me, which was big in content networking and web search and powering search for AOL, MSN, and Yahoo before Google had the lion's share. Ink2Me was acquired by Yahoo shortly after I joined. Display, video, mobile advertising at Yahoo and then Microsoft, in product leadership roles that got me into the data behind targeting and measurement.

The most notable Microsoft moment: getting Microsoft into programmatic before it was a big thing. 2009-2010. We invested in a small company called AppNexus and chose them to power our ad exchange. We made Hotmail and MSN inventory available through the exchange (premium publishers weren't doing this yet). My team managed how to make it effective.

Back to Yahoo, where the most notable accomplishment was putting forward the strategy to build the Yahoo DSP (omni-channel for web, mobile, eventually CTV and video). Then IAB Tech Lab as CEO for four years (OpenRTB, taxonomies for data collaboration, privacy-compliant identity resolution). Then Meta as VP of advertising ecosystem, on the other side of the Tech Lab table. Now Snowflake.

When the Snowflake opportunity came, I had to think about it. And it fits like a glove because it's the combination of data and AI, which with basically no hyperbole is the foundation of everything for the foreseeable several years. If you don't have your data story together, you can't power AI. If you don't have your AI story together, you'll be beaten by your competitors. As a consumer, you'll fall behind in what you can do day-to-day.

Snowflake is a platform, not a solution provider for everything. A data-plus-app ecosystem, and increasingly an agentic-workflow-and-conversational-UI layer on top. We rely on a great many partners to enable around 12,000 customers to do what they want to do across marketing, media, and the other industries I look after. The role is to help Snowflake create that ecosystem for the industries I care most about.

It has been a rollercoaster. At IAB Tech Lab we were sometimes pushing the boulder uphill trying to get the industry to do more to protect data, both for security and privacy. At other moments, oddly, the threat-slash-promise of third-party cookie deprecation on Chrome became wind in our sails. Motivation came with it to develop innovative solutions that didn't even rely on knowing the consumer (anonymously), even more aggregated data. Compliance is the baseline. Nobody decides I won't be compliant. The opportunity was to leapfrog beyond the bare minimum into innovation that changes how some things work in the industry.

I'm optimistic that AI will help us come back to personalisation and effective measurement without having to know absolutely everything about everyone, which is clearly non-compliant and isn't the way forward.

Publishers are the proxy for consumers in the media and advertising space. They need healthy relationships with the people who view their sites and use their apps. Service providers, brands that act as publishers when they build their own microsites or apps: that connection to the consumer is everything. If you don't establish and maintain trust with the consumer, it doesn't end well.

The buy side increasingly seeks transparency from the sell side, and from agencies as more principal positions are taken in media. That ride has been bumpy. AI can help on both fronts by giving consumers, brands, and publishers more insight, even if it's aggregated. Signals as to what's happening with my data, with my ad buy, with the ads on my sites and apps, who's buying what, what kinds of purchases the attribution side is driving. The modelling AI enables is particularly helpful here. And on the consumer side, agents could be helpful in protecting and guiding the consumer experience.

There is more data available than ever. The scaffolding for compiling identity graphs (even on anonymised information) is more present than ever. The consent layer is now attached in a more granular way than it had been historically.

Modelling and machine learning let you take the inputs of all the people you do know about (anonymised IDs or otherwise) and build on that data. Beyond lookalike audiences, with more advanced machine learning you can take an input as a seed and expand to understand new attributes or behaviours you may not have thought of. Operating at machine speed you can see that in this zip code there's a higher propensity for X, in ways humans couldn't have understood previously. Know your consumer, know your prospects, and step back to look at novel slices of the data that AI enables.

The dynamics around AI come back to the regulatory environment and how (or whether) AI will be regulated, which differs by country. We don't yet know all the effects. Tariffs have had a lesser impact than feared initially, and it isn't over yet.

Geopolitical considerations around privacy, data, and AI regulation create a patchwork of compliance. Marketers and publishers have to know more about the law than they would have expected. The decision: behave differently in every single market and situation, or take a more global view that leapfrogs the regulations and innovates ahead of them. The latter is protection against being disrupted again and again as a new US state introduces a new privacy law or AI regulation. Forced innovation for good, for the consumer and for the company that does it well.

It's happening very quickly. Meta probably invested more in data science than any company and largely created the notion of marketing science with hundreds of people in that function. Meta has pared that function back. Other companies are realising they need that function. Not so fast: AI is kicking in.

Conversational AI talking to your data is already happening. Business users can talk to data much more directly without having to go to a data scientist to request a report and wait days or weeks. The progression from question to answer can happen in minutes.

That changes the role of the data scientist dramatically. You need fewer data scientists, and those who remain are more like architects: this additional type of data would be very helpful to my business audience; this AI model is misleading the business users, so we need to stop. Checks and balances plus enablement. The role evolves from serving the needs of the organisation to being strategic and pioneering. More satisfying work too.

It's circumstantial. Incrementality matters: understanding what's moving the needle versus what's adding to a pie you already had. Identity matters a great deal as long as you're operating in the realm of privacy. Identity is the foundation for planning, targeting, understanding who you're going after, and for closing the measurement loop. You don't need to know every person; you can model, sample, use synthetic data.

What I'm optimistic about: bidirectional relationships between the buy side and the sell side. If data collaboration works (in clean rooms or in more aggregated form when it doesn't need to be a clean room), the marketer can give insight to the publisher about what happened on the publisher's properties (attribution), and the publisher can give insight back to the marketer about which audiences worked best. Agents can package them. That flywheel creates innovation and iteration faster than we ever have.

Ad tech and martech convergence is partly about the actual technology converging, partly about the paid and unpaid channels being more flexible and connected (the orchestration layer). It comes back to identity. If I know you and have a relationship that lets me email you, then text you, then serve you ads, that hasn't been quite as connected as it should be.

The approach at Snowflake is composability. Composable customer data platforms, and composable solutions across marketing and media. Discover the data, connect to it, discover the apps that operate on top of it (best-of-breed point solutions that interoperate), and install them. Agents on top to orchestrate. Plus the graphical workflows so a buyer can assemble it themselves.

It's not the monolithic solution that does everything. It's more like using your phone: choose the apps that make the most sense for the job to be done. A mix across providers of apps and providers of data. Some partners are realising they can be the orchestration layer and don't have to build every point solution. Why reinvent email as a channel if there's a best-of-breed solution already? That's where convergence will start to drive.

I have a little view of the existing standards from IAB Tech Lab. The disruptive question: private marketplaces in the advertising space (a publisher packaging inventory and offering it only to selected buyers). Why would that not be done with agents? The publisher agent says I have these things. The buyer agent says I want these things. Where do they overlap and how much am I willing to pay or sell for? That doesn't require the same protocols. Large language models are all about language and can understand it.

Real-time bidding and the protocols around it will need to change. Other standards will evolve. A lot of it has to do with how agents communicate (some advertising-specific, some not, some marketing-specific). Maybe the AI can do the standards work for us.

I have gotten on the agent bandwagon very quickly, having been candidly sceptical. There's an 80-20 rule: it isn't all going to happen perfectly as quickly as some hypothesise. Five years out, this industry will be almost unrecognisable in some ways because of how much is happening in an automated and agentic way.

This is a faster time of change than the move to programmatic, which was the other moment in my time I'd highlight as wow-that-was-disruptive. This is a hockey-stick kind of change.