Michael Bailey, senior director at Google Media Lab, walks through several AI solutions Google has prototyped on behalf of our own brands. While Google’s AI journey is not yet complete, we are seeing real promise in areas like measurement, creative, and media.
Here at Google Media Lab, we’re always looking for ways to apply AI on behalf of Google’s brands — and, increasingly, we’re finding them.
Our team oversees all facets of Google’s advertising campaigns for our own products — including media, creative, and measurement. We’re bullish on the potential of AI, but knowing when, where, and how to introduce it to our workflow can still be a challenge, especially as we seek to progress from AI hype to execution to, ultimately, business impact. That’s why we take a structured approach, with clear governance and rules of engagement. Our ambition is to go from “a thousand flowers blooming” to an “AI garden” that’s a little less wild and a little more manicured, with a clear path to delivery.
While we don’t have all the answers, we thought it might be useful to give Think with Google readers a behind-the-scenes look at how Google Marketing is putting AI to work using Google’s own products, along with our key learnings. The tools featured below represent some of our early promising marketing experiments across three areas: measurement, creative, and media. Taken together, these tools also reflect our efforts to transcend point solutions and adopt an end-to-end approach capable of spanning various marketing disciplines and ways of working.
While we built these tools to support Google Marketing campaigns, it’s important to note that any advertiser can build similar tools to support their own campaigns using Google Cloud, Gemini, and other AI models.
Measurement: AI supports creative performance
Measurement is historically backward-looking, something many marketers do only after a campaign has ended. And traditional measurement can be laggy, with insights arriving too late to be actionable. We wanted to apply measurement insights much earlier in the process, so we built an internal prototype using the Google Cloud Platform suite of AI marketing tools that can help surface real-time creative insights and inform future decision-making.
Beginning this year, we used AI to generate a catalogue of features for our huge library of Google ad creatives spanning Google brands such as YouTube, Pixel, Android, and more. These features could be in almost any format (text, audio, or image-based) and could range from basic (“brand logo top right position,” “has a clear call to action,” or “characters are having a dialogue”) to highly specific (“contains natural landscapes, e.g. mountains, lakes, rivers, etc.” or “language includes words that evoke surreal or abstract concepts”). In all, our prototype can capture over 50 different types of features1 within every creative we produce.
Using this catalogued metadata, we then used predictive AI to train a model using ad performance data for those same creatives. This predictive model can score how likely a new, untested creative is to deliver lift against a specific KPI. In this way we are able to align our AI-generated creative variables to key outcomes, while reducing our reliance on more traditional forms of creative performance evaluation. It’s still early days, but internal testing has found approximately 70% overall accuracy of our Creative Performance Predictions, defined as how often the model correctly predicts whether a creative will drive brand lift.2
Another powerful example of AI supporting creative measurement pertains to qualitative data. We are using Google Cloud’s latest generative AI models to sort and summarise the qualitative feedback from our creative testing panelists, including open-text responses. AI can summarise both positive and negative feedback into a set of key themes, rate how much people like the creative, then make creative recommendations for improvement.
For instance, we applied this qualitative feedback analysis to our latest Pixel ad for the UEFA European Championships. Despite testing very well in-lab, Google Cloud’s generative AI model was able to identify that about half of respondents who mentioned music disliked our selection. Its recommendation: “Explore alternative music options or offer variations to cater to a wider audience, as the current music choice received mixed feedback.”3
At Google Marketing, well-organised data is our foundation and a necessary step before applying AI solutions. Once we did the pre-work of making our data manageable, building these models was relatively quick.
Creative: AI speeds up campaign development
Culture moves fast. Sometimes you see a trend coming, scramble to come up with a brilliant idea, and move through production and approvals, only to find that the trend has passed. Could AI help speed up this process, resulting in fewer marketers missing out on culture-driven opportunities?
To help solve for this, we’re experimenting with internal prototypes powered by Gemini models to catch trends and to generate campaign ideas. These tools work together in interesting ways.
We built a trends and insights assistant to surface trends from several sources in real time and to visualise spikes and dips in public conversations. We can use it to take the temperature of the culture broadly (What’s buzzing at the moment?) or to go deep on a specific topic or keyword (insert your favorite sports team). It can also generate a detailed summary of a topic with several views, including a timeline, key players, and more. Without real-time data, this could take weeks, not minutes.
Once we identify a trend we’d like to include in our messaging, we can use AI to generate campaign ideas. This tool doesn’t generate a ready-to-go idea but, rather, provides thought starters. Once we find an inspiring thought starter, we’ll request to expand on it. The tool then offers specific suggestions to bring the idea to life, for example, through activations with creators and media partners. One perk with this tool: It has no ego and doesn’t get tired. We can request and reject an unlimited number of ideas before choosing a selection of ideas to build on with our agency partners.
AI can supercharge your creativity. With the time saved identifying trends and coming up with thought starters, you can refine and polish your creative ideas and storytelling.
Media: Using AI to find new customers
The audience signals that advertisers once relied on are becoming harder to use, and in some cases disappearing altogether. Our team wondered, can we use predictive AI to understand who will find our message most helpful?
To find out, we are building an internal prototype that uses our first-party data to discover users who are likely to perform a particular action that we care about, such as installing an app or activating a new phone. We use these insights to reach a broad audience of similar users on services like YouTube, Google Display Network, and Search.
Early internal tests of this prototype have shown a 72% increase in incremental outcomes at a 43% reduction in cost.4
To do this at Google, we created a cross-functional AI team and process to advance experiments like AI Audiences, with representation from marketing, legal, privacy, and product.
Advancing AI together
We’re not doing this work alone. Developing prototypes like those above requires close collaboration with our agencies and other partners.
While we’re early in our AI journey and don’t have all of the answers, the results we’re seeing from these and other experiments suggest a bright future for AI-powered solutions. We invite you to join us in our industry’s AI journey by experimenting, building prototypes for your own needs, and collectively elevating best practices across the full spectrum of marketing activities.
A few final tips as you get started: First, start with embedded AI solutions that your team is already using. I might be biased, but Google Ads is a good place to start. Next, actively experiment with those solutions to build muscle in areas like writing prompts and processing approvals for AI-generated content. Once you’re ready, you can build your own new solutions like we’ve showcased today, aligned to your high priority needs. Finally, be sure to collaborate closely with partners along the way to unlock technical capabilities and identify opportunities to customise solutions to serve your distinct marketing goals.
In the meantime, check out these six practical tips to help you put AI to work.
