From the beginning, marketers have measured the success of their digital advertising in very inefficient ways, like how many people click on an ad. However, there has never really been a correlation between people who click on a banner or text ad and the marketer’s objective, whether increased sales on the website, more foot traffic in the store, or better perception and awareness of their brand. Now you can correlate them.
“We’re all in on our technique that we call ‘incrementality measurement’ for advertising,” says James Hassett, CTO at Martin, a modern media buying and measurement platform for marketers. “We use it to help our clients differentiate between actions that consumers were going to take, regardless of whether or not they saw advertising, versus actions that consumers took because they saw the advertising.”
The marketing discipline combines art and science. Great campaigns can be born in a burst of inspiration, but at the end of the day, success is driven by metrics. As a demand-side platform (DSP) for advertising, Martin is committed to empowering marketers and media buyers with precision tools for measuring the true impact of their marketing.
“Metrics often don’t show a causal relationship between advertising and business outcomes,” says Lewis Rothkopf, President of Martin. “We think that’s crazy, so we help our customers understand the real-world impact of their advertising and then help them make quick decisions on the fly to optimize what’s already working.”
Managing their DSP platform required a cloud strategy capable of converting the art of advertising into the science of analytics, with the ability to handle intense compute requirements. In short, Martin needed to manage a lot of data and turn it into insights fast in an industry where responding to customer sentiment at lightning speed can make all the difference in campaign success.
To solve their business challenge, Martin turned to Google Cloud. One big selling point? Google Cloud’s history as the developer of Kubernetes, the industry-leading container orchestration system. Managing a heavy data load was key, so Martin needed to complement Google Cloud’s Kubernetes capabilities with other Google Cloud platforms and products, including BigQuery. Equally compelling was Google Cloud’s focus on cost-effective data management.
“Google Cloud has a well-earned reputation for being very good with Kubernetes, having invented it,” says Hassett. “That was a large part of the services that we wanted. We’re very compute-intensive, and we want to control that with Kubernetes. We also like BigQuery for the trillions of bid requests that come through every month. The prospect of putting that all in a data warehouse and not paying for it until you analyze it was very enticing.”
Martin’s engineers also found that BigQuery integrated well with a critical application they use called ScyllaDB, an open-source distributed NoSQL wide-column data store. ScyllaDB is designed for ultra-low database latency and high throughput. “Scylla is a cache for our hot data since it is accessible within milliseconds,” says Ryan Ross, Senior DevOps Engineer. “As billions of bid requests come in, we need to make bidding decisions very quickly. Scylla can respond to these requests in real-time. For a long-term solution and more long-term data storage, we rely on BigQuery.”
Additionally, Martin thinks Scylla and BigQuery are best suited to different use cases. They use BigQuery for data bound for analytics or AI because they can store as much data as needed and only pay when they analyze it. “But we also need to run the operational bidder that performs auctions, and there’s lots of information we need in under 100 milliseconds,” says Hassett. “We use Scylla to handle all those requests and respond to them within milliseconds. What’s great about being on Google Cloud is our ScyllaDB load is predictable, so we can use committed instances to save money.”
When Martin needed to make their bold play on Google Cloud, they approached SADA’s expert Solution Architects and Cloud Engineers for guidance. One advantage they noticed with SADA was how familiar the team was with various third-party tools they use.
We expected SADA to be knowledgeable about Google Cloud, and that’d be the extent of how they’d support us, but when we mentioned non-Google Cloud tools like Terraform and Airflow, the team proved quite knowledgeable of them and provided some support, as long as they were industry-standard tools.James Hassett | CTO at Martin
Also, Martin liked comparing learnings with SADA on a technical-person-to-technical-person basis. “It’s really nice to ask another human, ‘Have you seen this particular use case for this piece of tech?’ Cutting through the noise and gaining from someone else’s experience has been really useful,” says Ross. “That’s why we chose SADA.”
Hassett credits communication as key to Martin’s successful implementation with SADA. “In many situations with vendors, there’s an extra layer to get to the engineers when you have a very specific problem,” says Hassett. “That wasn’t the case with SADA; there’s just direct communication, usually via chat with engineers from SADA and Google Cloud, who we can describe the problem to because they’re familiar with our language and tools.”
As a result of working with SADA and Google Cloud, Martin has been empowered to take advantage of a cloud-native strategy without the burden of adapting legacy technology. Overall, by engaging SADA, Martin was able to:
- Achieve millisecond access to ad bid data
- Scale BigQuery to store billions of ad bids at no cost until analyzed
- Receive technical support for non-Google Cloud tools
- Connect directly to engineers via chat
“Google Cloud allows us to show results to clients in seconds,” says Hassett. “When our clients launch a new campaign on our platform, they see the spend, conversions, and effect of their marketing in real-time. By cost-effectively storing everything in BigQuery, we can report on more metrics than competitors. Because we’re transparent, we show our clients all the bid requests and associated costs. We can retrieve hundreds of terabytes of that data from BigQuery at phenomenal speed.”
Improvements in Martin’s platform have been qualitative as well as quantitative. “Another advantage is that we can run sophisticated AI on BigQuery,” says Hassett.
We can provide recommendations for clients’ campaigns as they’re happening, with sophisticated analyses like anomaly detection and pattern matching, which are difficult tasks without tools like BigQuery.James Hassett | CTO at Martin