Madgic: Powering quality-focused ad tech with no tricks, just growth

About Madgic

Madgic is a programmatic mobile advertising platform optimizing the performance and quality of ads and traffic for its supply and demand customers.

Industries: Technology
Location: France

Madgic grows from 3 billion to 30 billion requests per day in a year and a half by moving to Google Cloud Platform, extending its global reach, reducing latencies, and gaining flexibility to take on new partners.

Google Cloud Results

  • Reduces latencies by up to 200 milliseconds thanks to servers located closer to its clients
  • Keeps costs down with autoscalable preemptible VMs on Compute Engine
  • Powers a cost-effective way of screening ads automatically using Cloud Vision API

From 3B to 30B queries per day in 1.5 years

Digital marketing is being transformed by programmatic (automated) advertising strategies, which use algorithms to buy and sell advertising space through real-time bidding. This approach potentially offers more targeted and effective advertising, but it also requires a more complex supply chain. Founded in 2011, Madgic resolves concerns about brand safety and inventory quality by filtering incoming traffic from its supply partners and delivering higher quality ads.

"Madgic functions as a filter," says Olivier Chouraki, CEO at Madgic. "We receive 400,000 queries per second from our supply partners, but we only pass on 40,000 queries per second (QPS) to our demand partners. We block any traffic that is not profitable, including fraudulent traffic and bots, and we also enable our supply partners to control precisely which ads they display."

"Scalability is the main reason for our move to the cloud. In order to take on new publishing partners, we need to be reactive, with a platform that will adapt instantaneously to our needs, and we need to be able to do it with reasonable costs."

Olivier Chouraki, CEO, Madgic

"Of course, processing 400 QPS involves a number of technological challenges, including capacity and response times", continues Olivier. "Previously, we had dedicated servers in the United States and Europe, but there were certain global partners with whom we just couldn't work, because of latencies. " In order to expand its geographical reach, Madgic needed servers hosted worldwide, but it also wanted to adapt its platform architecture to become more flexible and scalable, while maintaining cost-efficiency. Google Cloud Platform (GCP) provided the solution.

"Scalability is the main reason for our move to the cloud," says Olivier. "In order to take on new publishing partners, we need to be reactive, with a platform that will adapt instantaneously to our needs, and we need to be able to do it with reasonable costs."

Creating a global, scalable platform

In real-time bidding, the highest bidder wins the auction, but being able to bid fast enough is equally important, and when volume constantly goes up and down, this is a technical challenge. Before, Madgic was spending money on dedicated servers that were not able to respond effectively to highs and lows in traffic. "With our previous partners, it could take up to two weeks to add additional servers," says Olivier. "Conversely, if there were lulls in traffic then our servers were underused and money was wasted, as they were already paid for in advance. We also needed servers hosted in Asia, which they were not able to offer."

"Using the Cloud Vision API, we analyze ads to detect any sensitive content. We can use this information as a set of keywords for filtering and targeting ads. It's a very efficient, fast, scalable, and cost-effective way of assessing ad quality."

Olivier Chouraki, CEO, Madgic

To get the scalability and performance it needed, Madgic moved its servers to GCP, spread across five regions worldwide: U.S. West, U.S. Central, U.S. East, Europe, and Asia. "Bid requests from our supply partners all over the world are processed by Cloud Load Balancing, which then sends on requests to Compute Engine. That's the most important tool for us," says Olivier. Once the auction has taken place and the ad has been displayed, reporting information is then sent to Cloud Storage to be stored before being loaded into BigQuery. "We upload around 50GB a day, but we don't store all the ads and clicks, and we aggregate the data before uploading it. That helps keep costs down," says Olivier. "We can then use that data to monitor our platform, to optimize our fill rates, and for reporting win rate."

In order to filter ads for sensitive content, Madgic uses the Cloud Vision API. "Using the Cloud Vision API, we analyze ads to detect any sensitive content. The API can recognize ads for alcohol, dating, or gambling. It's really impressive, it even works with video ads," says Olivier. "The API recognizes logos and various objects — such as cars, balls, and animals — and retrieves all the text appearing in the ad. We can use this information as a set of keywords for filtering and targeting ads. Thanks to the AutoML feature we can feed the Cloud Vision API with our own quality tags set by our Ad Quality Community and after some time it will send back these tags automatically. It provides all the benefits of a custom Machine Learning solution with minimum work and at a very low cost. It's a very efficient, fast, scalable, and cost-effective way of assessing ad quality."

Keeping costs under control

As well as the flexibility to add and remove servers automatically according to changing traffic, shifting to a scalable infrastructure also offers the opportunity to keep costs under control with smart usage.

"Since moving to Google Cloud Platform, our traffic volumes have increased from 3 billion requests per day to 30 billion requests per day. Our objective next year is to double the volume again. With the scalable tools offered by GCP, that kind of growth is now possible."

Olivier Chouraki, CEO, Madgic

"With GCP, we make the most of the opportunity to use Preemptible VMs," explains Olivier. "In addition to our data servers, we use a group of preemptible VMs for our ad servers that autoscales throughout the day, according to our need. Those servers only store temporary data, and are 3.3 times cheaper for us to use. Using preemptible VMs helps us to maintain our margins, as infrastructure is our biggest cost."

Doubling the volume of traffic

Since moving to GCP in September 2016, Madgic has gained in flexibility and agility. "Now, rather than waiting up to two weeks to add servers, we can scale up in minutes," says Olivier. "That means we can take on partnerships with large publishers with no delay and accept more volume whenever they are ready to scale, which is a huge business advantage." Thanks to the global network of GCP, Madgic has also eliminated its latency problems. "Having servers worldwide means we can bid faster: with servers close to a partner, we gain up to 200 milliseconds, compared to having servers overseas," says Olivier. "We're much closer to our customers."

Now, Madgic is planning to use more of the tools available on GCP: using Kubernetes, Docker, Prometheus, and Stackdriver to make its architecture even more flexible, reliable and cost-efficient and moving its MySQL database to Cloud SQL, extending its use of Cloud Vision and growing the business further. "Since moving to Google Cloud Platform, our traffic volumes have increased from 3 billion requests per day to 30 billion requests per day", says Olivier. "Our objective next year is to double the volume again. With the scalable tools offered by GCP, that kind of growth is now possible."

About Madgic

Madgic is a programmatic mobile advertising platform optimizing the performance and quality of ads and traffic for its supply and demand customers.

Industries: Technology
Location: France
Google Cloud Platform logo

12 Months FREE TRIAL

Try Kubernetes Engine, BigQuery, and other Cloud Platform products with $300 in free credit and 12 months.

TRY IT FREE
Google Cloud Platform logo

12 Months FREE TRIAL

Try Kubernetes Engine, BigQuery, and other Cloud Platform products with $300 in free credit and 12 months.

TRY IT FREE