AntVoice: Speaking to new customers with predictive targeting on Google Cloud

About AntVoice

AntVoice helps brands and merchants to identify and target new clients with a potent mix of behavioral analysis, powerful algorithms, and advanced AI techniques. AntVoice helps its clients turn digital ads into smart ads.

Industries: Technology
Location: France

Antvoice uses Google Kubernetes Engine and BigQuery to power its advertising recommendation engine, helping companies provide users with the content they want to see.

Google Cloud results

  • Provides more relevant ads customized for each end user, quadrupling its client base in just 18 months
  • Scales automatically to accommodate up to 8,000 requests per second with Google Kubernetes Engine
  • Reduces infrastructure costs and maintenance, allowing developers to focus on building the best products they can

Handles 1 TB of data a day with BigQuery

Online advertising can be an affordable, effective way for a company to get its message out to as wide an audience as possible. However, because most ad selection works by analyzing customers’ recent shopping history and targeting the same kind of products at them, customers often see ads for products they’ve already bought. This runs the risk of wasting money on redundant ads and even tarnishing the reputation of the brand.

French startup AntVoice aims to solve this problem with its “predictive targeting” AI. “Traditionally, ads are targeted around what people have already bought or browsed for,” explains Sergii Salata, Technical Lead at AntVoice. “That’s not what we’re trying to do. We’re finding new customers for our clients, who might not have seen the product but will be interested in it.”

“For our ad recommendation technology to work, we need to be able to process large amounts of data very quickly, scale on demand, and quickly respond and adapt to changes in the environment. For us, the best infrastructure to achieve all that is Google Cloud.”

Sergii Salata, Technical Lead, AntVoice

AntVoice’s proprietary algorithms help to predict what its clients’ customers likely to need in the near future. For instance, customers who have browsed for holidays in July might be served ads for summer clothes. The goal is to recommend more interesting ads for more relevant products, providing a win-win for retailers and customers.

AntVoice began as a social media application publisher in 2011, before pivoting to work on its recommendation algorithm for ads. In 2018, the company went to market and today counts some of France’s biggest brands as clients. But AntVoice couldn’t deliver world-class service without a world-class infrastructure. To get to this point, AntVoice turned to Google Cloud.

“For our ad recommendation technology to work, we need to be able to process large amounts of data very quickly, scale on demand, and quickly respond and adapt to changes in the environment,” says Sergii. “For us, the best infrastructure to achieve all that is Google Cloud.“

Power and flexibility with Google Cloud

When Antvoice began research and development work on its new solution in 2014, it had to set up its own infrastructure for the first time. Choosing Compute Engine virtual machines gave the company an easy-to-use, fully functional and affordable infrastructure to use for research and prototyping. But as time went on, it was clear that the company needed a more heavy-duty solution for the final product.

AntVoice’s predictive targeting methods rely on more than impressive algorithms. By treating users as individuals rather than part of wider demographic groups, AntVoice can recommend products tailored for each user’s unique needs, but even just collecting and storing the data needs considerably more resources than a traditional solution.

Being able to access and store all that data is no use if it can’t be used effectively. For the recommendation algorithm to be effective, AntVoice has to handle its data at very high speeds. Whenever a user goes to a website that serves ads, companies like AntVoice have a window of between 50 and 80 milliseconds in which to bid for ad placements. In that short time, AntVoice has to decide what ads to bid on, what would be the most relevant ad to display and how much to bid. That requires the ability to ramp up compute power and memory as and when the company needs it. At the same time, AntVoice needed a low-maintenance solution that kept costs to a minimum.

“We moved to Google Kubernetes Engine because it allows us to scale up and down on demand. It gives us a lot of power, but also a lot of control. We can share resources between different services and alter the CPUs or the memory, depending on what we need.”

Sergii Salata, Technical Lead, AntVoice

Although it was already using Google Cloud, AntVoice kept an open mind and evaluated other cloud providers and their tools. Thanks to its pricing and ease of use, Google Cloud ultimately proved to be the best fit for the company. With BigQuery, AntVoice can easily store and analyze huge volumes of data. For integrating new data, the company was relying on third-party messaging service which required a lot of manual oversight. By migrating to Pub/Sub and Dataflow, AntVoice can ingest large streams of data without having to worry about provisioning new servers or manually adding more memory.

Aside from the data handling, most of the infrastructure was previously running on virtual machines. But after going to market, and as loads became greater, AntVoice began to look for a more scalable approach. Today, although some stateful applications still run on Compute Engine or App Engine, most of the infrastructure has moved to Google Kubernetes Engine( GKE).

“We moved to Google Kubernetes Engine because it allows us to scale up and down on demand,” says Sergii. “It gives us a lot of power, but also a lot of control. We can share resources between different services and alter the CPUs or the memory, depending on what we need.”

“With Google Cloud, we’ve been able to help our clients find new customers and make sure that their existing ones see more relevant ads. It’s helped us grow as well. We have four times the number of clients now than we did at the start of 2018.”

Sergii Salata, Technical Lead, AntVoice

Scaling on demand, minimizing maintenance

When AntVoice went to market in 2018 with its predictive targeting solution, it quickly made an impact and gained clients straight away. Thanks to Google Cloud, the company was able to handle the additional stresses on its data infrastructure with ease and currently ingests around one terabyte of new information every day. As well as the increase in data storage, AntVoice is keeping up with an explosion in overall activity. Transitioning to GKE has helped the infrastructure easily handle an increase from 200 requests per second at the start of 2018 to its current average of 8,000 requests per second.

Meanwhile, AntVoice has been able to keep costs and time spent on maintenance to a minimum with autoscaling and managed services. Developers can focus on creating the best possible recommendation algorithm for the clients, who in turn pass the benefits on to their customers.

“With Google Cloud, we’ve been able to help our clients find new customers and make sure that their existing ones see more relevant ads,” says Sergii. “It’s helped us grow as well. We have four times the number of clients now than we did at the start of 2018.”

AntVoice expects to continue growing its infrastructure to provide for an estimated 20,000 requests per second by the next six month. The company has also made plans to expand the business to international markets, specifically across Europe and the United States. Google Cloud has simplified the expansion through its global network of data centers, which comply with local data security requirements. Meanwhile the integration with third party tools such as Terraform allows AntVoice to replicate its infrastructure quickly and easily across multiple locations.

“We’ve received really good support from the Google Cloud team and we’re continuing to get great assistance as we expand,” says Sergii. “We’re really happy to work together.”

Graph: RPS vs CPUs

About AntVoice

AntVoice helps brands and merchants to identify and target new clients with a potent mix of behavioral analysis, powerful algorithms, and advanced AI techniques. AntVoice helps its clients turn digital ads into smart ads.

Industries: Technology
Location: France