Online Fraudulent Account Detection with SendinBlue

Efficiently detecting and preventing fraudulent operations are table stakes for online businesses today, and they increasingly require AI and data-driven solutions.

Any online business is exposed to fraud risks by its very nature, from bots to phishing, scams to wire transfer frauds, and everything in between. The huge array of risks – not to mention the potential consequences of these risks – demands human intervention. But hiring more staff to conduct manual audits is expensive and often inefficient. What’s more, for online businesses used to efficiency and immediate delivery, manual audits considerably delay and decrease the user experience, damaging the site’s reputation.

Therefore, the key to decreasing fraud risk is leveraging AI and machine learning for automatic audits to scale with the growing demand. This enables audit teams to focus primarily on ambiguous cases that actually require a review.

Launched in 2012, SendinBlue was created to become the simplest, most reliable and cost-effective marketing platform. This all-in-one solution now powers marketing campaigns for more than 50,000 companies around the world. With offices in North America, Europe, and Asia, the SendinBlue team supports the product in six languages. Their platform integrates with the top e-commerce and CMS tools, and their system delivers over 30 million emails and text messages per day.

Image result for sendinblue

Challenge 

SendinBlue is a relationship marketing SaaS solution that allows customers to execute email and SMS campaigns. Because the core of its service is the successful sending of customers’ messages, SendinBlue must maintain exceptional delivery scores and therefore cannot allow the use of stolen, false, or purchased email databases that would decrease its email reputation. So when customers upload databases, SendinBlue must ensure all contacts on the list are opted in. To be considered as opted- in, these contacts must have been in touch less than two years ago. They must be able to unsubscribe and can only receive emails related to their initial queries.

SendinBlue used to rely on manual validation of new customers, checking (among other things) the quality of their databases, but it was a painful and long process that required a large workforce. It also severely delayed account validation for customers and was therefore detrimental to SendinBlue’s reputation. Faced with a growing customer base, SendinBlue could not continue down the path of manual validation and feasibly scale.

To move away from manual validation and allow for rapid growth, SendinBlue turned to Dataiku, leveraging their expertise in the area of automated fraud detection. With Dataiku, SendinBlue built a scalable solution using historical data coming from more than 1 billion emails and associated events (clicks, opening rates, bounces, etc.), thousands of blocked accounts, and hundreds of fraud criteria (IP addresses, history, behaviors, etc).

[embedded video: Anomaly Detection 5 steps – Lumen5 pilot video]

SendinBlue uses Dataiku to: 

  • Analyze new customers and automatically classify them as “good,” “bad,” or “uncertain.”
  • Once the new customers are classified, an algorithm determines the customer’s credibility by taking into account sending volume, the scoring of the contacts, etc.
  • Depending on the customer’s risk score, they may be blocked, validated, or sent to customer care for manual analysis.

Throughout the process, Dataiku proved paramount in deploying the data product into production. SendinBlue chose Dataiku as their preferred solution because of its flexibility to easily handle huge amounts of different datasets and its ability to design, test, and develop a solution in less than three months.

Thanks to Dataiku, SendinBlue teams have been able to:

  • Scale their team and customer growth by serving customers more quickly.
  • Save time and rationalize the global validation process.
  • Provide a better customer experience by considerably shortening validation delays.
  • Handle most of the new accounts through the use of a machine learning model.
  • Be more accurate in their validation, thereby increasing their email reputation; the first iteration of their model leverages Random Forest and has an 83 percent precision rate on email address classification.

Dataiku helped us to massively scale our team’s productivity while delivering a better user experience. We are now going to leverage the power of machine learning to aid our customers in real time while they use SendinBlue. As the fraudsters’ systems always evolve, we’ll soon leverage a new generation of advanced algorithms to detect new kinds of fraud.

Armand Thiberge-Sendinblue
CEO| SendinBlue

BGL BNP Paribas: Improving Fraud Detection

See how BGL BNP Paribas was able to improve fraud detection and democratize the use of data across the organization while maintaining their high standards for security and data governance.

Read more

Go Further

Santéclair: Detecting Fraudulent Claims More Effectively

See how one health insurance company was able to implement a machine learning-based fraud detection system that is 3x more effective.

Learn More

Marlette Funding, Best Egg Loans: Fraud Detection with ML

Marlette Funding, Best Egg Loans improved their fraud detection capacities by 10 percent by switching to a machine learning-based model.

Learn More

Anomaly Detection for the AI-Driven Business

Across industries and use cases, there is perhaps no other data science strategy more important to leverage than anomaly detection.

Learn More

Bringing AI to Marketing

The evolution of AI, machine learning, and data science have been an increasingly integral part of the transformation of many industries, and marketing is no exception.

Learn More