The sales events that mark the start of the holiday season (Black Friday in the U.S. and elsewhere, Buen Fin in Mexico) are pivotal moments for ecommerce merchants. Opportunities for unprecedented sales collide with the challenges of heightened fraudulent activities, and merchants are rightfully preoccupied with ensuring that every component of their online sales flow is functioning at maximum capacity.
Anti-fraud providers in particular are crucial to ensuring a healthy online checkout flow. A poorly calibrated anti-fraud service can stifle holiday sales and cause customer frustration to surge due to inaccurate declines, or it can open the floodgates to chargebacks by allowing fraudsters to obscure their behavior among the increased sales activity. Even worse, an insufficiently provisioned system might not be able to withstand the increased order volume, which could result in blocked orders across thousands of merchants.
How to prepare for Black Friday
At Signifyd, we recognize the importance of Black Friday and the entire holiday season for merchants, as well as our own vital role in ensuring success for our customers. That’s why we’ve enacted numerous, painstaking measures to fortify our capacity and decision accuracy. We’d like to share some details about the preparations undertaken to ensure that our system excels during Black Friday and the holiday season.
Our preparations are focused on two main axes:
- Accuracy: Maximizing approvals on good orders and rejecting as many fraudulent orders as possible
- Availability: Ensuring our system is able to provide anti-fraud evaluations in a timely and reliable manner, regardless of merchant order volume
Upholding our bar for excellence on either front is no trivial task. As far as decision accuracy is concerned, promotions like Black Friday pose an inherent challenge for machine learning anti-fraud solutions because customer behavior during these periods is so aberrant from the norm. The models that underpin our anti-fraud service assess order characteristics to determine their statistical similarity to either valid orders or those generating chargebacks. However, Black Friday shopping patterns introduce complexities as legitimate customer purchases deviate significantly from typical patterns, with orders appearing less conventional due to unusual timing and atypical discounts. Furthermore, many characteristics align more with fraudulent behavior, such as a high volume of orders placed in a short period by users with minimal or no order history with a merchant.
More than a decade of fraud protection experience gives Signifyd an edge
Signifyd employs a variety of mechanisms to overcome these challenges. First, we have an inherent advantage as a long-standing anti-fraud provider who’s weathered more than a decade of Black Friday events. Our proprietary training datasets contain many years of promotional events whose output is incorporated into our machine learning models’ decisions. As a result, our models “understand” to a certain extent how to interpret Black Friday abnormalities due to prior years’ behavior.
But furthermore, during the holiday period, our Risk Intelligence and Data Science teams are working double-time to forecast and simulate purchase behavior and tune our risk thresholds accordingly to ensure optimal performance. This involves assessing historical data as well as incorporating merchant-specific context into our decision mechanisms, which requires close coordination with our customer-adjacent Implementations and Customer Success teams.
Finally, during Black Friday itself, our Customer Success teams closely monitor live merchant results and coordinate with Risk Intelligence and Data Science to make any fine-grained adjustments to model thresholds and rules if additional tuning is needed. Ultimately, between our product’s capabilities that support precision performance tuning in real-time and our preparatory risk calibration and model training, Signifyd enters the holiday rush with confidence in its ability to provide highly accurate decisions that delight customers and frustrate fraudsters.
Holiday shopping volumes call for pressure-tested operations
Yet providing accurate decisions is only half of the battle. We also need to ensure that we can provide hundreds of decisions per second, without compromising on accuracy or system availability. During Black Friday, the surge in online orders requires a robust system to prevent any degradation in our decision latency or availability during peak traffic periods. Fortunately, our Engineering team is at the forefront of tackling this challenge, implementing rigorous load testing and proactive measures to maintain our operational status even in the face of heightened demand.
One of the main challenges of the Black Friday shopping event is the immediacy of the increase in order volume. Many promotions start at midnight, and order volumes can balloon from zero to thousands of orders from one minute to another. Although Signifyd’s cloud-based infrastructure is highly scalable, no system can scale quickly enough to avoid any degradation of service with a purely reactive scaling approach.
As such, Signifyd preemptively scales its infrastructure prior to the holiday period to ensure sufficient capacity for any merchant needs. Then, rigorous load tests are conducted, simulating the anticipated surge in user activity to identify and address any potential bottlenecks. These bottlenecks are identified and remediated, and the load test is repeated until we achieve a satisfactory level of performance. This year, we were content after testing our system at roughly 2.5 times the forecasted Black Friday peak load — ensuring a healthy margin of additional capacity in case of any unforeseen additional traffic.
Code freeze ensures smooth operations during retailers’ most critical time
Finally, in an effort to limit the variables that might alter our system’s behavior during this critical merchant period, Signifyd imposes a “code freeze” on a number of our production services in the weeks leading up to Black Friday. Our system is locked into place to minimize the risk of any anomalous behavior during this pivotal phase for our customers, although certain merchant-critical changes and adjustments with well-scoped, limited impacts (such as risk threshold adjustments) are permitted. That said, code freeze doesn’t mean that development and productivity comes to a halt for Signifyd’s developers; changes are constantly being made on upcoming features and improvements, but they are not deployed to our production environment until after the Black Friday surge has passed.
With the holiday season upon us, Signifyd strives to provide merchants with peace of mind knowing that their order flow is supported and secured. Our years of experience and months of dedicated preparation ensure that we are a reliable ally in the complex ecommerce landscape. We look forward to supporting our customers during the most high-stakes online sales period of the year, and we are proud of the service that our teams come together to provide.
Photo by Getty Images
Looking for pressure-tested fraud protection for the holiday season and beyond? Let’s talk.