Real time anomaly detection is critical for eCommerce companies first and foremost due to the fact that it can detect and alert on costly issues in the time series data of business metrics as they’re happening, long before the effects of lost revenue, negative social media shares, embarrassing news articles, or calls from upset customers start to mount.
The metrics most commonly tracked by eCommerce companies tend to include aspects such as purchases, page views or unique website visitors, failed payment transactions, or abandoned sales carts. Each of these is usually broken down by product category, geographical region and the device, operating system or app used at the time. These are the metrics which are most closely tied to revenue for eCommerce businesses and thus directly impact the bottom line.
Real-time anomaly detection is a significant part of a business’s success, and there are a number reasons for that:
Reason #1: Complex business big data requires advanced anomaly detection
Real time anomaly detection requires sophisticated machine learning algorithms which not only instantly and accurately identify anomalies, but are also smart enough to adjust to long term shifts and other changes in those metrics, because in eCommerce, nothing is ever static: products are constantly added and removed, discounts and rebates from manufacturers and retailers start and end, competitors attempt to outmaneuver each other with ad buys and social media marketing campaigns, and so on.
With all these moving pieces, the last thing your business needs is to be overwhelmed by alert storms or burdened with setting and then constantly adjusting static thresholds, but traditional business intelligence (BI) tools which rely on manual anomaly detection leave you no other choice.
Reason #2: Real-time anomaly detection helps eCommerce companies scale up
Real-time automated anomaly detection helps eCommerce businesses scale up by allowing them to avoid the delays, errors and costs of manual anomaly detection via traditional BI tools. In the fast-paced world of web-enabled commerce, detection delay is perhaps the worst drawback of manual anomaly detection since hundreds of customers can take advantage of a price glitch in a matter of hours. The faster a problem is detected, the faster it can be fixed, resulting in less money lost. Less money lost means higher profits, which then can be reinvested in the company, fueling growth.
Reason #3 Manual anomaly detection is neither fast nor scalable
As web development platform company Wix discovered, relying on manual anomaly detection requires prioritizing, since that approach can never scale to thousands or millions of metrics. This forced its analytics team to make, at best, educated guesses about which metrics deserved their limited resources. Critical evidence of important business incidents, however can and do occur as anomalies in seemingly insignificant metrics – and not just in the handful which Wix was actively monitoring – which is why you need to monitor them all. The needle can be in any part of the data haystack.
Even with a skilled team of data analysts scrutinizing a few metrics, Wix’s manual anomaly detection approach still couldn’t catch anomalies in real time. Delays between anomaly occurrence and anomaly detection ranged from hours to days, a lag with potentially catastrophic consequences in a price glitch or payment processor API breakage scenario.
Now that they’ve begun using Anodot’s platform, hours and days of detection delay collapse to minutes, and the entire organization, not only the BI teams, have a unified platform for discovering important incidents everywhere from sales to R&D.
Right for eCommerce big data
Anodot’s real time anomaly detection system is a must-have for eCommerce retailer success simply because eCommerce occurs in a very fast-paced environment – the Web – where delays can be outright disastrous. Our real time anomaly detection system will find the critical signals in your data by monitoring all your metrics – without astronomical headcount or unacceptable delay.