Another holiday season will soon be upon us, and many retailers and eCommerce businesses are already making plans. As you take inventory of what you learned last holiday season, let’s start with some lessons learned by the entire retail industry this time last year. In addition to stocking up on hot items and planning your promotions, the most competitive sites found that using AI/ML to optimize customer experience not only kept customers happy, it dramatically increased their revenues.

Small and slow leaks sink ships – by analogy, slow and small leaks can also cause significant losses for any business if not detected and fixed early. Are small leaks interesting? Suppose an eCommerce business sees a decline of 50% of purchases in the last day – the entire company would be called in – from the CEO all the way to R&D, Support, to figure why it happened as quickly as possible.

In our previous article (How to Scale and Manage Millions of Metrics), we looked at correlations in terms of name similarity, but there are other types of similarities that occur between metrics.

What’s surprising to see today is how business operations struggle to get an integrated view of all business metrics. With greater volumes of data being collected, data analysts just can’t keep up with the pace. This state of affairs alone doesn’t hit as hard as the fact that many in data analytics have just come to accept this situation as a norm and simply bear with this daily struggle.

With businesses collecting millions of metrics, let’s look at how they can efficiently scale and deal with these amounts. As covered in the previous article (A Spike in Sales Is Not Always Good News), analyzing millions of metrics for changes may result in alert storms, notifying users about EVERY change, not just the most significant ones. To bring order to this situation, Anodot groups correlated anomalies together, in a unified alert.

As product managers, you’re ultimately the one held responsible for the entire product. So the last thing you want to assume is that someone else has got monitoring and alerts covered. In the first days of a release, all eyes are on the new product or latest feature. Just a few months later, when you introduce a brand new feature, the old one might break in the process. At times like these, you want to be ahead of your users, and not hear from your users that something isn’t working.

Why could a spike itself not always be good news? Why is it so important to find the relationships between time series metrics at scale?