What were you doing when you heard of Michael Jackson’s death?
2019 June marks the 10th death anniversary of Michael Jackson. Yes, time flies! Who is not a fan of his energy, music, and moves? He was a generational icon that most people remember where they were and what they were doing when they heard the news of his death. Well, what was I doing? Our team was trying to find out why the sales suddenly spiked on store.michaeljackson.com. I was employed at Amazon.com those days managing thousands of 3rd party businesses on webstore by Amazon. The eCommerce store of michaeljackson.com was one of those online businesses that we managed. It was a sad, almost perverse reality that within seconds of the news of his death the sales graph took off almost vertically and stayed at the top of the charts for months. We were alerted immediately on our phones about this. Our first thoughts were that there was a DOS attack on our server farm. But when we noticed it was just this one site that was spiking and after checking out the news, we knew it was real. That is how I learned of his death. Those days it was an interesting exercise to analyze and correlate bumps or dips in sales to mentions of the product or leadership appearances on popular shows like Oprah or Ellen. We could feel the marketing power of those shows acting as infomercials as we saw the sales move up or down in real-time even as the show progressed. We used to ask for advance notice of any such planned marketing events, sometimes we got those, sometimes we did not and had to sleuth it out. It was fun nevertheless.
Proactive alerts and predictive analytics
One of the key reasons for Amazon’s success is this continuous obsession with the customer and the urge to stay on top of whatever was happening on the website. There was a culture of continuously measuring and alerting any variances to what was expected at that time of the day, day of the week, time of the year. This was considered an absolute necessity because each second the web site is down or impaired, it meant not only revenue losses but also a huge hit to the brand that is the poster child of online commerce. This meant collecting analytics and trends to predict a range of where the sales volume should be at any given point. This was something that was done pretty effectively at Amazon in 2009 itself even when the industry did not hear the buzzwords analytics and proactive alerts. How should your business set this up? The raw inputs to the model are the sales trends (GMS – Gross Merchandise Sales and order volumes) by the time of the day. From these the model has to deduce the expected volume for the next 24-48 hours, adjusting for seasonality, annual growth, and catalog growth. While this forms the basis for Holiday preparedness, there are several other complexities so holiday readiness is a topic for another post.
This predicted sales volume has to be projected as a range around the actual value derived from the model described above. There are several types of models such as ARIMA (Auto-Regressive Integrated Moving Average) or LSTMs (Long Short-Term Memory networks). These models have to be refined for your business, based on the specific parameters and has to be periodically tweaked based on the feedback from the actual sales to limit the error rate of future predictions.
Proactive alerting needs to be in place to ensure any deviance of the actual sales graph from the projected threshold range is immediately notified. In progressive businesses, On-call support was put in place to investigate these deviances by an operations team. This operations team also puts together these metrics for a review each week to ensure the sales are on track for the weekly, monthly and annual targets as compared with the MOM YOY trends. Such a review by the leadership consisting of the Business, marketing and IT facilitates corrective action in a timely manner so business goals are exceeded or met. The latest wave, however, is AI Ops (or applying AI to technical Operations). Vendors like Dynatrace (with Davis) and Newrelic (with SignifAI) are promising to do root cause analysis by analyzing all the signals to locate the offending line of code. Cloud natives are already being offered similar services by Google, AWS, and Azure. However what truly excites me is the potential when this AI-Ops goes beyond technical ops to business ops, business leaders will have root cause analysis with the ability to drill down to the signal that is bumping or pulling down revenue in real-time. Coming back to the MJ story, if such an AI (business) Ops was in place 10 years ago, we would not have to manually do our due diligence and instead just received an alert that your business has increased by x% because of the news article on MJ’s death!
tvprasad 