The last couple of months have been disruptive for all of us, to say the least. That demonetisation-driven turbulence has travelled into the online world as well. Digital advertising, online traffic, social media interests—nothing is normal anywhere anymore.
Based on what we are seeing at programmatic ad exchanges through Cartoq.com, online campaigns have reduced (by almost 50%, according to our estimates). User traffic patterns show inexplicable peaks and dips. Online traffic spikes at night usually, but that’s not always the case these days.
What’s driving social media behaviour today is temporarily unfathomable. We are seeing periods of stronger-than-ever engagement, followed immediately by intervals of utter user indifference.
Times are weird, and the environment is very unpredictable. It’s easy to succumb to panic-driven counter reactions.
There is this memorable moment in the movie Gladiator, where Russell Crowe and a rag-tag bunch of slaves are thrown into the Colosseum. They are about to fight the well-trained gladiators of the mighty Roman army. In front of frenzied masses baying for their blood, Crowe’s team are all first-timers. In the middle of this giant killing arena, they do not know which side the attack will come from.
Just at that moment, before all hell is about to break loose, Crowe roars out to his compatriots: “Hold the line!” He wants them to stay in their formation. And as his valiant crew gradually prevails upon the violent mayhem around them, Crowe keeps bellowing out this advice: “Hold the line!”
That’s an apt lesson for our turbulent times—we need to hold the line as well. And that means digging into our past data, knowing intimately the underlying business trajectory we were on, and then figuring out how to get back to that path.
To do that, we need to leverage our data for a slightly different purpose.
Predictive analytics to the rescue
In simple terms, predictive analytics is about using your current and past data to make predictions about the near future. To do this, several sophisticated approaches are available, from simple data mining and statistical analysis to modelling and machine learning.
I am going to focus on what’s rudimentary and easily accessible to most of us.
To begin with, we need to know if the sudden change in our environment is really disruptive for our business or if it is a temporary flux. Unless you are a Paytm (for whom this disruption is fundamental and highly desirable), for most businesses, the current demonetisation entails a temporary cash crunch and is not likely to impact the trend-lines much.
If that is indeed the case, then two things become very important. One, we need to reacquaint ourselves with our core business metrics—how these have been evolving in the past 6-12 months. We need that crucial reference point to go back to.
And then, we need to juxtapose this data with our current numbers to establish the extent of the variance. From that understanding will emerge more calibrated, and informed responses.
Let’s take Cartoq as an example.
The turbulence in context
Before we got into programmatic selling of ads on our auto media site, we had two fairly independent set of business variables to optimise: traffic and ad sales. With real-time selling of ads, things got a little intertwined.
Our traffic planning (a function of marketing and content publishing), now had to sync strongly with the ebb and flow of ads in ad exchanges. We had to maximise our ad inventory (a function of page views generated) when ad rates were the highest, and the volumes were also large.
So matching ad inventory to page inventory became an inter-dependent set of activities. In real-time. Now on top of this, consider the current environment, where all the data, be it about traffic or advertising, has gone haywire. Where do we start? Should we spend less in marketing? How much of a drop in ad volume should we assume this month?
We had new decisions to make on various fronts.
Predictive analytics in action
We started by drawing up our three-month and six-month trend-lines for all important metrics.
What was our fan acquisition cost on Facebook in the past three months? What has our daily cost been in the past one month (since demonetisation)? How was our key ad price and volume data behaving in the first 10 days of the month for the past six months? What was that data telling us in the first 10 days of December?
We built a series of trend-line numbers for our key metrics, and then we compared these with our most recent data. We established the extent of divergence. This variance from the underlying trajectory provided us with a fresh starting point about what change we have to plan for.
What gave us further confidence in this approach was that the variance levels were similar across metrics.
This told us that we had a fair measure of how much we need to calibrate our responses for December, and possibly further too.
With the new baselines at hand, we have now built new projections for this month and the next month. Predictive analytics has become the basis for all our planning now. It’s going to be a choppy ride for sure, because the environment is still in flux, but without bringing in predictive data, we didn’t have a compass to navigate our way through.
It seems paradoxical but it’s true—past data has become more valuable just when the current data has diverged dramatically from the past.
That’s because disruptions are mostly just that, temporary turbulence. To ride that out, you need to dig hard into your data and leverage predictive analytics even harder to minimise panic responses.