In order for any business to thrive in modern day setting, I believe that rich consumer insights are going to be a crucial element for growth and success. I believe that by having a complete understanding of these insights will help any management team with their decisions as to which direction they want to take for their company. Either to move forward with their current strategy or pivot to avoid a dead end. It is a sign of a management team's maturity once these insights are converted to actionable metrics. 

But how are insights generated? Is it simply looking at raw data and interpret it directly as it is? Or is it taking big data into a different context and looking at it at all possible angle? 

I remember this one discussion I had with one of my clients where I took his big data and correlated the relationship of the organic website traffic to the total number of products he sold on his site. I also look at the impact of selling quality products and how it attracts high quality customers. 

At the beginning he was doubtful that I could find the relationship because his business operates in a very agile environment. 

Like any Algebra problem, this was a case of finding X. Do you still remember your algebra problems way back in high school? Where every problem seeks to find X, let me give you an example:

2X + 7 = X + 18

In this case, X is equal to 11. The process of identifying X is the same process on how to approach a cohort analysis. 

Cohort analysis is a type of longitudinal analysis that was developed to observe a set of behaviour from the same group of people. Longitudinal because it requires you to look at a specific set of data for a long period of time. By observing these people in a certain time frame, you'll be able to see a pattern on their buying behaviour. 

So how do you start with a cohort analysis? Firstly - identify a cohort. A cohort would mean identifying a group of people you will observe for a period of time with the same characteristics. For example, people born in 1984 can become a cohort. As you can see, this kind of cohort will only analyse people who were born on that year and it doesn't count on any other factors. But this doesn't mean a cohort can only be one characteristic at a time. You can go further by grouping people born in 1984 who are male and with a minimum weight of 100kg. This criteria becomes another cohort of analysis. The only thing that you need to be constant with is that these people should be the same all through out during the analysis. You can't change your sample once you've identified them. 

Unlike any other analysis, I believe that doing a cohort analysis will actually give you more potent results. The reason is that you are not just getting information directly from the survey consumers tend to hastily answers, but you are actually observing buying behaviour right in front of you through the data collected over time. And with this, you'll be able to predict with a very high probability the next action the consumer will actually do. 

While it is true that a cohort analysis will give you more potent results, there is also a possibility that your analysis will be flawed if you don't ask the right questions and look at your data the way it is presented to you. What do I mean by this - by looking at information and trends just the way they are, this becomes a Vanity Metric specially if the trend is going to your favour. 

I'll give you an example, you analyse a cohort to see if there's a growing trend of females visiting your store for athletic equipment. Over time, it shows that you have and it grows on a month-on-month growth rate of 2%. You hurriedly printed the report and showed it to your boss. But is that enough? Initially your boss will be excited and if your boss is not the type who reviews data, he will make assumptions that since you have a growing database of female visitors, he then now orders twice the female athletic equipment to be sold in your branch. After 6 months of being displayed in store, the equipment is not moving. So what's wrong?

You probably guess it right, it may not be the reason why there is an increase of female visitors in your store. There could be other data you are not looking into hence the initial metric has given your boss a false reason to buy the additional inventory. 

Hence, vanity metrics poses a great threat to your analysis. 

How often do you hear it from your boss that you need to execute a certain plan just because it worked at a certain branch? And when you did it yourself, did it work? Was it a great success? If no, why do you think it didn't work?

In order to avoid the situation in the previous section, a longitudinal study will help you decide which actions you should take for your business. Do not simply present vanity metrics, go beyond it and ask yourself how your big data are connected to each other. Ask yourself why your vanity metrics is going up? What contributes to the growth? Is the growth related to an increase in purchase? Is the inventory moving? What items are sellable? Is the purchase cycle of a certain commodity decreased hence you need to allocate more resources in item procurement? 

All these questions would help you lead in identifying reasons conveyed through behaviour that will eventually help you make calculated risks. Calculated since every decision is a risk but mitigating loss is what should every business owner or decision maker always strive for because in every business, risk will always be there. 

At the end of the day, I believe that insights are valuable in making calculated risks. It is better to know than not know what you are getting into. No matter how much your gut tells you that you are doing an awesome job - always verify with your acquired data. There's nothing more important than validating your learnings and applying it to your business to move forward and grow as you have envisioned it.