By Manan Vyas
The beauty of data science is best understood through an example on how it helps real businesses. The Indian Economist teamed up with the data science team at Atidiv to give our readers an exclusive inside view on a data science project that helped rescue a failing company.
And thus on a sunny Monday morning, I decide to pay a visit to Atidiv’s state of the art office in Pune. Rahul Madhavan is a senior data scientist in the team, and he has promised to take me through the process they followed while diagnosing and fixing a company through data science.
The office hums with activity. The open office floor plan reveals scores of young men and women intently at work on the rows of computers that occupy the tables. The crowd is mostly young. The energy in the office palpable. Music blares on a loudspeaker in the corner of the room. Atidiv describes itself as a full stack data science company. This means that they help companies collect data, they engineer data platforms to store this data and they employ data scientists to analyse this data. The office I am in has a diverse crowd. From entry level data entry operators to the software engineers to the elite data scientists – this office has them all.
I am introduced to Rahul Madhavan, an enthusiastic data scientist who has had stints at global financial companies including Barclays and Nomura. He will be my guide for the day. He leads me to his desk, which is lined with an intimidating array of monitors. “For visualisations” he explains, seeing the somewhat bewildered look on my face.
“So tell me something about the company you worked on”, I begin.
(Note: Client names and statistics have been changed to maintain client confidentiality)
Rahul begins, “The client company, let’s call it Great Toys, is based in the United States. It operates in the subscription e-Commerce business category. This is a business in which customers pay a fixed amount each month to get a set of predefined products delivered to their house. From razor blades to clothes to toys, the subscription ecommerce category delivers a wide range of products to customers. Great Toys delivers action figures and memorabilia. The assured, recurring revenue stream from a customer makes this a potentially lucrative business.”
“The business model looks great. What went wrong?” I asked.
Rahul pulled up a chart on one of his large monitors.
“This chart below shows the company’s performance in 2016. The yellow line represents the revenue while the blue bars represent the costs” he said.
“Costs appear to exceed revenue by a rather substantial margin” I remarked.
“Absolutely. At this rate, the company will shut down in less than a year” he said.
“So did do you do next?” I asked.
“The key to bringing out effective data science insights” he explained, “Is to know which questions to ask. In a world with unlimited data, it is easy to get lost. A good data scientist approaches the problem by asking the right questions for that context” he said.
“So what are the questions that you began with?” I ask.
“In any ecommerce business in general, and in a subscription ecommerce business in particular, there are two drivers that drive your growth and profitability – Customer Lifetime Value (LTV) and the Customer Acquisition Cost (CAC). If the Lifetime Value of your Customers is greater than the Customer Acquistion Cost, you have a potentially profitable business, because your customers are adding greater value than they cost to acquire” Rahul explains.
“So what was the situation at Great Toys like?” I ask.
Rahul deftly pulls up another chart. The situation was evident. The company’s customer acquisition cost far exceeded the value it derived from its customers.
“We calculated that the company was losing 60 cents on the dollar for each customer” Rahul explained. “No company can sustain with those numbers”.
“The chart makes the situation fairly clear. Where did you go from here?” I asked curiously.
“The next step”, Rahul continues, “Is to dig deeper into each driver. We break down the factors that constitute the Customer Acquisition Cost as well as the Life Time Value and we look for insights in each”.
“So where do we begin?” I asked.
“We’ll begin by deep diving into Customer Acquisition Cost” said Rahul. “Here is a framework that will help you understand the approach better” he said, pulling up a chart. “These are the three constituents of Customer Acquisition Cost we are going to tackle”.
“The advantage of a digital business is the ability to target your spending to the demographic that is most useful to your business. So that is what we looked at” Rahul said.
“And what did you find?” I asked.
“We found that men buy significantly more from the business as opposed to women without costing significantly more to acquire. We recommended a more focused targeting on men for future advertising” Rahul explained.
“When it comes to analysing geographies”, he continued, “Nothing works as well as a map. Check the map below. The colour indicates profitability and the size of the ball indicates the number of customers. I have left this image unmarked for you to take a shot. What can you make out from the chart?” he asked me.
The chart seemed clear enough at a glance. “There are some markets that are driving losses. Denver and Boston for example” I said. “And some markets that seem to be profitable, including Indianapolis and Charlotte” I continued.
“That is correct” he confirmed. “We recommended increasing advertising investment in high performing markets, while exiting ad spend in low performing markets. This helps us attract profitable customers” he said.
We them moved on to analysis of customer groups based on the discounts they received.
“We looked at the discount rates at which customers were being acquired and divided customers on that basis. We then analysed the months to cancellation for those customers, indicating their value to the company. We found that more than 70% of all customers entered through discounts of between 50% to 100% but cancelled in less than a month. These customers added no value to the business. On the other hand, only 9% customers entered through no discounts or low discounts. These customers stuck around for longer” Rahul explained.
“Which means that the company is losing money with its heavy discounting strategy” I summarised.
“That is correct. We therefore recommended the company to move away from heavy discounting as a customer acquisition strategy and deploy those funds in other marketing areas” said Rahul.
We then moved on to analysing the performance of different kinds of coupons. Rahul pulled up this chart.
“There were certain coupons that were bringing in undesirable customers that had a very high cancellation rate. The key is to isolate and remove these coupons from the system” he explained.
The concluded the Customer Acquisition Cost aspect of the project. Advertising became more targeted – men were targeted in cities that yielded profitable customers, heavy discounting as a strategy was scrapped and certain coupons were phased out.
“We now turn our focus to increasing the lifetime value of a customer. It is not enough to simply reduce costs – we also need to increase the value we extract from each customer” he said.
“So what exactly is Lifetime Value? How is it calculated?” I asked.
“In a subscription business, a company earns a defined amount from a customer for a month. A company also loses a certain percentage of customers each month in a process known as churn. Dividing the customer’s monthly value by the churn rate provides the lifetime value. For example, if the monthly value is $9 and the churn rate is 25%, the lifetime value is $9/0.25, which equals $36.” Rahul explained.
“So it is critical to reduce the churn rate” I ventured a guess.
“Yes. 1 minus Retention Rate equals the Churn Rate. Therefore, to increase a customer’s value, you need to improve the customer retention. Which brings us to the problem facing Great Toys. Check this chart” he said, pulling out yet another chart.
“The numbers on the bars represent the percentage of customers still remaining in December. The months at the bottom represent the month in which they were acquired. For instance, only 65% of customers acquired in November remain customers in December, indicating a churn of 35% in the first month itself. By the end of the 3 month mark, represented by September, more than 64% of customers have churned out. It takes 6 months for a customer to become profitable. Customers acquired in May start yielding profits in December. However, only 18% of May’s acquired customers now remain. 82% of other customers have churned out at a loss. This indicates that retention is as large an issue as high customer acquisition cost” Rahul explained.
“So retention is the focus area. That is fairly clear. But what steps can we take?” I asked.
“The numbers you see represent overall retention rates. Just as we saw in our CAC analysis, the numbers start looking different once we slice and dice them. Isolating the problem areas can help the company take the right decisions on areas to increase and decrease spend” Rahul explained.
The chart he pulled up next indicated that there product types were loss making. Unfortunately, these three product types also constituted the largest part of the business.
“The company has certain products that are striking a chord, and retaining customers customers for more than 3 months. However, its 3 largest products are not doing well. In terms of a future investment, we recommended the company to look at moving out of loss making product types identified, and move towards the investing more in marketing the profitable products identified” Rahul explained.
We then moved to the final bit of analysis for the day, from the summarised view of the project that Rahul was giving me. “Customers with default behaviour are the ones with the weakest intent” he explained. “We looked at the behaviour of customers on the basis of whether they changed the default option provided in the buying process. Look at this chart for example” Rahul said, pulling up a chart with this analysis.
“Even cheaper customers are more engaged?” I asked.
“Yes. All customers that changed the default option are more engaged than others. So our recommendation involved removing a default option an encouraging customers to engage more in the buying process” Rahul completed.
“That sounds fascinating” I remarked. “Did we get a complete picture of the process?” I asked.
“There is a lot of work that goes behind the scenes in terms of bringing the data to a stage where it can be easily analysed. In this instance, we looked only at the analysis aspect of the case. In the past, Atidiv has helped companies collect data, process it for use and then analysed it. Once you know the set of questions you want to investigate, the process becomes smoother” Rahul explained.
This had been quite the morning for me. Here, in this room, Rahul Madhavan and his whiz team of data scientists dig deep into reams of data to pull out insights that seem elegant and simple at the same time, but mask the complexity of pulling it all together. Around me, the buzz of activity continued, as the Pune outpost of the global data science company continued to generate business changing insights for companies across the globe.
Note: This was a sponsored post created in collaboration with Atidiv.
To learn more about Atidiv’s work, check out their website: https://www.atidiv.com/
Stay updated with all the insights.
Navigate news, 1 email day.
Subscribe to Qrius
