As a data-scientist, you must be a good story teller
Let me tell you a story… :)
The Devastation
The aftermath of the 2008 crash claimed millions of casualties big and small. Whole countries in Europe went into a depression and are still struggling today to come back to normal (Greece, Spain). Millions of people lost their jobs and saw their retirement accounts shrink in half or disappear. In the US, many banks disappeared forever and the bigger ones only survived because of the US government’s charity (or the taxpayers’ depending on how you look at it.)
Our protagonists in this story are the small business owners in the US who have not been typically highlighted as victims. They were caught in the perfect storm. Customers were spending less, leading to lower sales & revenue, collaterals were weakened and banks on the whole became risk averse, lending less money to small businesses. With all these problems, our protagonists do not have any access to working capital, credits or loans -necessary to sustain or grow their businesses. Small businesses as a group were more susceptible to the economic swings and the tightening of regulations, and community banks going under were not helping matters any. Why Small-Business Lending Has Not Recovered
Yet, small businesses are the very engine of the US economy. “They produced 46 percent of the private non-farm GDP in 2008. Small firms accounted for 63 percent of the net new jobs created between 1993 and mid-2013.” (Ref: Small Business & Entrepreneurship Council) Helping our protagonists thrive again was at the very center of helping the US economy recover from this point of no-return.
The Opportunity
Rob and Kathyrn wanted to help. But there was one big problem. They were not banks. Besides, these storied institutions that had been present for centuries were finding it hard to lend to small businesses. What hopes did they have? Peeling off the layers to the problem presented a glimmer of hope. Banks had a very labor intensive and non-scalable underwriting process. A business that needed a loan would have to walk into the bank, with a tonne of paperwork in tow. Income tax statements and filing paperwork, company incorporation documents, sales, revenue and account books, credit records - to name a few. An underwriter in the bank would labor over all the data, make some background checks and calls to perform due diligence and generally try to estimate the ‘credit worthiness’ of the business. All of this could take weeks if not months. Worse, the bank might come back with a ‘no’ at the end of the due diligence. Onward to the next bank to try again. Then, the Aha moment! What if this whole process could be automated? What if the human underwriting element was removed and the decision making process codified using sophisticated machine learning algorithms? What if the customer’s data including business banking accounts, credit information, and transaction channels like Paypal/Stripe could be pulled using API directly from the providers with just some simple authentication by the business thereby doing away with cumbersome paperwork? Intuit’s Mint was doing something similar to pull customer data using API and presenting a holistic view of the customer’s finances. Rob and Kathryn knew that they were on to something. Kabbage Inc. was incorporated in 2009 with this idea. In 2011, the launch of the Kabbage platform seamlessly pulled online sellers’ data to assess the health of their businesses. Thanks to an automated, online process, users were qualified for loans instantly, with the entire process of applying and qualifying completed in fewer than seven minutes.
The Aftermath
Fast-forward to the present. In 2016 alone, Kabbage lent out more than $1.2 billion dollars in loan principals to small and medium sized businesses. Kabbage is currently Series E funded and is considered a ‘Unicorn’ with a billion plus valuation. Ref: Kabbage | crunchbase. It is considered a leader in the small business alternate lending space. It has partnered successfully with International banks like ING, Santander and Scotiabank to offer loans in other countries besides the US using the same automated process. It is also considered one of the top places to work in 2016. Ref: Working at Kabbage But, most of all, it is helping solve a key problem for our protagonists. Providing fast and easy access to working capital when they need it the most to thrive, succeed and grow -thus helping turn the wheels of the US economy. This is a story that used data and machine learning at its core to solve a very real problem. I hope I had your attention. The goal was not so much to extol the virtues of the company I work for, but to drive home the point about how a key business objective/problem was identified and a solution was identified and solved using data and machine learning at its core. Told via a story.
Why is story telling important to data science?
I would argue that story telling is always a great skill to have irrespective of your job description. Like others have already pointed out, the human mind has evolved over time and is attracted to good stories, while not very tuned to statistics, facts or data. The audience must thus be drawn in with a good story. And, for a data scientist, this is seldom a random audience. The audience are typically the key decision makers in the company. The CEO who calls the shots, the CRO who controls the budget in the company, the CSO who sets the strategy for the company. If you want them to go bat for you, you better sell them a story that they can understand and solve a problem that they truly care about. They rarely care for the inner details of how the solution works or is implemented. What fancy algorithm or big data architecture you used to solve the problem is not their concern. The results, the impact to the company bottom line and top line does however matter to them. Speak their language. An solution that is not ultimately implemented is a solution that did not happen.
How should I tell a data science story?
The story should first and foremost set the motivation or the business problem that is being tackled. The story should next entail the data that was gathered, the insights and the key experiments that were setup to try to solve the problem. Finally, present the business metrics that were impacted to quantify the benefit of the solution. Metrics like AUC or F1-score matter little to the business person. Lift to sales, customer lifetime value improvement, increased profits, improvement in customer satisfaction, product usage up tick, increased click through rates, reduced customer acquisition costs - these are the metrics that finally matter to business people. Spin all this into a tight and concise narrative with graphs and visuals to aid the story. Practice makes perfect. All the best!
The Devastation
The Opportunity
Rob and Kathyrn wanted to help. But there was one big problem. They were not banks. Besides, these storied institutions that had been present for centuries were finding it hard to lend to small businesses. What hopes did they have? Peeling off the layers to the problem presented a glimmer of hope. Banks had a very labor intensive and non-scalable underwriting process. A business that needed a loan would have to walk into the bank, with a tonne of paperwork in tow. Income tax statements and filing paperwork, company incorporation documents, sales, revenue and account books, credit records - to name a few. An underwriter in the bank would labor over all the data, make some background checks and calls to perform due diligence and generally try to estimate the ‘credit worthiness’ of the business. All of this could take weeks if not months. Worse, the bank might come back with a ‘no’ at the end of the due diligence. Onward to the next bank to try again. Then, the Aha moment! What if this whole process could be automated? What if the human underwriting element was removed and the decision making process codified using sophisticated machine learning algorithms? What if the customer’s data including business banking accounts, credit information, and transaction channels like Paypal/Stripe could be pulled using API directly from the providers with just some simple authentication by the business thereby doing away with cumbersome paperwork? Intuit’s Mint was doing something similar to pull customer data using API and presenting a holistic view of the customer’s finances. Rob and Kathryn knew that they were on to something. Kabbage Inc. was incorporated in 2009 with this idea. In 2011, the launch of the Kabbage platform seamlessly pulled online sellers’ data to assess the health of their businesses. Thanks to an automated, online process, users were qualified for loans instantly, with the entire process of applying and qualifying completed in fewer than seven minutes.
The Aftermath
Fast-forward to the present. In 2016 alone, Kabbage lent out more than $1.2 billion dollars in loan principals to small and medium sized businesses. Kabbage is currently Series E funded and is considered a ‘Unicorn’ with a billion plus valuation. Ref: Kabbage | crunchbase. It is considered a leader in the small business alternate lending space. It has partnered successfully with International banks like ING, Santander and Scotiabank to offer loans in other countries besides the US using the same automated process. It is also considered one of the top places to work in 2016. Ref: Working at Kabbage But, most of all, it is helping solve a key problem for our protagonists. Providing fast and easy access to working capital when they need it the most to thrive, succeed and grow -thus helping turn the wheels of the US economy. This is a story that used data and machine learning at its core to solve a very real problem. I hope I had your attention. The goal was not so much to extol the virtues of the company I work for, but to drive home the point about how a key business objective/problem was identified and a solution was identified and solved using data and machine learning at its core. Told via a story.
Why is story telling important to data science?
I would argue that story telling is always a great skill to have irrespective of your job description. Like others have already pointed out, the human mind has evolved over time and is attracted to good stories, while not very tuned to statistics, facts or data. The audience must thus be drawn in with a good story. And, for a data scientist, this is seldom a random audience. The audience are typically the key decision makers in the company. The CEO who calls the shots, the CRO who controls the budget in the company, the CSO who sets the strategy for the company. If you want them to go bat for you, you better sell them a story that they can understand and solve a problem that they truly care about. They rarely care for the inner details of how the solution works or is implemented. What fancy algorithm or big data architecture you used to solve the problem is not their concern. The results, the impact to the company bottom line and top line does however matter to them. Speak their language. An solution that is not ultimately implemented is a solution that did not happen.
How should I tell a data science story?
The story should first and foremost set the motivation or the business problem that is being tackled. The story should next entail the data that was gathered, the insights and the key experiments that were setup to try to solve the problem. Finally, present the business metrics that were impacted to quantify the benefit of the solution. Metrics like AUC or F1-score matter little to the business person. Lift to sales, customer lifetime value improvement, increased profits, improvement in customer satisfaction, product usage up tick, increased click through rates, reduced customer acquisition costs - these are the metrics that finally matter to business people. Spin all this into a tight and concise narrative with graphs and visuals to aid the story. Practice makes perfect. All the best!
Comments
Post a Comment