Fostering a Culture of Analytics

Editor's Notes

• #2: Introduction Who is OptionsHouse (bought by E*Trade) Quick history My team and what we do
• #3: These are the 4 things I want you to take away from this talk.
• #4: This concept comes from Malcom Gladwell’s The Tipping Point The Law of Context infers that epidemics are sensitive to the conditions and circumstances of the times and places in which they occur
• #5: Can anyone tell me something interesting about this data?
• #6: History Lesson: Subway system in New York in the late 80s and early 90s was a very dangerous place to be Close your eyes, let me paint a picture for you: You are standing on a dimly lit platform On all sides of you are dark, damp, graffiti-covered walls Train is most likely late due to either a fire or a derailment somewhere on the track When it does come the floor is covered in trash and both the walls and ceiling, top to bottom, are covered in graffiti. There is no heat in the winter and no AC in the summer When you were about to go through a turnstyle, it wouldn’t be uncommon to see someone jump over it instead of paying Felonies were about 20,000 a year Harassment and beggars had caused most traditional riders to stop taking the train It was once called the ‘transit version of Dante’s Inferno’ Then came Bill Brandon and David Gunn By the end of the 90s felonies had fallen 75% Murders fell by two-thirds So how did they do it? Started by cleaning up graffiti. They poured millions into this effort. Never mixed ‘dirty’ cars with ‘clean’ cars Then they brought in the muscle and started arresting people. But who do you think they arrested? The felons? No. The Panhandlers? Not necessarily. They arrested the fare-beaters, the drunks. Arrested individuals for misdemeanors and focused on the smaller infractions - The Power of Context means that by tackling the small things. My making sure there are no ‘Broken Windows’ that give individuals to act in a way not up to your standards, you create an environment which has a very powerful and positive impact on people
• #8: Start by fixing the ‘Broken Windows’ Establish trust in you data. That’s the reason most people don’t use it Case Study: Completed Applications - Multiple Sources You could get 3 different answers depending on who you talked to Led to distrust in the numbers
• #9: Case Study - Standardization Get a standard process in place. Something you don’t need anyone else to do. But once you have a well documented process and you start sharing with people, it will grow organically and you will be seen as an expert
• #11: - Borrowing this from a great talk on TED about The Hidden Influence of Social Networks. But I think it pertains here. And notice first of all the highlighted dots. Compare B to D. B has 4 friends and D has 6. D has more friends than B, obvious. But compare B to A. Both have 4. But the difference here is the A’s friends are all connected to each other, forming a tight group where information con only pass outward by 1 or 2 of the 5 in the group. - Finally, compare nodes C and D: C and D both have six friends. But if we look at it this way we can see that their networks are vastly different. How? How big is C’s Network 2 degrees out? Compared to D? I bring this up because most analytical teams are embedded into different business functions. They become subject matter experts and begin to resemble group A which we will call Marketing. They only talk to themselves and share ideas in that tight nit group and very seldom does information leave and influence a different department. Don’t be afraid to branch out. Work your way into the fabric of the network of your organization. Become a group B, then a D, and then finally a C It’s the little things at first you can do to help. Things I’ve done Attended tech all hands and then followed up with various developers about how it could help analytics. Example here is the ELK stack Started including sales leads as an attribution channel. This little bit of information triggered a forced conversation with another department. We started getting together as a combined unit and going over the data. It helped build rapport with both teams The biggest problem is that people don’t know what they don’t know. So the best thing you can do is naturally interact with them and show them what a benefit this skill set it - If you keep at this, I promise you, you will find success. It not only brings analytics to the forefront and conditions others to think about the numbers, but it also gets you face time and builds those relationships.
• #12: To bring home the point. If you look at patent filings of Tesla employees. The lines connect who worked with the main filer and the size of the dot indicates how many patents. You can clearly see (with some outliers) how those in the center of the network not only filed more patents and were more innovative, but were also vastly more collaborative.
• #13: Need a volunteer who has done some machine learning. Quick, in 30 seconds or less explain to me how a decision tree algorithm can help to predict customer value How about, picture a shabby tree and each branch is a potential indicator of the customer’s value. Well we work our way around the tree removing branches until it looks like another tree we are comparing it too. Overly simplistic? Definitely. But you don’t need them to know anything more than that. What you need is just enough buy in - without turning them off - to get excited about the idea. Example: Give Elliott’s first pitch example.
• #14: Numbers are numbers until they can be used to do amazing things like increase revenue, or know when to interact and establish a relationship with a customer If you don’t buy in to it 100% they won’t at all 3) People get skeptical when you talk about step changes and revolutions. They’ve heard years of failed promises. Instead, sell them on aspirin, which is used to fix pain, and vitamins, which are used to supplement their lives 4) Don’t bother with the individuals who don’t buy in at all. Focus on those who are undecided. Once you get them the atheists will follow out of self-preservation 5) When you have a believer, do everything you can to empower them. Go out of your way to learn about their job and their pain points. Provide them with an extra solution even though they may not have known to ask for it. Case Study: Datawarehouse and how we became a process leader
• #15: Need a volunteer who has done some machine learning. Quick, in 30 seconds or less explain to me how a decision tree algorithm can help to predict customer value How about, picture a shabby tree and each branch is a potential indicator of the customer’s value. Well we work our way around the tree removing branches until it looks like another tree we are comparing it too. Overly simplistic? Definitely. But you don’t need them to know anything more than that. What you need is just enough buy in - without turning them off - to get excited about the idea. Example: Give Elliott’s first pitch example.
• #16: So Remember these 4 Things