Big data vs small data peter zhang
- 1. Big data vs. Small data From a start-up to $100MM EBITDA: the power of data driven strategies Peter Zhang Director of Customer Analytics, Sears Holding Co. Founder & President, Simon Peter LLC
- 2. Background • High risk, below-the-sub-prime industry. • Fragmented market. • A start up in online consumer lending. • An efficient on-line platform, a strong database team, and, a lot of data. • An innovative founder/leader.
- 3. Old Strategies • Purchase the leads (loan applications) from leads providers. – costs were high • Organic traffic (loan applications online) from company’s web page. – traffic was low • Risk-based lead purchasing strategy. – high default • Dilemma: Driving up volume vs. cut default Marketing vs. Risk
- 4. What is the goal? • Risk management: cut default rate • Marketing goal: increase lead/loan volume Show me the money! Maximize long-term profit: Optimize both marketing and risk management on one platform
- 5. Strategy: Use data-driven strategies to maximize profit, no matter data is big or small Improve risk management: go deeper in data- mining Instead of using off-shelf credit score, build customized model based on target population to improve prediction of default. Dig into the details of data. “Devil hides in details. ” Improve marketing: go broader in data-mining Build profit model based on short-term and long-term profitability. Use profit score to decide lead price and underwrite loans. Drive up good lead traffic. “We want market share, but only the good market share.”
- 6. Strategy: Continuously optimize operations using data-mining • Improve organic web traffic: Search engine optimization, Google search rank, improve the cost- efficiency of the banner ads. • Improve operation efficiency: Use historical data to optimize the call center resource allocation. • Improve collection operation: Build collection model to decide when, what channel and how to do collection. • Improve debt selling strategy: Build debt evaluation model, Sell the bad debts that is not worth to collect in early stage and get better return.
- 7. Strategy: Continuously optimize operations using data-mining • Improve organic web traffic: Search engine optimization, Google search rank, improve the cost- efficiency of the banner ads. • Improve operation efficiency: Use historical data to optimize the call center resource allocation. • Improve collection operation: Build collection model to decide when, what channel and how to do collection. • Improve debt selling strategy: Build debt evaluation model, Sell the bad debts that is not worth to collect in early stage and get better return.