Building and Managing Analytics Teams Stanford SC Forum Roundtable Event Creating Business Value with Analytics and Big Data Thomas Olavson Director, Operations Decision Support Google July 20, 2011
My perspectives colored by my experiences PhD, Management Science & Engineering, Stanford, 2001 Previously - 10 years as analyst and director of Strategic Planning and Modeling (SPaM) at HP Currently Director of Operations Decision Support, Google Member of INFORMS Roundtable of OR group leads since 2007
Discussion topics - building and managing analytics teams Functional Focus and Reporting Lines Customer Engagement & Funding Model Team Skills Team Size & Structure Data Problem-Solving Approaches
Functional Focus and Reporting Lines Focused functional expertise: supply chain, customer analytics, etc. Centralized or decentralized analytics org structure? Stay close to customers to be relevant and grounded Stay connected with other analyst teams for knowledge sharing & career development
Customer Engagement & Funding Model Maintain autonomy to define what to work on for greatest impact Be customer focused: partner and co-develop with customers and decision makers in any major project What do you offer: consulting services, data mining, tool/ process development or R&D? Funded by one source, multiple sources, or by project? Trust and credibility are sacred be the neutral third party, and do consistently excellent work Use project plans to document customer agreements: scope, timeline, resources, deliverables
Team Skills Analytics Skills Domain Expertise Consulting Skills Generalists: everyone well-rounded, good in all dimensions and great in some Top talent: 3-6 mo. recruiting per hire Mentorship training model: 12+ mo. learning curve Management: respected technically; provide team direction and 1-1, coaching, and then get out of the way Minimum technical requirement for all (analytical MBA or PhD)? Include software engineers? Primarily IE/OR or statisticians? Unique job family for quant analysts?
Team Size & Structure Sweet spot: 8-15 people, min 4 per location Flat organization Let analysts work in pairs Allow people variety in their work Single or multiple locations? Low cost locations (India)? Types of Lead roles for more senior analysts (focus area lead, account manager, etc.)?
Data Be willing to do dirty work : request data, clean data and make do with imperfect data Own data sources, define regular data sources owned by others, or request data on project-basis?
DATA - analysts would prefer to spend less time on data gathering and cleaning % of Analyst Time Spent Gathering, cleaning and vetting data and modeling assumptions Spotting trends & asking the right questions Problem solving Time 9
Problem-Solving Approaches Good problem framing - are we working on the right problems? Good communication with decision makers - can we communicate intuition and insights in simple terms, rather than black box models? Balance between rough cut McKinsey analysis, detailed model development, and software tool development? Team modeling languages of choice: spreadsheets, R, python, etc.?
Problem-solving approaches the right tool for the job High For recurring problems, automate or add model complexity over time One-Time Analysis Ambiguity of Problem* Low Decision Support Tools Short (days, weeks) Time Horizon of Problem Long (years) * Ambiguous problems are those in which alternatives, constraints, objective function and uncertainties are not well understood.
Summary - best practices and choice points for analytics teams Domain Area Focus Focused functional expertise: supply chain, customer analytics, etc. Centralized or decentralized analytics org structure? Customer Engagement & Funding Model Maintain autonomy to define what to work on for greatest impact Be customer focused: partner and co-develop with customers and decision makers in any major project Trust and credibility are sacred be the neutral third party, and do consistently excellent work Team Skills Generalists: everyone good in 3 basics : analytic skills, consulting/leadership skills, and domain expertise Top talent: 3-6 mo. recruiting per hire Mentorship training model: 12+ mo. learning curve Team Size & Structure Sweet spot: 8-15 people, min 4 per location Flat organization Let analysts work in pairs Allow people variety in their work Data Be willing to do dirty work : request data, clean data and make do with imperfect data Problem-Solving Approaches Good problem framing & communication with decision makers are essential. Are we working on the right problems, and can we communicate intuition and insights in simple terms? What do you offer: consulting services, data mining, tool/process development or R&D? Funded by one source, multiple sources, or by project? Minimum technical requirement for all (analytical MBA or PhD)? Include software engineers? Primarily IE/OR or statisticians? Unique job family for quant analysts? Single or multiple locations? Low cost locations (India)? Types of Lead roles for more senior analysts (focus area lead, account manager, etc.)? Own data sources, define regular data sources owned by others, or request data on project-basis? Balance between rough cut McKinsey analysis, detailed model development, and software tool development? Team modeling languages of choice: spreadsheets, R, python, etc.?