Peer to Peer Magazine

61 WWW.ILTANET.ORG » Predictive: Questions that make predictions about future data based on current data EXAMPLES: • What will the total number of hours worked be on this maer? • Will fees billed increase or decrease for this client? • Is this document relevant for discovery? » Causal: Questions that look to prove whether a change in one particular variable, on average, leads to change in another EXAMPLES: • Do secondments lead to increases in business? » Mechanistic: Questions that describe the actual mechanism of interaction between two variables (mechanistic studies are rare outside of physical science fields like engineering) Two Exploratory Examples The following are two examples of the power of question-based analytics. » What different client archetypes exist? A law firm that has a hundred or more clients might want to discover which clients are most like each other. Clustering, a type of unsupervised machine learning, can look at your client data and group the most similar ones based on predefined algorithms. When using clustering, consider the data you feed into the algorithms as the clusters will reflect the relationships between clients in terms you select. Once your clients are grouped with others that are most similar, the real analysis to discover what the groupings mean can begin. Is there a stand-out group with high profitability? Further investigation could reveal something in common driving high performance. Does one group have a large number of hours worked but poor profitability? This group might need legal project management to identify and address discovered issues. The Analytic Power of R for Decision-Making FEATURES » What does a typical maer in this portfolio look like? For a large, multi-year portfolio that could contain hundreds of maers, advanced graphing capabilities and various options of complex charts help you visualize current states and trends. R can summarize data directly from your sources, plot data from two different sources on the same chart (e.g., budget vs. actual) or automatically assign colors to different categories of values or split them into separate charts. Advanced Tip: If you want to look at the legal spend of each of your maers, start with a CSV file that has timecard information, and name that file as the data source in a ggplot2 chart. At this point, you can decide whether you would like lines, points or bars. If this portfolio contains multiple types of work, R makes it simple to split them off into separate facets to show the curve for each type. Adding a Loess smoother will give you a picture of the path a typical maer takes over its life. This analysis can be useful in identifying a portfolio's outliers for further analysis and as a benchmark when renewing the portfolio agreement. Once your clients are grouped with others that are most similar, the real analysis to discover what the groupings mean can begin.

• Cover