• “Hacking” or programming: This emphasizes the ability to access data and to think algorithmically. Data comes in a huge variety of forms and is stored in many ways. Team members need the skills to deal with this. They need to be able to get at this data and clean and manipulate it. This does not mean being an expert in computer science who can design compilers. Indeed it does not mean being an expert in database design and maintenance. It is a little bit of everything needed to get the job done. They need to recognize common database structures. They need to model data sensibly. Once data has been shaped into a good format, analytics is usually performed. This is where knowledge of algorithms and awareness of computer science topics such as complexity and data structures is advantageous. An appropriate data structure and an algorithm design can be all the difference between a feasible execution time and analyses that take hours to finish. The former is obviously the priority in Guerrilla Analytics.

• Substantive expertise: This is what is often termed “domain knowledge.” The more your team knows about the business domain they are working in, the better placed they are to find, understand, and add value to the data they must work with. A geologist estimating yields from a resource knows the order of magnitude of a sensible result. A marketing analyst should know the typical take-up rates of a particular type of campaign. Without this substantive expertise, the analyst is placing blind faith in their program code and their tools.

• “Math and statistics” knowledge: What Drew calls math and stats means some awareness of statistics as would be used in many machine learning and statistical analyses. If you want to get some insights from data beyond a ranking and a table of counts, you will need to apply a statistical analysis or machine-learning algorithm. Typical tried and trusted methods include regression, neural networks, Bayesian networks, decision trees, clustering, and association rule mining. To use these algorithms successfully, you need to understand the conditions under which they operate, the appropriate situations in which to use them, and the techniques for evaluating and tuning their performance.

• Communication: Communicating with customers who are not data literate is critical. It is also important that team members can work constructively with one another.

• Visualization: Developing minimal, insightful, and beautiful visualizations to tell a story can often be the difference between analytics being understood and used by the customer or being discarded.

• Software engineering: Team members need an understanding of version control, build tools, documentation, specifications, and deployment. Without these techniques, the complexity of analyzing data can quickly turn into chaos. Second, if the team is ever to effectively consolidate its knowledge and increase its effectiveness, team members should appreciate how to build tools that can be reused. A sophisticated fuzzy match algorithm’s value to the team is limited if it can only ever be used by its creator.

• The data environment: A lot of data resides in databases and most Guerrilla Analytics is done in a DME. While database tuning and management is a specialized skill, team members do need to understand how to set up database users, permissions, logs, and table spaces as well as how to connect to a database. Web applications are ubiquitous now. They are the front-end to many data sources and they are increasingly the presentation medium for analytics reporting. Team members need a working understanding of how web applications are structured, deployed, and developed.

• Mindset: Team members need to be able to prioritize work to add value incrementally. They obviously need a Guerrilla Analytics mindset where they can focus on data provenance despite project disruptions.

• The customer. You must be able to describe your analyses and their importance in simple business terms. A customer rarely cares about the sophisticated algorithms behind results. They only care about the bottom line and how your analyses support the decision they need to make. Know when to impress with knowledge of decision trees, data contortions, and data velocity stats, and know when to tailor communication at the right level.

• Team members. In a healthy Guerrilla Analytics team, there will be disagreements between team members and challenges to a particular approach that you are advocating. It is important to be able to understand where these challenges are coming from and present a reasoned non-emotional argument for your position. Teams damage themselves irreparably because of arguments over technology and process. A good communicator can influence their technical peers, superiors and reports, and get their message across.

• Management. The higher up people go in management, the more removed they become from the details of Guerrilla Analytics. Over time they may forget how difficult it is to estimate analytics jobs before getting stuck into the data. This worsens as technology moves on. They forget how surprises are always lurking in every dataset. This causes problems when management expectations of delivery are not in tune with reality. A Guerrilla Analyst must be able to identify and communicate issues early and present incremental approaches to a problem. This makes management easier because managers have frequent visibility of progress and a “safety net” of delivery they can rely on even if the more advanced analyses and hypotheses do not come to fruition.

• Independent variables. These are the inputs to a statistical model.

• Dependent variables. These are the outputs from a model – the things that are predicted.

• Transformations. How to change and transform variables appropriately to make them suitable for an analysis.

• Choice of variables. How to assess whether variables should be included in a model.

• Performance. The concept of false positives and false negatives, and how to measure algorithm performance.

• Testing. Model validation as discussed in Chapter 15.

• Changes over time: Here you want to show how one or more quantities change over some time period. Bar charts and line charts (perhaps with multiple series) are a good option.

• Static proportions: In many cases you want to show the breakdown of a population. If the population is static, a pie chart or a tree map is a good option. A bar chart is also useful if various categories are being compared.

• Changing proportions: When data proportions change over time, an area chart or a stacked bar chart with time on the horizontal axis are good options.

• Correlation: This is when you want to demonstrate relationships between variables. For example, when variable A increases then variable B is seen to decrease. The scatter plot is the classic way to demonstrate correlations.

• Distribution: Sometimes you want to know how a population breaks down into buckets. This is a distribution and is best represented by a histogram.

• Comparisons: In these cases you want to compare a variable in terms of its distribution and key statistics such as median, mean, and range. A box plot offers a concise but information-dense way of doing this.

• Outliers: Here you are looking at particular variable values that are significantly different from the rest of the population. Again the box plot is very useful here. If outliers as a combination of variables are of interest, then a scatter plot quickly demonstrates interesting cases.

• Geography: If locations are of interest then the obvious solution is a map. You can color regions, or place bar and pie charts at locations to bring out variables at that particular location.

• Connectivity: Connectivity often arises when you look at networks and relationships. For example, you might consider email or phone call traffic, or connections between web pages. In these cases a network diagram is best.

• Hierarchy: The canonical way to represent a hierarchy is with a tree diagram.

• Change in rank: In some cases you would like to know how the rankings of items have changed between two points in time. This is where a slope chart comes into play.

• Sequential additions and subtractions: In some scenarios you would like to understand the cumulative effect of sequentially introduced additions and subtractions. This is where a waterfall chart is very useful (Rasie, 1999).

• Curiosity: They must relish tearing data apart, and twisting it inside out to find insights. Many of their lines of attack will lead to nothing but the Guerrilla Analyst must enjoy that exploratory process rather than expect a specification of what they need to do.

• Passion: They must care about presenting results in a beautiful and insightful way so that they influence customers. Fancy algorithms and data gymnastics are irrelevant to telling a data story. The Guerrilla Analyst must want to explain their analyses and improve them.

• Discipline: Maintaining data provenance requires discipline. An analyst needs to keep versioning, testing, conventions, and quality code in mind despite a sometimes chaotic project environment. Hacking around datasets carelessly without regard for the data provenance implications creates technical debt.

• Patience: Data is like a box of chocolates – you never know what you’re going to get (Zemeckis, 1994). Some might naively think it is just a matter of “add another column.” The reality is that there are always surprises lurking. A Guerrilla Analyst needs to be able to deal with the unexpected setbacks and persevere with wrangling insight out of the data they have been dealt.

• Data analytics is a wide ranging diverse set of activities. What defines a Guerrilla Analyst is that they have to produce insight in dynamic circumstances while being quite highly constrained.

• It is unlikely that any individual would have all the required skills of a guerrilla data analyst; however, you should ensure your team overall has as many of these skills as possible.

• The main skill areas for a Guerrilla Analyst are: