Data scientists and analysts use various methods to understand and derive insights from data. Within the Analyzing Data domain, the first level of inference is drawn from data using various statistical tools, technical visualizations, or data models to understand the patterns. Whether such indications from data are of business relevance and lead to true business insights is determined with appropriate analysis in the Interpret and Report Results domain. For example, there are some surprising insights that were discovered by combining structured and unstructured data when the density of Uber rides was merged with the crime rate for the city of San Francisco. It was observed that the highest number of Uber rides originated from high crime neighbourhoods. Although it is a fascinating correlation, demand prediction for Uber rides should not be modelled on the crime rate without stronger evidence of a relationship. Analysts use a mix of sound statistical judgment and explanatory analysis to translate data patterns to useful insights, especially when the findings are counter to common business practices.

Analysts use multiple visualizations to derive insights from the data collected. Visual models are developed with a variety of data visualization tools. Visualization from a technical perspective differ from visualizations that are intended for business stakeholders. For example, an error residue graph, a technical visualization which shows a decrease in prediction error as the number of predictors increases for a revenue forecasting problem, may be useful for determining the optimum number of variables to use for revenue forecasting. A marketing stakeholder will likely be more interested in a visualization that shows how ad spends relate to overall revenue.

To effectively understand the insight, analysts adopt a design thinking perspective to the visualization and data story explaining the visualization. Inputs from 2.4.3. Determine Communication Needs of Stakeholders play a key role in thinking through the type of visuals or other methods used to clearly articulate the insight and make it business relevant. Both standard (bar graphs and line graphs) and custom visualizations are used to assure meaningful, usable analytics for the business are communicated.

Organizational skills, systems thinking, design thinking, creativity, attention to detail, stakeholder orientation, and industry knowledge are all important skills required to process information and review and assemble the results in an organized fashion. Analysts also require the ability to view results from a holistic viewpoint.

Visualization Best Practices The ability to effectively derive and explain insights largely depends on visual communication. There is no one size fits all approach to visualization. Forms, graphs, dashboards, and reports are all useful for explaining business insights. When developing effective visual communications, analysts keep the following practices in mind: When there is only a single or a couple of metrics involved, simple text may be a more effective way to communicate the metrics. For example, ROI, profits, percentage, and average values. Text, strategically placed to highlight important facts, is a great way to focus attention. Communicate individual insights through their own individual graph increases clarity. When there is a limited set of metrics, a tabular summary can be more effective than a graph. The simplest forms of graphs and charts highlight the focal message. Complicated graphs, which are used for visual appeal only, may end up complicating the message. For example, a pie-chart where the audience needs to interpret arc lengths and angles can be replaced with a simple horizontal bar chart. 2-dimensional graphs with appropriate colours and labelling can be more effective than 3-dimensional graphs as it is difficult to visualize depth. Superimposing graphs with a secondary axis is generally not a good idea. For example, a vertical bar graph showing quarterly revenue and a line graph showing a trend for profit in a single graph with a secondary y-axis will lead to confusion. Depending on the context, an interactive or a static graph may be more suitable. Statistical or mathematical parameters used in a visual should be explained. Maps and diagrams are other visuals that can be used besides graphs. Prototyping is better suited for prescriptive/predictive models whereas graphs are a good way to represent descriptive analytics. Apart from these basic principles on visualizations, analysts should be well versed in the design concepts and frameworks for visualization. For example, a good visualization might include 6 core principles from Gestalts' theory of design: proximity, similarity, enclosure, closure, continuity, and connection.