Bi servers use ________ to determine what results to send to which users and on which schedule.

To build momentum and prove use cases for self-service analytics tools, organizations should look for quick wins first, said Lyndsay Wise, director of market intelligence at Information Builders, a BI and data management software vendor.

"This means identifying key outcomes or metrics and creating self-service applications that align with taking action -- making business decisions -- based on the analytics and visualizations delivered," she said.

One example could be operational dashboards that help supply chain professionals route materials based on factors like weather, traffic and so on. Similarly, BI dashboards for the C-suite can provide immediate bang for the self-service buck.

"Executive dashboards also provide a great self-service access point by providing insight into overall performance, but [they] let people drill through visualizations to evaluate situations to make better decisions by leveraging more insight," Wise said.

As a bonus, these dashboards would also be a great way to gain buy-in from key executive sponsors needed in order to push out an organizationwide self-service program. When they see the benefits in their daily lives, they'll be more likely to understand the value that self-service BI tools hold for other users across the enterprise.

Successful self-service BI applications require a foundation of effective data governance and management. Experts like Moffo believe that organizations must enable business analysts and users to get creative with how they correlate and visualize data, without sacrificing proper governance. He offered a few first steps in supporting data readiness.

"Tighten up data quality standards so that all who interact with the data have clean data," Moffo said. "Exploratory data is a great way of finding new ways to grow your business, but it still needs to have quality standards so that you are not making decisions off of inaccurate or unclear information."

At the same time, he said it is beneficial to loosen up data governance "just enough" to let business analysts and other users explore data streams that otherwise might not be available to them.

"Open up the data faucet gradually and continually train everyone who is working with the data," he said. "The more they understand, the better the results will be."

One of the ways to build successful self-service is by having IT work collaboratively with the business analytics team to identify how data will be managed. Lyndsay WiseDirector of market intelligence, Information Builders

Self-service BI best practices include a high degree of collaboration among three major stakeholder groups: the business users who will utilize the self-service tools, the BI analysts who support them and IT professionals.

"In general, one of the ways to build successful self-service is by having IT work collaboratively with the business analytics team to identify how data will be managed. This way, the analytics team will develop solutions and IT will manage the overall data assets," Wise said. "In some organizations, IT manages all analytics projects. The challenge to doing so effectively is that most IT departments are focused on technology and infrastructure. Successful self-service requires a team dedicated to solving business challenges by leveraging technology."

Justin Butlion, an analytics and BI infrastructure specialist, said he also prefers a model where BI analysts take the lead in building out data modeling processes and data visualization capabilities, while IT handles the back-end infrastructure.

"[Analysts] know the [data] consumers best and need data visualizations themselves to provide their services. IT [is] generally not familiar with the core business at the level that is needed for mapping out the needs of the business users," said Butlion, who is also the founder of ProjectBI, a community for data and business analysts.

From there, BI analysts should identify power users in business units and departments to help build out tooling, analytical data models and visualizations that work best for their daily workflow.

"Visualization is creative work just as much as it is technical," Butlion said. "Taking an agile business/tech-working-together approach is one way to get the most out of the time spent designing and developing critical visualizations that will eventually help shape the direction of your organization or enterprise."

While a few isolated pilot projects are great for showing proof of concept and gaining quick wins, the only way to sustain self-service BI across thousands of users is to build the program with scalability in mind from the start.

"In many cases, teams develop solutions that meet the needs of their teams or departments. Self-service is seen as a way for various groups to gain a lot of insight quickly," Wise said. "Unfortunately, many decisions at this level are made at the business level without collaboration with IT."

This creates technical debt, unreliable data and compliance nightmares.

"In order to scale, companies need to understand their data assets, how they interrelate across the organization, what infrastructure currently exists and what is required on the platform level to scale," she said. "Basically, in order to scale self-service across the enterprise successfully, businesses need to use a proactive approach and evaluate solution providers that can support the level of future scalability and not simply current use cases."

It's not just a technology problem, either. Groups throughout the organization need to plan for process scalability if they want self-service analytics to truly take hold within their user base. Some organizations, like Morgan Stanley, are driving scalable analytics deployments by formalizing self-service BI best practices for workflows, interdepartmental relationships and more through center-of-excellence or center-of-enablement approaches.

IT and BI leaders must recognize that they're going to encounter tensions they'll need to balance out in the long run as they manage a self-service BI program.

For example, one big conflict that crops up is between BI analysts spending time training users on existing self-service capabilities and building out new ones.

"People treat the analysts like a candy store and constantly want to get their hands on more tools, reports and dashboards," Butlion said. "Adoption is more important, and you need strong analysts that can push back against strong managers that are demanding new and shiny toys, instead of using the tools already available."

Another big conflict is between speed and optimization, Butlion said.

"The BI and operations teams are always pushing for efficiency and optimization within the organization. The issue is that this philosophy might be against the overarching strategy of the company," he said. "If the goal is growth and everyone but the ops team is focused on that, then you end up with a lot of head-butting. It's challenging to find that balance, and it can be extremely frustrating for the ops people."

Organizations that have embarked on or are thinking of adopting self-service BI initiatives need to think seriously about associated data security and privacy strategies, stressed Anees Merchant, global head of digital and applied AI at analytics vendor Course5 Intelligence.

"Data can be misused, or biases can creep into the picture when organizations look at data elements which aren't required for an individual to take steps or actions based on the insights they generate," Merchant said. Personally identifiable information that isn't needed for analysis uses should "be completely kept out of the self-service analytics roadmap," he warned.

BI managers and their teams also need to consider how much data they want to make accessible to different individuals and at what level of depth, as well as how frequently the data should be updated, Merchant said.

Keeping up with compliance requirements on data protection and privacy is becoming a bigger challenge because of new regulations that are being enacted. The most obvious examples are GDPR and the California Consumer Privacy Act; in addition, measures similar to the CCPA have been proposed in several other states in the U.S.

"Staying compliant isn't a straightforward task," said Gene Yampolsky, a BI and data visualization developer who currently works at Wells Fargo under a consulting contract. "However, the thought that data protection and compliance must be a core part of all business practices makes perfect sense." The various regulations provide "rules and guidelines that help organizations protect their systems and data from security risks" as part of self-service BI environments, Yampolsky added.

Deploying a new technology or initiating a new program often requires significant training and change management. That's especially true with self-service BI and analytics initiatives, which means business teams and users need to be involved from day one, Merchant said.

"The plan needs to ensure that the business teams are committed and motivated to make the program successful," he explained. Toward that end, BI managers and business leaders "need to ensure there is adequate time and budget allocated for onboarding and hand-holding as needed."

Equally important, the training and onboarding process should emphasize how business teams can use the insights generated by the business intelligence tools to meet the organizational goals of the self-service BI program.

The focus shouldn't simply be on how to use the BI platform, Merchant said. Instead, he recommended that it be "more aligned to working on what questions the business teams need answers to daily and how the self-service platform can enable them with that."

Self-service BI enables business users to become more self-reliant and less dependent on the IT and BI teams, from data discovery and data preparation to querying, data visualization and reporting. To be successful, a self-service environment "has to support the need for a personalized and collaborative decision-making environment for the information workers," Yampolsky said.

However, Merchant cautioned that things can go wrong if an organization isn't careful about monitoring deployments. Self-service programs can quickly get out of hand, he said, citing the potential for runaway costs due to uncontrolled scaling; the risk of faulty conclusions and insights from inconsistent data; and breaks in the process "that can get wider if not arrested in time."

To avoid these pitfalls, self-service BI best practices include setting up processes that allow the BI team to monitor, manage and control a program without hampering the ability of users to do required analytics work. That should enable the BI program to efficiently scale as needed and achieve ongoing business success, Merchant said.

Line charts are familiar to most people, and well-designed ones can be easily analyzed at a glance. "Nothing is better than a good old line plot when monitoring numerical attributes over time," said Paolo Tamagnini, data scientist at open source analytics software vendor Knime AG.

Line charts can show different measures of categorical attributes on separate colored curves so users can quickly compare them. In addition, making a line plot interactive can help users browse through a large number of curves that might be confusing in a static visualization.

During the COVID-19 pandemic, for example, line charts have been the most used data visualization technique to illustrate the spread of the disease and to compare the effectiveness of containment measures country by country, as in the interactive example provided by Tamagnini that's shown below.

The bar chart is one of the simplest and best known data visualization techniques. "Humans eyes are really good at comparing the length of sorted bars -- not angles, not color gradients, not curved shapes," Tamagnini said. "That's why I still love to use bar charts whenever I can."

Patrick Miller, who leads the data reporting and visualization team at consulting firm West Monroe Partners, said bar charts can be understood by virtually anyone without training or explanation. But watch out for the temptation to make them too busy, he cautioned.

The pie chart is another well-known type of data visualization, in which different percentages of a whole are represented as slices of a pie. They make great eye candy, but visualization experts said they don't convey the differences between data as well as other techniques do. As a result, they can be hard to interpret.

"Pie charts rank low in precision because users find it difficult to accurately compare the sizes of the pie slices, although such charts can be helpful when you're giving a high-level message," said Manjula Mahajan, director of data and analytics at storage vendor NetApp.

Bubble charts are used to express three dimensions of data based on the x-y location and size of each bubble. With minimal explanation, a bubble chart can provide useful information about relatively complex data sets, said Chris Adams, vice president of product management at SparkPost, maker of a predictive analytics platform for analyzing email outreach efforts.

Adams recommended, though, that the labels for the various bubbles be clearly visible. Also, he said the bubbles should be sized so they don't bump into each other and are different enough to quickly convey insights about the data they represent. For example, in the SparkPost visualization shown below, the bubble sizes indicate the relative frequency in which different sentiments were used in the subject lines of an organization's email campaigns.

But not everyone is a fan of bubble charts. Mahajan said they aren't always a good fit for BI dashboards and can require too much mental effort to understand "due to their lack of precision and clarity."

Daniel Chalef, vice president of data science at SparkPost, said histograms offer an effective way to visualize the distribution of the values in a data set, which may help users when analyzing the data. Averages are often misused and can be misleading, Chalef said. For example, the average deal size in a sales pipeline can be skewed by several large deals. To provide more accurate information, an analyst can use a histogram to show the number of deals within different price ranges, as in the example below.

Histograms look like bar charts, but they're specifically designed to illustrate data distribution. The equally sized numerical ranges that the data values get grouped into are called bins, some of which may not have bars if no data falls into those ranges. One potential issue with histograms is ensuring that the bins are properly sized to convey useful and relevant information.

A heatmap uses color coding to show the magnitude of data elements in two dimensions. CloudCheckr's Davis said that can help his company's customers understand the temporal characteristics of their cloud infrastructure. For example, CloudCheckr uses heatmaps to visualize the use of cloud resources in different time periods so customers can see good times to shut off or downsize servers to reduce costs.

But Joshua Moore, principal technologist for cloud analytics at NetApp, said the heatmap is his least favorite visualization technique, at least in most situations he has seen it used. He often sees dozens or hundreds of key performance indicators (KPIs) on things like the operational health of servers tracked on a grid in one view, with all but a few of them showing as green.

"This just creates clutter and noise," Moore said. "The healthy KPIs require neither action nor attention, so why does anyone need to see them? Conditional formatting like this is best kept to a handful of top-level KPIs, not huge heatmaps."

Scatter plots are useful to display the relative density of two dimensions of data. Well-designed ones quantify and correlate complex sets of data in an easy-to-read manner. "Often, these charts are used to discover trends and data, as much as they are to visualize the data," Adams said.

For example, the SparkPost scatter plot below enables digital marketers to correlate the number of characters in email subject lines with email read rates to help them identify best practices and plan future campaigns.

Knime's Tamagnini likes to use scatter plots to show relationships for single data points -- for such uses, they're easier to read and interpret than bar charts and other visualization techniques, he said. However, he added that scatter plots struggle when analysts try to display more than two dimensions. Also, packing too many data points into a scatter plot can make it hard to decipher.

This technique, formally known as t-Distributed Stochastic Neighbor Embedding (t-SNE), uses a machine learning algorithm to model high-dimensional data sets as two- or three-dimensional data points for display in a scatter plot, using colors, shapes or other visual elements to represent the third dimension. It was developed to address some of the limitations of conventional scatter plots.

Data scientists tap t-SNE to transform relationships in the raw data so they're easier to visualize. "The adoption of data transformation techniques, like t-SNE, is the direct consequence of the increase in data literacy and data science expertise in the data visualization field," Tamagnini said.

Sankey diagrams represent data and process flows via lines and arrows with different widths that illustrate the magnitude of individual flows. SparkPost's Chalef said they're a "beautiful tool for illustrating flows in a network," using a directed graph that runs left to right.

A Sankey diagram can be visually overwhelming at first, West Monroe's Miller said. But it's a flexible visualization technique that can work even with massive fluctuations in the underlying data set, he added. For example, he has used it to visualize how applications move through a complicated, nonrigid workflow.

"A Sankey diagram is not easy to understand and requires explanation to the user in order to gain insights," Miller said. He recommended that designers include instructions or help text in a visualization or a dashboard where it's embedded, ideally in a hover-over information field.

A treemap displays hierarchical data using nested blocks that are sized differently based on the data values they represent and can be packed into other blocks to show large data sets. For example, IT administrators often use treemaps to track the use of disk space, memory or CPU resources in systems. The blocks can help users identify data trends, although fluctuations in their sizing and how they're ordered are potential complications that visualization designers need to consider.

Davis said CloudCheckr has started experimenting with treemaps to visualize cloud security vulnerabilities based on the severity and scope of individual security issues. It can be hard to communicate the size and scale of a particular vulnerability to customers using other data visualization techniques, he said, adding that the treemaps show promise in helping users to understand the severity of the issue at hand and the level of urgency it requires.

A circle packing chart is a treemap variation that use circles rather than blocks to represent the relationship among data objects. See Ho Ting, senior director of software engineering for CommScope's Ruckus line of networking products, said he finds them useful for showing an overview of a network and illustrating the severity of performance issues in different parts of the network.

Plotting different-sized circles within larger circles makes it easier to display network data three layers deep, including the overall network, the individual controllers and the different zones within each controller, Ting said. That provides a helicopter view of a network's health for a quick scan by users, plus deeper visibility into problems, he added.

A network diagram represents the way data elements are connected by showing nodes and the link lines between them. "Network diagrams help visualize data that is most hard to grasp without visualization," LatentView's Hariharan said. There are many examples: networks of friends and the strength of their relationships, data transmission between systems and devices, financial networks, disease propagation, transportation and people movement, criminal activity and more.

One of the challenges of designing a network diagram is deciding what to show and what to hide. "Many times, the default network diagram comes out like a ball of mud with no obvious patterns," Hariharan said. Designers need to include a mechanism to show the big picture and then progressively drill down and see more details as a user zooms in on the diagram, he advised.

"Simple charts are powerful, but combining them together is even more powerful," Knime's Tamagnini said. For example, he combined a bar chart, a scatter plot and a parallel coordinates plot in a single dashboard, shown below, to provide a more detailed and interactive view of the underlying data for an analytics application to predict customer churn.

The combination of the three visualizations enables users to select all customers who churned or didn't churn by clicking on the associated column in the bar chart. Data on the selected group of customers is automatically displayed in the two plots so the users can compare numerical attributes to see whether there's a correlation between them and customer churn. It's an example of how the visual analytics mantra coined by UX and visualization pioneer Ben Shneiderman in 1996 can be applied to all sorts of data visualization techniques: "Overview first, zoom and filter, then details-on-demand."