Once data is in the data warehouse and converted into usable formats, it’s ready to be analyzed.
Analysis is fundamental to deriving value from data and enabling information-driven decision making.
Here’s what you need to know about this.
Online Analytical Processing (OLAP)
Day-to-day business relies on Online Transaction Processing, or the creation and use of data to support operations. Online Analytic Processing (OLAP), on the other hand, refers to the use of data for analysis and intelligence.
If the data warehouse is the back end of business intelligence, OLAP represents the front end.
OLAP tools allow users to access the data in the warehouse and use it to run queries and generate reports. For example, if a user wants to see a comparison of products that were sold in California in September versus those that were sold in New York at the same time, OLAP can perform the processes necessary to retrieve and display that information.
As the foundation for many types of business intelligence applications, OLAP offers the capabilities for complex analysis and trend modelling. A key aspect of OLAP tools is that they store data in multidimensional, rather than relational, databases.
A relational database stores data in two dimensions – think of a spreadsheet, which groups data based on static rows and columns.
With a multidimensional database, however, each attribute of a record is stored as its own dimension in the database.
That allows for much greater flexibility in making comparisons, tracking trends and looking at data from different points of view. In other words, OLAP is what allows data to be used to answer any questions decision makers may have about their business.
These multidimensional databases are often referred to as OLAP cubes.
OLAP cubes are designed to allow for business to make queries using plain English, and the data is organized to allow for minimal processing time.
Different methods of manipulating data in an OLAP cube include:
- Slicing – Pulling out one subset of a cube (think of it as a rectangle) and using it to create a new cube with one fewer dimension. This is used to isolate only the criteria necessary for a given query. For example, if you have data about the sales of products in every state for each month in the last year, you may not need to look at the monthly data.
- Dicing – Producing a smaller cube by pulling out specific values from multiple dimensions. For example, you may only need data about specific product categories.
- Drilling up/down – Moving among levels of data, ranging from the most detailed sets to more summarized sets of data. For example, one level may be about total sales in each state over time, while the user can drill down to see more detail for each product.
- Rolling up – Summarizing data by combining attributes based on a hierarchy. For example, data about sales in each state could be rolled up into data about larger geographic areas.
The goal of business intelligence is to use data to answer questions and drive decision making.
Analytics, in a broad sense, can be defined as the conversion of data into useful information and intelligence. That’s done by using data to answer questions or to spot patterns and trends that the initial questions may not have asked about.
It can be performed by both people and technology. Technology is used to perform analysis and spot patterns that a person may not be able to see – for example, by analyzing lots and lots of Big Data – as well to organize and report data in ways that make it easier for people to spot patterns.
Analytics performed using technology is often based around statistical modeling.
Statistical models use historical data to determine the probability of events occurring based on certain criteria. For example, data about customers who have jumped ship for competitors can be used to create a model to predict which current customers are at the greatest risk of leaving.
Businesses typically use three types of analytics depending on what kinds of questions they’re asking:
- Descriptive Analytics – This is about learning more detail about historical facts. The basic questions answered are: What happened in my business? Why, when and how did it happen?
- Predictive Analytics – This is used to ask questions looking ahead. For example: What is likely to happen to my business or in my industry in the future?
- Prescriptive Analytics – Analytics can also recommend actions based what was discovered using descriptive and predictive analytics. Basically, prescriptive analytics answers the question: What should my business do in response to what has happened or what is likely to happen?
The specific process of getting from raw data to useful analysis is often referred to as data mining.
Despite what the term may sound like, this step doesn’t involve mining or extracting data – that’s already been done during the ETL process. Rather, data mining refers to the extraction of patterns and knowledge from that data. Data mining is also known as knowledge discovery and data discovery.
Analytics and data mining are essentially what turn raw data into information and then actionable intelligence. When we say data, we mean any facts, numbers or text that can be processed by a computer. Through data mining and analytics, data is turned into information about patterns and trends within the data and intelligence, which is knowledge about historical trends and future patterns that can aid in decision making.
Knowledge discovery refers to the process of finding useful patterns in the data that can be used for intelligence. How does knowledge discovery work? Essentially, it’s based around looking at relationships within and among sets of data.
Generally, four types of relationships are sought, according to a paper by UCLA’s Jason Frand:
- Classes – This is data that’s already contained in predetermined groups. For example, a restaurant might want to know at what times customers visit and what they order at those times.
- Clusters – This refers to data grouped according to logical relationships. For example, a business may separate data based on customer segments or geographic areas.
- Associations – This is when data is mined in order to identify unexpected relationships. For example, a supermarket may perform data mining and find out based on associations that when men buy diapers they also tend to buy beer at the same time.
- Sequential patterns – This is used to anticipate future trends and behaviors. For example, Netflix and other content providers often use sequential patterns to predict what content users will like based on what they’ve accessed previously.
For business intelligence to work, the intelligence needs to get into the hands of business users and decision makers. That’s where reporting comes in.
While the processes and tools described earlier are involved in turning data into intelligence, reporting is the way intelligence is accessed and distributed. Basically, business users have questions, and the business intelligence software generates reports that answer those questions.
That doesn’t mean the system simply spits out a spreadsheet listing all of the relevant data. Beyond those simple operational reports, analytical reports contain information specifically targeted to aid in strategic decision making, presented in ways that the user can easily understand.
Since the end goal of business intelligence is to get information to business decision makers and have them act on it, reporting is a critical aspect of any business intelligence system. Here are some of the reporting features organizations should look for:
- Self-service – Business users are able to access the reports they need to answer their questions quickly, without having to go through someone in IT. Self-service is becoming a more integral aspect of all business intelligence tools.
- Flexible reporting – Different types of reports can be presented to different groups. For example, some people may want to view information in different ways. Also, your organization may need to run reports in different languages, so be aware of those requirements, too.
- Drillable reports – Many reporting tools are interactive and allow users to see additional information by clicking on part of the main report. Often, the answers to questions will lead to more questions, so it’s important that users have an easy way of accessing additional data when they need it.
- Interactive and customizable reports – Business intelligence users should be able to view their reports in different ways. Interactive and customizable reports allow users to change the data they’re viewing on the fly so they can more easily have their questions answered. For example, if you’re looking at the performance of different product lines in various contexts, you may want to compare just the top and bottom performers. An interactive report would allow you to remove all the other data that you don’t need to see.
- Sharing and collaboration – One of the benefits of using business intelligence is that it can help companies take a more holistic approach to making decisions about the organization. Therefore, it’s important that reports can be shared with stakeholders throughout the organization, so they can offer their input and understand the reasoning behind decisions. Those decisions are never made in a vacuum, so report sharing is an important way to aid collaboration.
- Mobile reports, so users can access their intelligence when and where they want. In all aspects of running a business, a lot of work is done when people aren’t sitting at their desks. Mobile reporting offers greater flexibility so users never have to wait to get their questions answered.
A key aspect of intelligence reporting is how the data is presented to the user.
The goal is to show the data in ways that are accurate, yet easy for the user to understand. Lists of numbers are rarely the best way to do that. Instead, visualizations are used to present data in a digestible format.
Common types of visualization include charts, graphs and maps, as well as more advanced types such as infographics. Some visualization tools also offer animated and dynamic visualizations that users can interact with.
Seeing information visually is a good way to spot trends or notice warning signs and potential opportunities without the expertise of a data scientist. While different types of reports will require different kinds of visualization, what good visualizations all have in common is that they enhance the viewer’s understanding of the data.
However, it’s important to keep in mind that visualization can sometimes distort the truth. For example, a chart can be set up to make certain trends seem more significant than they really are. When evaluating data visualization tools and coming up with visualization strategies and techniques, it’s important to keep some keys in mind:
- Aim for visualizations that are both accurate and easy to use and understand.
- Consistency is also important; the same design principles should be used for data visualizations throughout the organization.
- Go beyond basic tables and charts. Organizing data into a table isn’t really a visualization; it’s just a different way of listing statistics. For example, if your goal is to compare the sizes of numbers in different categories, use a visualization that uses different-sized objects to make the comparison obvious.
- Make them interactive. Today’s visualization tools use interactive capabilities to pack a lot of information into each visualization without cluttering up the presentation – for example, allowing users to click on a section of a graph in order to see additional information in text.
Many business intelligence and analytics systems come with their own visualization tools. However, external tools are also available if additional functionality is needed beyond what’s offered in the company’s current or preferred business intelligence software.
As self-service becomes a larger aspect of business intelligence, the business intelligence dashboard is becoming a critical tool.
The dashboard is typically the first thing a user sees after logging into the system. Customized for each user or user group, the page displays a collection of the most pertinent information, using various visualizations, for that person’s role.
The dashboard shows snippets of information the user can review quickly, with the option to choose different items for more detailed reports and visualizations. For example, a head of sales or marketing may log in to see a dashboard that includes a map of where leads are located geographically, a chart showing the source of leads, graphs showing the average cost per lead for each channel, etc.
A good dashboard will be critical to getting user buy-in for a business intelligence initiative, and for allowing BI to have an actual impact on company decision making. Here are a few dashboard best practices to keep in mind:
- Only show what’s relevant – The goal of a dashboard is to provide the easiest possible link from intelligence to action. Therefore, the focus must be on relevancy – i.e., giving the user all of the relevant information, and only what is relevant.
- Offer some customizability – Users should be given some control over what information they see, either by allowing them to customize the dashboard on their own or by getting their input when dashboards are designed.
- Incorporate strong visualizations – It’s also important to look for dashboards that are well designed and present information clearly, without any clutter, and using effective and accurate visualizations. To add greater accessibility, many dashboards are web-based, meaning that users can log into their dashboards from anywhere they have an Internet connection. Some systems also offer mobile dashboards that can be accessed from smartphones and tablets.
As with visualization tools, dashboard software is available bundled with larger business intelligence systems or as a standalone product. Businesses will need to evaluate their dashboard needs and capabilities and choose the right tools accordingly.