A recent survey by SAS, the provider of business analytics and business intelligence software, shows the number one reason organizations are not planning on implementing Big Data is their lack of understanding of how Big Data applies to their business. The solution is predictive analytics. Predictive analytics is the application of Big Data.
It is easy to marvel at the amount of data that is currently being collected and organized. The total data supply in 2012 was 2.8 zettabytes (ZB) or 2.8 trillion gigabytes (GB). However, the volume of data available isn’t nearly as important as applying data to improve and automate existing business processes. Here are CAN’s Do’s and Don’ts for successfully applying Big Data using Predictive Analytics:
1. Focus on the application not the technology. It is easy to focus on technology. After all, companies spend millions of dollars in marketing to get you excited about their technology. Often, they use this excitement to mask the specs and limitations of their technology and it can take a lot of hard work to break through the marketing hype. Unfortunately, the only way to realize a return on investment is by applying the technology to solve a problem. So, don’t focus on the technology but how it will help improve your business.
2. Don’t focus on data collection. Only 3% of the 2.8 zettabytes – 2.8 trillion gigabytes – of data available in 2012 was ready for manipulation, and only .5% was used for analytics. While it might be fun to talk about the “Big” in Big Data, your focus should be on applying data instead of collecting data. This is why CAN focuses on predictive analytics. Predictive analytics allows a company to know what data to collect and how to apply it to the current business processes.
3. Focus on existing data. Many organizations delay pursuing predictive analytics because they think they don’t have enough data. The truth is most companies have enough data for predictive analytics, especially if they have been in business for more than three years.
Also, collecting new data sources is not bad, but stable historical data sources are better. Since predictive analytics uses patterns in the historical data, it requires a stable historical data sources. Learn how CAN organizes your data for predictive analytics.
4. Focus on specific business processes not the entire business functions. It is an error to apply predictive analytics to an entire business function such as marketing, sales, or human resources. There are many parts of a business function that require humans instead of machines. The right way to scope a project is to try to automate and improve one routine decision process inside a business function at a time.
As an example, sales is a function. Identifying the best leads is a routine decision process of the sales function. Predictive analytics is great at generating and identify quality leads because successful lead generation depends on making unbiased decisions using hundreds of variables to select the best sales opportunity among thousands of options. However, making sales presentations is best left to the humans. Success requires focus.
5. Start with familiar business processes. There are many unique uses for predictive analytics. Predictive analytics is, after all, a form of artificial intelligence that automates and improve routine decisions by identifying nuanced patterns in historical data. However, it is a good idea to start automating what is familiar, and expand into unfamiliar territory only after you have had success. You are already aware of the nuances of the processes your business uses, you can use that knowledge to test the predictive models to make sure they reflect reality before using them to make decisions.
6. The goal is to automate and improve decisions not deliver more information: Predictive analytic is about automating and improving routine decisions. While this is the same goal of traditional business intelligence’s reports, dashboard, and data warehouse; predictive analytics allows for an improved approach.
Traditional business intelligence treats decisions as calculations with right answers; instead of judgements. With calculations, having more information makes the calculation more accurate. However, most business decisions are judgments with no “right” answer, only a gradient of better answers, none of which are wrong.
Judgments don’t require more data; they require experience and intuition. The power of predictive analytics is that it is able to merge this experience and intuition with the historical data to create better decisions. They are able to model the decision process of your top executives.
7. Don’t isolate non-technical stakeholders by using math talk. Mostly because you need them to help create the improved decision making models talked about in #6; non-technical stakeholders have the experience and intuition required to build a predictive model.
According to a 2013 Survey of Big Data by SAS, 51% of businesses are not implementing a Big Data strategy because of a failure to engage non-technical stakeholders. The following are the reasons that companies are not implementing or planning on implementing Big Data:
- 21% Don’t know enough about Big Data
- 15% Don’t understand the benefits
- 11% No reason
- 9% Lack of business support
- 8% Poor data quality in current systems
- 6% Lack of executive commitment
- 6% Cost/financial resources
In order to avoid isolating non-technical stakeholders it is key to allow them to participate in the model development process without having to understand the math. While there are many ways to do this, CAN created a system called Portal to give the non-technical stakeholders a place to interact and explore the data without having to understand the complex math. To learn more schedule a demo.
8. Budget for implementation and testing. Implementing predictive model requires passing operational data through the predictive model. A predictive model is a series of variables, coefficients/values for each variable, and the relationships between variables.
Predictive models are converted from mathematical notation into computer code using Predictive Model Markup Language (PMML). This allows a model to be run against a database as long as the variables in the model match the data columns in the database.
This is the hardest part of implementing and testing predictive models, because developing a predictive analytics model is very complex. It requires merging data from many different sources. There are many databases with many tables that need to be merged into a single table, meta data added, editing and deletion of duplicates, and the creation of new variables by transforming existing data (figuring out percent change per year). While these transformations are essential, they make operationalizing predictive models difficult because the data fields used to develop the model are derivatives of the original data fields available in the operational systems.
If you are building a predictive analytics application from scratch it is important to budget 3 to 4 times the cost to develop the model for implementation and testing. You can build great models, but they are only useful if the results are applied. Learn how the CAN simplifies the model development and implementation process from 30 days to less than a week.
9. Frame success in hours instead of dollars. It is very common for businesses to reduce decision to dollars. However this over simplifies the impact of predictive analytics. The value of predictive analytics is that it frees people to do more valuable work by improving and automating routine decisions. The impact of making better decisions represents a significant monetary return on investment, but is minor compared to the impact of having more time to spend on non-routine decisions, being creative and communicating with customers, the public and employees.
A good example, CAN built a model to help a company reduce employee turnover. The model helped the company save $1.7 million per year. In dollars the impact of the predictive model is minor compared to the company’s $8 billion per year in revenue. However, the impact is significant when expressed in hours. CAN’s model allowed the company to avoid hiring 600 new employees each year. The value of the time needed to find, hire, and train 600 new drivers far surpassed the 1.7 million in revenue. HR and management now have the time to build and improve HR, management and training for their current drivers.
10. Don’t forget about policies, privacy and security. Predictive models contain the knowledge, experience and intuition of the business process modeled. This makes security very important. A security breach doesn’t mean just the loss of data, but also intuition and experience. CAN goes to great lengths to protect your data and so should you.
Most companies we work with have a need to create policies for how to use predictive analytics. A useful policy is that the methods should support the goal. The methods you use to acquire new customers should increase sales by themselves. If you are using predictive analytics to increase employee retention, don’t use predictive analytics to analyze and monitor employee emails. Instead, use predictive analytics to identify that you can increase employee retention 10% by avoiding 5 recruiting sources and providing employees with an extra $70 worth of work each month. Employees are proud of their company and love getting more money, but they don’t like to be pandered to or snooped on.
In conclusion, Gartner published that the number one struggle around Big Data for organizations in 2013 is knowing how to get value from Big Data. The best use case of Big Data is to use predictive analytics to improve and automate routine decisions so that employees can focus on being creative and communication. Use predictive analytics and finally get an ROI from Big Data. Schedule a demo to learn how to get value from Big Data.