For the last twenty years or so, the mental model of analytics maturity has been along the lines of the diagram below, starting with basic gathering of data and culminating in proactive, automated use of advanced algorithms.
Gartner, for example, refers to four levels of capability: descriptive, diagnostic, predictive, and prescriptive analytics.
Do You Have to Do Them in Order?
Organizations are normally assumed to evolve from left to right as they become more mature in their use of analytics—but the reality is more nuanced.
The horizontal axis shows the complexity of the analytics technology rather than time. While many organizations struggle to cross the “chasm” to predictive analytics because of limited data readiness and skills, it is possible to use a combination of the different technologies simultaneously for different needs and projects.
After all, the biggest barriers to widespread use of self-service analytics are typically organizational and cultural barriers, rather than technology or data quality. In these cases, embedding sophisticated predictive analytics in a discrete business process associated with lots of high-quality data may be an easier project.
For example, many finance departments struggle to automatically match invoice data with corresponding bank payments, because of different reference numbers, amounts, and so on. New applications are now available that use machine learning to increase the matching rate, based on the existing data in the finance system.
Because these types of applications are targeted at narrow, complex decisions executed hundreds or thousands of times a day, they can potentially be implemented even in organizations that don’t have an overall high level of analytics maturity.
In addition, machine learning can allow brand-new applications that weren’t previously possible. For example, companies that sponsor sporting events typically use stopwatches to track how long their logos appear on the screen during event coverage in order to get an idea of their return on investment.
But new machine-learning powered applications can automatically analyze the video footage in much greater detail. Clearly, this type of solution can be implemented no matter where you are on the overall analytics maturity curve.
Categorization by Technology and Usage
The traditional maturity curve obscures some big practical differences between the technologies on the left and right of the chasm line. When it comes to thinking about your company’s analytics maturity and how best to move forwards, it’s useful to think about the technologies in a different way.
- Analytics powered by humans for humans
- Analytics powered by math for processes
- Analytics powered by math for humans
- Analytics powered by math for human interaction and autonomous systems
Analytics Powered by Humans for Humans
On the left hand side, descriptive and diagnostic analytics is “traditional analytics,” powered by humans for humans. Experts collect relevant data, translate it into meaningful business terms, and expose aggregated information to business people through reports, dashboards, or data discovery interfaces. Those users then make decisions based on what they’ve seen and make the appropriate changes in the business.
All of these steps are typically a separate process from the operational activities of the organization, with integration only at the user interface level.
For example, marketers might have a dashboard that shows which campaigns are the most successful. They would then act on those insights by changing something in the next campaign planning process.
Organizations often struggle to determine in advance the return on investment of these types of analytics projects because “you don’t know what you don’t know.” It seems reasonable to assume that better insights will lead to more efficient processes and new opportunities—but until you’ve implemented the systems, you can’t determine exactly where they will be.
Analytics Powered by Math for Processes
On the right hand side, predictive and prescriptive analytics are powered by math for processes. This requires a very different approach. For example, instead of making the data intuitive for humans, it needs to be prepared in a format suitable for processing by algorithms. This typically means denormalizing and flattening the data, and retaining as much detail as available. The process takes longer, since large quantities of high-quality data are required to build and train the predictive model.
These advanced analytics tools are getting easier to use—by 2020, Gartner believes that more than 40% of data science tasks will be automated, resulting in increased productivity and broader usage by citizen data scientists.
They can be used for sophisticated one-off analyses, but more often the goal is to create a predictive model that is then automatically applied to new data as part of a automated business process. For example, when a bank uses a predictive model to check the possibility of fraud each time a customer uses a credit card, or when a retailer determines what product to try to cross-sell you when you reach the cashier’s desk.
The return on investment of these types of project is usually easier to determine in advance, because the decisions to be automated are associated with known costs or opportunities.
Recent advances in powerful machine learning, available at lower cost than ever before, have resulted in a huge surge in this type of analytics use. It’s increasingly being delivered as a standard part of almost every business application and process.
Analytics Powered by Math for Humans
Advanced analytics powered by math for humans is becoming a key part of new interfaces designed for business users. In this case, the algorithms are used to augment traditional analytics processes. For example, they can be used to help correlate and cleanse data more intelligently, automatically spot what’s interesting or unusual about the data studied, group data, etc.
These types of “smart discovery” features are increasingly a direct part of commercial analytics software.
The advances in human-machine interaction also allow users create queries and ask questions using natural language interfaces—for example, “what is the budget vs actual for my department?” By next year, Gartner believes that 50% of analytics queries will be generated using search, natural-language query or voice, or will be autogenerated.
Analytics Powered by Math for Human Interaction and Autonomous Systems
Advanced predictive models also allow analytics powered by math for human interactions and autonomous systems—for example, self-driving cars that use constantly-updated algorithms to navigate with little or no human intervention.
Arguably, both of these cases aren’t really “analytics” at all, since although advanced algorithms are a fundamental part of the system, only the results are exposed to the end-users, with little or no control over how they were obtained.
Next Steps: Analytics Maturity Isn’t About Technology
By thinking about new analytics technology in these four categories, it’s much easier to see how and where predictive analytics could be used in the organization. Above all, there are now very real opportunities to implement powerful predictive technologies as a seamless part of augmented business intelligence, or embedded into business processes—even if you’re still struggling with traditional business intelligence.
But, of course, your analytics strategy should start with your business goals. The biggest barriers to effective use of information are people, processes, and information culture. If organizations really are interested in analytics maturity, their frameworks should be primarily based on these factors, rather than the underlying technology used.
With Great Power Comes Great Responsibilities
Finally, the new powerful opportunities also introduce new responsibilities. Using predictive technology essentially means that you are outsourcing decision-making to algorithms. It’s important to make sure that you understand the ramifications for your overall information governance framework.
For example, the AI provisions of the new European General Data Protection Regulations (GDPR) state that you have to be able to explain in detail any differences in treatments between different groups of customers—and this may limit the types of algorithms that can use.
Organizations such as Partnership on AI are a good source for tips on ethics and other considerations when you’re moving to use more predictive technology.
For more on topics like theses, check out the rest of our blogs in the Machine Learning Thursdays series.
I’d like to hear what you think. Connect with me here or on Twitter at @TimoElliott