Analytics 3.0: How to Become Prescriptive
How to create a road map to move your enterprise from descriptive analytics to prescriptive analytics.
[Editor's note: The author is leading a session on Analytics 3.0 at the TDWI Conference in Orlando December 7-12, 2014. In this article, Krishnan explains how your organization can move from its current analytics program to adopting prescriptive analytics.]
By Krish Krishnan, CEO, Sixth Sense Advisors, Inc.
On a hot sunny afternoon in Chicago this year, I participated in a good discussion about where we as an industry are heading in our analytics journey. Do we know what this journey is and how to engage in it? The question was not whether the journey is needed but how to make this a successful one, especially considering that past analytical forays have been long and extended journeys.
As I kept thinking through the mindset of the discussion and wondering the about answers, one project I had just finished working on involved digital meters and prescriptive analytics that we built to measure and analyze power usage came to mind. In this project, consumers were given historical and trend analytics every week about the usage of power. Predictive consumption models were presented to show customers how they could conserve power and reduce their bills. The power company wanted to change its approach from being predictive to becoming prescriptive, which meant that they had to understand the customer and their sentiments well enough to create a successful strategy to become a partner with the customer, creating trust and gaining more confidence.
The goal to become prescriptive led the utility company to ponder a key question: how could it accomplish its goal? They still needed to continue to use the current-state analytics (including predictive models) as they added sentiment analytics, social analytics, and competitive analytics into the model framework to create the best outcome and influence.
To become prescriptive from an analytical perspective is not a small step. As we started to lay the foundation for the shift, we discovered we needed to create a road map from descriptive analytics to prescriptive analytics that included descriptive (business intelligence layer) analytics, predictive analytics (including all models), and finally prescriptive analytics (predictive plus recommendations). The first task we executed was to build this road map and associated checklists, which helped us identify what the outcome from each analytical phase would lead us to (in terms of the next step in the process) and what the key items for success were. Let me provide a simple scenario that will show how we created a road map.
If we read an energy bill in its current state, it tells us the end date of a billing cycle and how much power we consumed. It provides a comparison chart of how our neighbors used power and if we are in line or out of line with them. In a smart meter system, we want to provide this information to a user on demand. We will be descriptive and we can also be predictive and prescriptive because the interaction will be near real-time.
The question is: what are the rules for becoming predictive and prescriptive? Who manages these rules and their outcomes, and how can we be governed and advisory at the same time? If the customer is unhappy, how do we provide more solutions?
In a traditional road map, we would handle these questions as project requirements. The difference here is that some of these requirements need to be discovered with machine learning and mathematical algorithms and presented to rules engines to make decisions about what to present or provide as data points to the customer. In this scenario, which was a learning experience for our large team, we created a section in the road map called machine learning. It was positioned in the first phase of the road map as a set of algorithms we could use. Each of the algorithms had a playbook that described the inputs, the processing, and the expected outputs.
By doing this, we created a team of internal experts who could work on this phase. Because 90 percent of the team consisted of on-site employees, facilitating the process was easy.
As a result of our work, we provided a bottom-up perspective on the road map process designed for success and easy adoption. Training and change management steps became more aligned, and we learned a few important takeaways that an organization can embrace.
- Don't get lost in technology
- Don't get lost in systems integration
- Communications technology teams need to be partners and collaborate
- A well-defined governance structure is needed for sustained success
Here is how we defined our road map.
For smart meter predictive analytics, we needed to:
- Understand the social and economic perspectives of the population
- Understand the local government policies and support
- Check how many customers have moved to smart meters
- Verify the near-real-time and historical power consumption status
- Graph the current state and prior state to show the savings and predict a future state for the customer and neighbors who have moved to smart meters
- Predict the future state
For smart meter prescriptive analytics, we needed to:
- Understand predictive model outcomes
- Understand current state descriptive analytics
- Execute a recommendation-filter algorithm using customer segmentation, demographic, and economic criteria
- Combine outcomes and execute business rules using prescriptive models
- Generate visualization and share prescriptive analytics recommendations
This mode provided us a starting point and helped us accomplish the initial proof of concept and proof of value phases without many glitches. Now we are moving well along this path with more specifics added to each phase for each algorithm, which is managed through the process of "analytical" governance created on the same realms of data governance except that we were KPI driven.
As we completed the initial phases, we realized that with prescriptive analytics, things would be executed differently. Algorithms were unleashed on data with only minimal rules telling them what to do, and the algorithms were programmed in such a way that they can take over and adapt based on changes in established parameters. With algorithms optimizing automatically, their ability to predict the future becomes better over time and creates the prescriptive platform from which decisions can be suggested. This learning has become a platform for innovation and governance aligned and combined for success.
In fact, this is the foundational concept that drives a new course at TDWI. In this course, the discussions are divided into:
- Foundations of analytics: What are the different types of analytics and how have we implemented these analytics
- Predictive and prescriptive: A case study on healthcare analytics and Google Car
- Organizational capabilities: How to create a road map and a team for this journey, and more about the data scientist team and its skills
- Infrastructure: Questions to explore, such as: will we need Hadoop? What are SAS, IBM, SAP, and R looking at from Hadoop and beyond?
- Mahout: What do we have in algorithms and where do we go next
The outcome of the class is to open the new world of prescriptive analytics, a new look at the road map and the journey, risks, and benefits along with case studies and examples for each analytical phase. The biggest benefit of the discussions in the class will be oriented towards prescriptive analytics, how to improve the efficiency and performance of the algorithms, and models as enterprises.
Krish Krishnan is the CEO of Sixth Sense Advisors, Inc., an independent management and technology consulting organization, and a TDWI World Conference session leader. He can be reached at firstname.lastname@example.org.