From Big Data to Big Insights: How to Make Better Business Decisions with Predictive Analytics
Extract value more rapidly by using a structured, four-step approach to move advanced analytics closer to the data.
Although many businesses tout «data centricity» as a critical success factor, most only realize a fraction of the value they could be getting from their data. Investments in data acquisition, storage, and analysis often fall short because most businesses lack the tools and techniques needed to adopt a successful data-driven strategy. Hiring more data scientists and other scarce resources to perform descriptive analytics or ad hoc analyses may help with reporting and business intelligence, but it doesn’t provide insight into what might happen if a business takes a specific action, nor does it provide a prescriptive path for what the business should actually do.
As FICO’s vice president of analytics, some of the challenges I hear most frequently include:
– We’ve collected lots of data but haven’t found a way to put it to use.
– Accessing and wrangling the data is too difficult and time-intensive.
– We have a team of data scientists, and although their findings are interesting, we haven’t found many real, practical uses for their insights.
– Our machine learning results are hard to understand and implement.
– It’s easy to validate what we already know, but how do we discover new, previously unknown insights that deliver measurable (even disruptive) value to our business?
– Acquiring data is expensive. We cannot justify the cost unless we can clearly show the ROI.
To overcome these challenges, businesses need both the right methodology and the right tools to execute that methodology. The real key is to find a cost-effective, repeatable way to add intelligence to the volumes of data that already exist within the organization. One way this can happen is by using data lakes and streaming data sources from anywhere inside (and outside) the organization to operationalize data-driven actions and support real-time outcomes.
By moving analytics closer to the data, you can extract value more quickly. Here are four steps today’s innovators are adopting to streamline their analytics supply chain.
Step 1: Start with the objective
What business challenge or objective is most pressing? How have you historically used data (and analytics, and even decision-making rules) to address this need? What metrics demonstrate success?
The more goals you can map out, the easier it is to gain political and organizational buy-in and streamline the project’s execution. Many organizations start with a single, well-defined problem that helps them fine-tune the process over time and improve it iteratively. Note that this mapping process will also help define resources — not just data scientists, but also domain experts and others in the business (even across functional boundaries) who can contribute to, and benefit from, the evolution from big data to bigger insights.
Step 2: Collect and enrich the data
Once you have clearly identified your objective, you need to access, understand, and assess the viability of all possible data sources. Many data sources, whether internal or external, might seem promising on the surface but need to be assessed for the breadth and depth of information they contain, as well as potential value. You may be starting with large volumes of raw data, so enriching the data requires machine learning algorithms to extract features (also known as attributes or characteristics) in a comprehensive and efficient way.
Step 3: Detect the signals
You’ll also need a comprehensive, efficient, and scalable tool that computes, prioritizes, and visualizes the valuable qualitative and quantitative intelligence often hidden in the data. Machine learning algorithms can rapidly provide a comprehensive and prioritized analysis of the various signals that indicate whether a given data set is valuable, including relationships, linear and nonlinear patterns, and outliers. Review these signals and ask yourself whether they make sense. This is when the intelligence of the machine needs to be challenged and explained by the intelligence of the human expert. This iterative collaboration of human and machine intelligence can yield new, breakthrough insights.
Step 4: Take action
This step doesn’t always mean implementing a predictive or prescriptive model. The business problem may call for the insights gained from this exercise to be used in a variety of ways, such as modifying product design or customer service tactics. Prescriptive models can go further — predicting what will happen and prescribing specific actions that will yield the most profitable outcomes. With multichannel customer engagement increasing in importance, predictive and prescriptive models can power consistently great experiences across the customer life cycle.
A Real-World Example
Once in place, these steps will add numerous insights to the decision-making process without slowing it down, yielding sustained competitive differentiation for the business.
Here’s an example of the steps in action. At FICO, we worked with one of the largest financial services firms in Latin America to identify data-driven insights that would help them deliver personalized offers to their best customers and build the value of those relationships. The firm needed to analyze behavior at both account and customer levels, targeting data-driven treatments to the specific level that would provide the greatest impact.
Historically, the firm had been amassing big data from multiple sources. There was a general expectation that more (and more diverse) data would help the company transition to a customer-centric business model and improve competitiveness. Yet the company’s managers knew this massive data contained far more «noise» than «signal» — irrelevant data rather than data that could help them make better customer decisions. How could they quickly extract meaningful data, and how could they identify the most potent insights for predicting customer behavior and create retention and cross-selling strategies based on these predictions?
The answer started with data exploration and wrangling. They eliminated data redundancies and resolved gaps. They structured and normalized data from different sources so it could be compared, combined, and evaluated for relevance. The team then enhanced the data, making its existing big data even bigger.
Machine learning techniques identified thousands of potentially valuable signals that were prioritized and refined to prepare the data for model development. These signals were decision-ready building blocks for designing predictive models and treatment strategies. They included «traditional» data variables (such as customer age) as well as transactional banking activity information (such as daily account balances or time between logins).
Machine learning can be powerful when detecting relationships across data sources. For example, we found interactions between certain types of banking products, customer age, and maximum balance that crossed over three data sources. Interactions across a network of disparate data sources are often missed by a human expert but easily discovered by machine learning algorithms.
Still, these discoveries are not usable if they don’t ultimately make sense to the human expert. Hundreds of thousands of these insights were then analyzed, allowing the bank to plan its next actions, which often occurred in the form of a predictive model or decision strategy. The business context for this particular project was to improve customer retention, but creating a repeatable and flexible «data that leads to insights which, in turn, leads to action» methodology made it possible for the bank to easily pivot its business goal to predictive credit, risky customers, or improving portfolio profitability.
A Final Word
Conventional thinking about machine learning focuses most of a team’s energy on the development of a model or strategy, but it needs to be viewed instead as a powerful accelerator of innovation, one that leads the business from data to insights to action. Organizations that adopt a structured approach to turning data into measurable value — that leverage the right tools to help them make the leap while also engaging their business to play an active, ongoing role — are the ones that will be best positioned to consistently benefit.
Fuente: https://tdwi.org/Articles/ by Chisoo Lyons
9 Distance Measures in Data Science
1. Euclidean Distance
We start with the most common distance measure, namely Euclidean distance. It is a distance
6 Ways AI is Transforming the Finance Industry
Scope of Artificial Intelligence in managing Finances
The sector has been witnessing unprecedented growth in term
Can You Trust Your Data?
The Data Trust Gap
In The State of Data Management — The Impact of Data Distrust, a rec