Analytics & Big Data
Mar 20, 2019

Data science – three steps to transform data into action

Today, data presents enormous value to organizations. This includes structured data that can be readily processed and analyzed by traditional means, as well as unstructured data that lies in the text and images of documents, emails, contracts, and invoices, which companies can now harness through natural language processing, computer vision, data extraction, and data classification technologies. When analyzed, these datasets can show opportunities to improve customer experiences, add new services or products, adjust existing operations, and drive business growth.

The process for deriving these insights is data science, which includes three transformative steps: identification, fusion, and data-to-action.

Data identification – finding the right data

After identifying a high-value business challenge, the first step is to determine which data can be useful. It could be operational data on how the business is running. For example, in the manufacturing industry, a data scientist can look at the movement and procurement of product components as they move through the different points in assembly. On the customer-facing side, there could also be data on what products are selling well, with an analytics goal of understanding relationships among customer segments.

Deciding which data to leverage should be anchored to a strategic goal and reason for analysis. As mentioned above, it may be to improve customer experience, launch new products, or transform an existing process. For instance, a photo- and video-sharing platform may want to know which of its filters is the most popular among users. Every time someone takes a photo, it becomes data that the platform can review and analyze in order to determine the next wave of filters to develop and roll out.

Sometimes a company may want to look to external sources of datasets to get a well-rounded picture of its business or market. It can purchase outside data or mine the information from an open community, which may have data from competitors. The more data that data scientists have to leverage, the more accurate they can be in their subsequent analysis.

Data fusion – putting it all together

The purpose of the next step, data fusion, is to find the relationships between different datasets. What data reinforces each other? What data provides contradictory information? What is immediately actionable? For the fusion to work correctly, the data must typically go through cleaning – imputation for example – which helps account for blank or missing values.

This is often an iterative process, requiring several passes of fusing and cleaning before a company can get unified datasets that are aligned to the strategic goal set out in the identification step. The result is formidable layers of data, such as associations among stock keeping units (SKUs), product attributes, sales, and customer preferences, which can be mapped out to obtain actionable insight.

Data-to-action – sensing and signaling

With fused layers of data, data scientists can use statistical analysis to uncover things like, “What are our customers really concerned with right now? Where are there gaps in our operations that can be filled or completely transformed?" These insights provide "a-ha" moments, as well as ongoing awareness to what is happening internally and externally at leading companies.

Many are starting to use machine learning to find new opportunities to take action. Just like a GPS reviews real-time coordinates to guide drivers to turn and move toward a destination, machine learning can review enterprise data, recognize signals, and provide recommendations for a data scientist to act on.

Sometimes, machine learning helps observe subtle clues and inferences that a human may otherwise overlook. For instance, there can be insight around customer preferences tied into data around demand in the supply chain. With machine learning, data scientists can detect these things earlier and proactively get signals out of their data. By design, the machine will “learn" by working with more data, and its algorithms and recommendations will become smarter and better over time.

As more organizations recognize the value of data, data science is critical to achieve strategic goals. Companies will need to know what data they need, clean and fuse the datasets, and run statistical analyses to unlock the underlying, interrelated information within their enterprise data. Machine-learning algorithms will prove to be powerful tools to get to the pertinent signals and transform data into action. And a hallmark of artificial intelligence is just that: action, which is enabled by unprecedented observation capability via machine learning. In this way, data scientists help advance the state of the art to extend what can be achieved across each industry area.

About the author

Armen Kherlopian

Armen Kherlopian

Ph.D., Chief Science Officer & VP, Analytics & Research

Follow Dr. Armen Kherlopian on LinkedIn