Data science has become mission-critical to many enterprises. Companies in the United States will spend billions on third-party audience data in 2018 and even more on third-party solutions to use that data, as they move toward becoming model-driven, Internet Advertising Bureau projections indicate.
It’s not just a matter of harvesting data and then crunching it; companies have to map data against their business model in order to get the most out of it.
CRM and sales software vendors have begun offering products that help sales and marketing teams adhere closely to their company’s model.
“It’s an existential issue,” said Mac Steele, director of product at Domino.
“According to a recent McKinsey survey, companies that can consistently deploy models in production have a 7.4 percent profit margin advantage over their peers,” he told the E-Commerce Times.
Such model-driven companies develop more breakthrough products while simultaneously and constantly driving iterative improvements in their core operations and customer experiences, Steele said.
“Companies like Amazon, which put models at the center of their business, are accelerating past competitors,” he observed. “It will get harder and harder for laggards to catch up.”
Getting Started With Data Modeling
Modeling involves predictive and prescriptive analytics, also known as “advanced analytics,” said Doug Henschen, principal analyst at Constellation Research.
“You’re creating models to predict out into the future what’s likely to happen,” he told the E-Commerce Times, “and with business context, how you might react to that prediction to get to a better outcome.”
Companies have been adding third-party data such as demographic, psychographic, weather and industry data, to account for outside influences and get to more accurate models, Henschen said.
They’ve begun using machine learning and deep learning approaches that create models based on the data itself as data stockpiles have grown.
While models have been used in focused areas such as lending, claims processing and other high-value areas, applying this data science expertise throughout a corporation requires developing “dozens, scores or even hundreds or thousands of models,” Henschen noted. “That requires automation and robust model lifecycle management capabilities.”
How to Become Model-Driven
The biggest challenge to moving toward a model-driven business is developing the data engineering and data science talent required to drive and maintain it, Henschen said.
“There is far too much data, and there are far too many opportunities to turn that data into business value, but there are far too few data scientists,” observed Don Schuerman, CTO at Pegasystems.
“In order to democratize the power of AI and analytics, we need to pass the laboratory tools and create model-driven factories designed for a savvy businessperson, not just mathematicians and programmers,” he told the E-Commerce Times.Pegasystems’ products leverage omnichannel artificial intelligence that lets users apply real-time analytics across marketing, sales and service, and “all of our customers take this approach,” Schuerman remarked.
Shortcuts like relying on out-of-the-box model-driven capabilities baked into commercial off-the-shelf applications, or analytical capabilities provided as a service in the cloud are better than simple reporting and visualization of historical information, Constellation’s Henschen acknowledged, but they “are not a substitute for building internal expertise and capabilities that can be applied across a company starting with its biggest business challenges.”
Model-Driven To Do List
The first thing companies should do when planning to adopt a model-driven approach is get buy-in, both from business leaders and IT, Pegasystems’ Schuerman said.
Build up a data engineering and data science team, Henschen suggested, and then get to repeatable processes and scaling up “by exploiting emerging automation and model-lifecycle management capabilities.”
Model governance and visibility are crucial, as regulations sometimes require transparency in model-driven decision making, he noted. Also, the data science team “has to learn how to streamline the time-consuming challenges of data management and iterative data movement and model testing, so they can focus on innovation with, and exploring, cutting-edge modeling techniques such as learning.”
Take an open approach to infrastructure, Scheurman advised. “No tool is a silver bullet; tool agility and governance is the most important objective.”
However, openness shouldn’t be adopted at the cost of control, he cautioned. “A proper approach enables one to know exactly which tools and packages are used on each project and at each step of the lifecycle, so all experiments can be reproduced.”
Perhaps the biggest challenge to overcome in taking a model-driven approach is the tendency of companies to view data science as a technical skill rather than an organizational capability, Steele said.
“Hiring more people and buying more tools won’t get organizations to where they need to be,” he said, noting that there are four major challenges to becoming model-driven:
- Static infrastructure — the infrastructure data scientists work with is too static for their constantly evolving needs;
- Work methodology — the way data scientists work is optimized for the individual, so it’s very hard to create a culture and process that enables collaboration and knowledge sharing;
- Production issues — It’s very difficult to get models into production, and it’s even harder to get feedback on how a model is working or if it is even being used; and
- Model liability — Models have to be validated and monitored to ensure they keep up with changes in the real world, and that they’re being used properly.
Companies should “more tightly link development and production across people, process and technology,” Steele advised, and get data scientists to work collaboratively on projects from the beginning.