Data and particularly machine learning are often thought of as silver bullets that can solve every business challenge. Subsequently, many businesses have or are in the process of building the necessary infrastructure and talent to deliver on the lofty promises that they believe data and machine learning can provide. This is the right move — there’s no doubting that analytics, both descriptive and prescriptive, have the power to completely change the dynamics and efficacy of a business. Unfortunately, many companies struggle to reach the lofty promises that Data and Machine Learning represent.
As most of us know, a truly successful data science or analytics project is one that is inextricably tied to business outcomes and reaches across multiple teams, organizations, and domains of expertise. Building a model or stitching data together, while complex in their own right, are often the ‘simple’ part of operationalizing a data stream or algorithm so that it has a veritable impact on the business and drives a strategy forward. It’s incredibly rare to see a data project land on that level and it’s for that reason that many data projects are seen as a failure or go unnoticed altogether. Large, highly technical companies are not exempt from this challenge. I’ve witnessed first hand a Data Science team at one of the world’s largest technology companies that spent years to complete a model but was unable to effectively deliver that model to the end-customer in any impactful way.
The Complexity of Data Operationalization
How do we solve this problem? The preeminent answer to this question would have you believe it’s due to a talent shortage — i.e. if we just had more experts or the ‘right’ experts to solve the ‘data’ problem, we could solve it. This strategy is what many organizations turn to for difficult challenges. In scientific circles, this idea is called ‘local search,’ (Epstein, Range, p.173 — mentioned in my previous article) the reliance on specialists in a single knowledge domain who utilize the most well-worn methods to solve a complex problem. While there’s absolutely nothing wrong with specialization, the challenge of deploying descriptive and prescriptive analytics across multiple organizations or an entire company spans multiple domains of business knowledge, teams, and interlaces between business strategy and finance, data engineering, data science, and operations. With so many variables at play, the complexity of the ‘data project’ reaches dizzying heights. Renowned behavioral scientist Daniel Kahneman helps us understand that problems of this nature are “wicked.” Mr. Kahneman defines a wicked problem as “a problem where the rules of the game are often unclear or incomplete, there may or may not be repetitive patterns, and they may not be obvious, and feedback is often delayed, inaccurate, or both.” (Epstein, Range, p.21) The challenge of integrating and deploying data-driven solutions and products is filled with obstacles that are unique to each organization and require an ability to laterally connect knowledge and expertise across domains and business units. Psychologists call this “Analogical thinking,” the practice of “recognizing conceptual similarities in multiple domains or scenarios that may seem to have little in common on the surface.” (Epstein, Range, p.103)
Analogical thinking provides the toolset for how to approach the challenging task of operationalizing data across the full breadth of an organization. I’ve utilized analogical thinking in my own work to come up with a framework for solving this problem — one that I call the Data-Driven Operating Model.
The Data-Driven Operating Model
The Data-Driven Operating Model is a framework for building the processes and infrastructure that enables companies to operationalize data effectively. I’ve primarily used this framework in the context of the marketing organization, but I believe the principles here will apply broadly. The goal of the Data-Driven Operating Model is not to provide a one-size-fits-all solution, but rather a framework for creating your own plan to organize people, processes, and data infrastructure at your business to deploy highly effective descriptive and prescriptive analytics solutions. The model stands on 5 pillars:
- Customer Intelligence
- Business Intelligence
- Data Translation