Is Digital Transformation Strategy the Barrier to Data Science Maturity?
Digital Transformation has a lot of hype behind it but do the strategies support adopting machine learning?
Vin Vashishta | Originally Published: February 23rd, 2021
I have spent 6 years building Data Science teams and capabilities in startups to Fortune 100 companies. I have built and brought machine learning products to market with 9 figure revenue. At every company I work with, digital transformation strategy has built a culture of Analytics and called it Data Science.
Companies have had to unlearn all those behaviors to monetize Machine Learning and stay competitive. The wave of underperforming Data Science projects is a direct result of legacy strategy struggling to adapt. Digital transformation is the right goal, but a 20-year-old strategy cannot build a foundation for achieving it.
The strategies behind digital transformation do not support adopting Data Science and Machine Learning. They are software automation centric. There is no competitive advantage or innovation. It has become a shopping list rather than a strategy.
Digital transformation is an unnecessary step, often creating barriers on the path to modernizing the business model. The path from legacy to data is a straight line. Abandoning digital transformation is more efficient than putting it between the two points.
Step 1 for digital transformation should be the research phase of the Machine Learning Lifecycle. Research results in artifacts the business can use to inform strategy planning. Once a supported strategy has been built, each subsequent research iteration builds competitive advantage. Unique data sets and models create a high wall around the business model because they are difficult for competitors to reproduce.
Cost savings are immediate. Digital transformation creates a zero-sum game around increasing its share of budget allocation. Research is much less expensive and results are delivered quarterly. Returns are driven by research artifacts; tangible outcomes of adopting modern strategy. These artifacts improve profits through products and efficiencies.
Digital transformation has a disconnect between strategy and execution. There is no translation layer between business model innovation and tangible artifacts. Anything can be substituted for success when there is no SUPPORTED connection between business metrics and project metrics.
Digital transformation is riddled with opinions in search of supporting data. The connection between strategy and outcomes is proven after the fact. There is data involved but the process is not rigorous and does not stand up to peer review.
The research process:
The strategy planning process is a peer review of the thesis by senior leadership. The research phase creates tangible, reliable results.
Digital transformation stops at the hypothesis step and fails to build a supported thesis. In the research field, we call this an overextension of the data. Someone looked at the data and formulated an opinion. The result is an analytical or descriptive model which is their best attempt to explain what has happened in the past.
Those models cannot be used to predict or prescribe. There has been no validation to support extending the analytics for those uses. From a strategy planning and forward-looking perspective, analytical models are useless.
Digital transformation asks the business to test out their historical facing hypothesis by basing the 3-5 year plan on it. The business becomes the experiment which explains why digital transformation is such a hit and miss process. Digital transformation mislabels analytics as Data Science, creating a barrier to getting value from the technology. The result is overly expensive Analytics with the Machine Learning label slapped on.
The advantage of Machine Learning is the ability to build models that simulate business systems, processes, marketplaces, etc. Research happens rapidly based on iterative data gathering and experimentation. Bad assumptions and opinions get tossed out early. Experiments are designed to prove a model accurately simulates what we claim it does. Proofs support extending the model to predict and prescribe.
We expect to be challenged. We expect review, questioning, criticism, skepticism, and expectations of extraordinary proof for extraordinary claims. There is no higher bar than science and that is the foundation of our research. If leadership says, “I do not believe it” we propose an experiment. Will this experiment and these results prove it? No? OK, how about this one?
Researchers are used to this. It is our job to come up with creative experiments to prove or refute a hypothesis. We constantly work to support and improve our understanding of the systems we model.
Research is a gated, managed process. Researchers are used to working with set budgets, timelines, and deliverables. Produce results and publish artifacts or perish. That is the research agenda. Digital transformation’s agenda is to perpetuate itself.
The research approach to strategy is proven out in industries like FinTech. It is the reason hedge funds are willing to pay Data Science and Machine Learning Researchers extraordinary salaries. Artifacts support their understanding of business, consumer, and market performance. They provide an advantage worth billions to those who are setup to leverage research artifacts.
The good news is making the move reduces expenditures. Reorganization of the Data Science Team is required. Upskilling existing talent and hiring researchers are short term projects. It is knowledge intensive but not resource intensive.
Research answers immediate questions like:
Research driven transformation is both short term, low business disruption and long term change focused. Machine Learning is designed to automate complex tasks around transformation. The goal is efficient transformation which requires low impact.
Integration teams ask me, “How am I going to consume this?” My response is, “How do you do it now? What tools do you use now?” The technology does not matter. Customers and business users should barely notice the technology.
Most of what we use is open source. The results are complex model driven but the delivery is user centric. Human in the loop Machine Learning is intentionally not disruptive to operations.
Impactful change without disruption to operations is a new concept and a departure from forklifting strategies. Research is a simpler path to Data Science maturity.