When you get to the root of it, understanding what it really means to be data driven is not as clear as it could or should be for many. At Dataiku’s Everyday AI Conference in London, Robert Bates, Head of Decision Science at Curry’s, shared his personal experience of understanding the meaning and power of data-driven decision science through the creation and evolution of data-centric processes at Curry’s. In this blog, we highlight the themes of his talk.
It’s About More Than Making Models
Contrary to popular belief, making the most of data at an organizational level is not just about making more and more models to throw around. Being a strategically data-driven organization means that, at the core, your teams are finding ways to enable business-savvy decision makers with the insights gleaned from both project-specific as well as enterprise-level data.
Over time, as new and better technologies have entered the space, one thing about data science has remained consistent — the need to derive insights from data for informed decision making. This sentiment however, has often been misconstrued, with stakeholders throwing projects over to data science teams without care or attention to necessary briefing and follow through. Queue the dreaded saying…“Run the numbers!”
Bridging the Gap Between Technical and Commercial Teams
To unleash the full potential of decision science, the expectation gap that exists between the work of technical teams and commercial-facing teams must be addressed. The traditional role of the data scientist has evolved. Working in silos simply won’t work anymore and even teams involved in deeply technical processes should be cognizant of the technology touchpoints that extend all the way to direct customer interactions.
Robert shared that one of the most sought after and valued individuals at Curry’s is what they refer to as a translator. A translator in this context is someone who applies systematic thinking to data processes, extracts manageable bits of information, and uses that knowledge to make implementation decisions at scale. These key individuals turn technical data science complexity into plain language for business teams to absorb and apply.
“We're after those translators. We're after that layer of people who are going to set the question. It's not just about looking at what the underlying data is. It's about combining that data with systems thinking — breaking that problem down into manageable chunks which you can analyze independently.” - Robert Bates, Head of Decision Science at Curry’s
The Goldilocks Rule and a “Put Into Practice” Approach
In addition to having the right hands on deck, a smart approach also matters for success. It is highly unlikely that the first model you build will be 100% accurate, so it becomes about determining the point of inflection. You have to pinpoint where the level of complexity and cost involved in perfecting a model will overtake the level of benefit of the model making it to production. Much like the tale of Goldilocks finding a bed, you have to be particular about choosing when, where, and what models to deploy whilst determining the level of risk you are actually willing to take on.
In the beginning it is often easier to identify areas for gain in practice rather than during the theoretical conjecture phase of projects. When a model is live, with the right framework in place, you can pinpoint the exact areas where adjustments are needed to capture ROI and mitigate high risks.
However, in order to successfully apply this approach, a sturdy foundation needs to be built out with clear targets to bolster decisions throughout the flexible adjustment period. Questions to ask include: What exact metric should improve? How often and how extensively will models need to be refreshed? What data sources feed the models? These things should be outlined and teams should be briefed before projects take off so that there is a rough idea of what the impact of implementation is going to look like.
Of course, things never go exactly as planned, but having an organized plan for what you want will make optimization much easier. If you aren’t clear on what you’re optimizing for, then you will find yourself stuck in the depths of your data instead of being driven by it. Things should be explainable from step one all the way through post implementation.
Good for Business or Just Shiny?
A positive side effect of bridging the gap between technical and commercial as well as pushing to define an operationalization structure is the differentiation of shiny projects vs. projects that truly ameliorate processes. As a rule of thumb and in reference to what was just touched on above, if you cannot explain to an end user how the data project generates impact, then it is unlikely that the project was truly “good for business.” The big picture matters because it is a reflective outcome of all the little allocations that you make to finally deliver your solution.
There is a lot of complexity and hidden information in data that can prove valuable when accurately extracted and applied, but it must be centric to overall responsiveness to have meaningful influence on business processes long term. If there is not going to be a notable movement resulting from a project, then there is no purpose behind going through with it. This is where unleashing decision science holds the power to prevent loss of resources and time.
“By stepping back and breaking down the problem, we can give a coherent solution, which is far more powerful for the business than that singular model.” - Robert Bates
The End Goal of Decision Science
At Curry’s, a prime example of data and decision science in action is the application of demand forecasting models where insights have been applied to sales market analysis. With patterns and key metrics clearly visible, it became simple to ask important questions and apply ample evidence alongside change proposals. Single problems and projects could be strategically and holistically diffused to the business life cycle at large.
“All these parts are connected and we want to talk to the business as a coherent whole, understanding the linkages, what drives what, how does that problem link into others in the business, and what are those primary and secondary drivers of change?” - Robert Bates
It has, at the end of the day, been an entire transformation of business philosophy that began with examining the minutiae of improving the processes of the data science team and the evolution is still in its infancy in many respects.
As tool stacks grow and technologies advance, continuously evaluating the decision processes to ensure that all potential value is being unlocked in the most efficient, systematic way is top of mind for leaders at Curry’s. In this way, the end goal of decision science is actually constant improvement.