Data Science: Proving the Concept
The popularity of data science has never been higher, both to solve business problems, as well as for academic study. Searches for ‘data science’ reached record levels in early 2022, almost double where they were less than 12 months prior. The advent and rapid growth of data science and artificial intelligence (AI) in modern business and society has been termed the ‘Fourth Industrial Revolution’ (4IR) by the World Economic Forum and it’s easy to see why. More and more of our everyday activities are being augmented by intelligent applications of data: the ways in which we watch TV and consume entertainment; book tables at restaurants and the taxi or ‘ride-share’ that gets us there; the way we buy groceries, heat our homes, book holidays and even socialise with friends are all, now and increasingly, AI-enabled.
More organisations are beginning to explore how data science can be used to drive improvements in their operations, processes and decisions. Many people in strategic and executive roles have heard of the term ‘data science’ and understand that it’s a way to leverage data in new and innovative ways. They want to become ‘AI-enabled’, whether that means generating more revenue, finding and exploiting opportunities for operational efficiencies, increasing brand equity in highly competitive markets, mitigating business risk, or achieving and demonstrating regulatory compliance.
Despite this growing interest in data science and appreciation of its power, for many organisations it still feels out of reach: a future state or a perpetual ‘phase two’ that will probably, hopefully, one day happen. If this sounds familiar – don’t worry, you’re not alone. Data science is difficult. Solutions involving machine learning and complex statistical analysis can be hard to build and the assumption is that applying data science techniques to development comes with a high barrier to entry.
But it doesn’t need to be that difficult.
What is a Proof of Concept?
The common approach of ‘where can we apply data science’ isn’t necessarily best, especially for businesses where opportunities for improvement and more fundamental challenges take precedence. You wouldn’t wander around at home with a hammer looking for a nail to knock in, you’d spot a problem and go through the toolbox looking for the right tool for the job. Likewise, it doesn’t make sense to look for ways that data science could be applied; the better approach would be to try out a few approaches to solution development and add time, effort, and complexity as needed to add further benefit. This is where a Proof of Concept is most valuable: exploring whether a problem can be solved with data science or other advanced analytics techniques and, importantly, whether this adds more value than a more traditional rules-based or logical solution.
Data science, like any other scientific discipline, uses experiments to improve understanding. The Proof of Concept is a scientific experiment where a hypothesis (“can we solve this problem using data?”) is tested in a controlled and reliable way (“can the outcome be predicted based on data with a higher degree of accuracy than an alternative?”) to generate insight (“how can we apply this model to realise a real-world benefit?”). In the pharmaceutical industry, for example, the ‘alternative’ would be a placebo. In a business context, the alternative is the decision that would have been made if the model didn’t exist, usually based on descriptive analytics or interpretation of data in spreadsheets.
Benefits of running a Proof of Concept
There are many benefits to running a small-scale Proof of Concept before committing to development of a full-functional solution. These can be grouped into three main categories: risk reduction, value estimation, and functional discovery.
Risk Reduction
How much time, effort and budget would be required to take an idea from zero to complete? What assurances do we have that our commitment will solve the problem? How do we know whether it would work at all? These are all important questions to ask whilst exploring whether data science can be applied to a business problem and each one represents a risk. The Proof of Concept de-risks the project by reducing commitment to the smallest possible level, taking a fail-fast approach to solution development.
Value Estimation
What kind of benefit will we see if this idea is developed into a functional solution? How much return on investment should we expect? What can we do to maximise the potential for success? The Proof of Concept is typically a scaled-down version of the ideal solution that produces clear and quantified results. These results can be explored and combined with other information around finances, operations, and processes to estimate benefit. For example, can we predict how much customers are likely to churn in the next 6 months? What is different about these customers compared to those more likely to stay? What kind of interventions or systematic changes could we implement to minimise the risk of these customers churning? If this is X percent successful, how much revenue would be generated that we otherwise would have lost? In this way, the results of the Proof of Concept can be used to quantify how much ROI can be expected from developing the solution fully and in production.
Functional Discovery
There is value in the Proof of Concept outside of the actual results: the process itself reduces the need for detailed exploration and discovery that would have been necessary to jump straight into a fully functional solution. What data do we need to make this work? How does it fit together? Are there any data quality issues that need to be addressed? What kind of transformations do we need to make? We’ve found that answering these kinds of questions as part of a detailed discovery typically takes around 10-20% of the total effort required for a data science project. The Proof of Concept includes exploration and discovery, answering a lot of the key questions before committing to productionisation. This means that, at the point of developing a functional solution, much of the initial exploration and functional scoping has been done, leading to faster development time and reduced effort.
Summary
Data science is becoming more and more commonplace in our personal lives, but using data science approaches to problem solving can feel daunting to business leaders and executives due to concerns around skills and capability, technology dependencies, budget, effort, and uncertain outcomes. Our Data Science Proof of Concept service aims to make advanced data analytics techniques available to organisations at all levels of data maturity, and can reduce the barriers to entry, minimise risk, demonstrate value, and provide a plan for solution development in production for ongoing benefit. Discover how data science can benefit your organisation and book a free consultation with one of our experts today.
Blog Author
Adam Andrew, Innovation Manager, Simpson Associates