Three “musts” to get value from data science innovations in finance
All financial firms seek innovations that will provide a competitive edge, and the realm of data science and analytics offers fertile ground, driven by the rapid proliferation of ‘big data’ and machine learning techniques into mainstream business. Gartner predicts that by the end of 2024, 75% of enterprises will shift from piloting to operationalizing AI.
Most management teams acknowledge the importance of innovation in a general sense, and business theorists spill much digital ink attempting to formulate best practices for fostering innovation within enterprises. However, our work with clients on data enablement projects leaves us convinced that the data domain requires special considerations in order to maximise potential return on investment.
Data science must be targeted at strategic priorities
Common wisdom holds that most great innovations originate organically – in data science cases, from a data scientist discovering ‘signal’ in a dataset which can be used to develop a predictive model, which can then be used to automate or support decisions.
The underlying assumption, however, is that the data scientist is looking in the optimum place to begin with. Data science specialists are a scarce resource, and organisations face a choice of where to direct their focus. For example, they are prevalent in traditionally ‘quant’-oriented, front-office trading teams, but rarely seen working with back-office functions such as finance or IT operations, which are often viewed as procedurally-driven and an unappealing target for data science work. And yet, such functions reflect some of the most significant cost drivers for financial firms. Application of machine learning to optimise seemingly-mundane processes in IT servicing, financial reporting or similar could, in some cases, be far more impactful to businesses’ bottom lines than marginal improvements in market or customer behaviour analytics.
The key is that the deployment of data scientists on exploratory work should be demonstrably aligned to the priorities inherent in the corporate strategy. To do this, the strategy must, itself, be intelligently targeted – a strategic directive to ‘reduce costs’ or ‘bring in more revenue’ means little without consideration of how this should be achieved. However, data analysis can itself play a vital role in quantifying potential upsides and driving prioritisation. With this in place, data product owners should be challenged to show that their teams’ focus reflects conscious pursuit of the outcomes which have been determined to be of maximum strategic value.
Data science must be embedded within the business
The clear consensus is now that data scientists/analysts/engineers perform most productively when integrated within close-knit multi-disciplinary teams.
The benefits of such integration are both direct and indirect. Direct benefits accrue from improved communication – data scientists working side-by-side with their business customers can be far more responsive to business needs, reducing the friction inherent in handoffs between siloed business and analytics teams. The DataOps concept (steadily growing in popularity) reflects this idea, drawing heavily on the observed benefits of Agile and DevOps techniques in software engineering.
Indirect benefits accrue from the increased sensitivity that data scientists in embedded teams develop to the business context and priorities. This does much to bridge the gap to (and allow finessing of) the top-down strategic direction described above.
Data science products must be rapidly operationalised
A statistical model in the form of a piece of Python code or a Jupyter notebook delivers, in of itself, zero value to an organisation. The value accrues when that code is deployed in a calculation engine which feeds automated scoring or decisioning processes, and/or presents metrics to a user who acts on the insights it provides. The resulting improvement in business decisions embodies the real business value, in the form of reduced risk and improved financial performance.
A key measure of the commercial productivity of data science is therefore how efficiently the organisation can translate an idea for a data science project into a piece of code running on an operational system – going ‘from concept to cash‘. This goes beyond the domain of data science into data and applications architecture, governance, technology delivery and business process engineering, and necessarily involves a much wider pool of stakeholders across the organisation.
We generally find that while larger organisations have access to broader and deeper data science expertise, they can be less nimble than smaller, less mature organisations at operationalising the insights that data science provides. Predictably, forces are at work to close the gap in both directions: data science is becoming more democratised through the provision of more managed offerings (e.g. by public cloud providers), whilst larger organisations increasingly seek our help to implement more agile ways of working.
Key for an organisation seeking to mature in this place is to critically assess the journey by which data science initiatives progress from initial concept to production implementation, and seek to eliminate any unnecessary friction from that journey.
Whilst leadership teams increasingly acknowledge the role of data science to their corporate agenda, they should avoid over-simplifying the problem to one of hiring talented experts and just leaving them to get on with it. Realising net return on data science investment requires careful strategy and on-going attention to detail. Organisations which recognise this stand to gain a significant competitive advantage as the rapid digitalisation of financial services continues.