Five steps for building an in-house data analytics team

14 June 2018

Data analytics is essential for companies to be more impactful and increase their operational excellence. Having the internal capacity to utilise available data can even make or break a modern-day firm. French management consultancy Sia Partners has outlined five steps for clients keen to get ahead of the curve and build an in-house data analytics team.

Sia Partners data experts Jeroen de Laat – a Senior Manager – and Nabi Abudaldah – a Supervising Consultant – have put together a five-step guide for organisations seeking to build their own data analytics teams. The guide is inspired by substantial field experience, working closely with clients to build successful and stable teams that thrive over time. 

An expert guide

The first step is to start with data challenges, not data platforms. “Most software vendors and advisory firms will tell you to invest heavily in platforms before actually practicing data analytics,” says De Laat. This leads to huge spending on revamping the IT infrastructure before the team even understands the particular problems they are trying to overcome.

“Start by framing your strategic challenges and goals first,” advises De Laat. “Analyse where you are, and where you want to be as a company. Or rather, where your customers and other stakeholders are and where they would like to be.” Most importantly, try to refrain from any large (technology) investment decisions at this point. “More often than not, you don’t need an infinitely scalable and state-of-the-art technology solution to be successful.”Five steps for building an in-house data analytics teamOnce the strategic challenges are identified, the next step, says Abudaldah, is to find and organise internal talents. “Free up those individuals from their current day-to-day responsibilities to work on data analytics cases. Simply hiring externals to ‘do the job’ isn’t desirable because of their relative distance to your problems and the temporary nature of their employment.”

Leaders will typically have an idea of who in their organisation is passionate about analytics and has some degree of competence. “Design a small and mixed team of business and technology specialists and give them clear goals and timelines,” says Abudaldah.

Once the in-house talent is identified the third step is drafting in external support. “While some internal talents might not fully be able to get out of their earlier responsibilities for example, external talents can go full-speed ahead to make a successful start.” This rounded approach helps avoid false starts and brings fresh perspective to the team.

With the data analytics team set up and raring to go, the fourth step is to identify a suitable challenge to tackle first. “Don’t take on your most complex or anticipated data challenge first,” says De Laat. “Setting up a team is already a challenge. If you also take on a highly complex problem to solve, it might become rather a burden to any new team.”

“Start with low hanging fruit,” Abudaldah adds. “Don’t assume that small problems are too easy or even boring to solve. In practice, it always turns out the other way around.” Both consultants advise starting with small challenges which, handled successfully, have immediate impact. This way the team can under-promise and over-deliver, cultivating a solid reputation and the legitimacy to pursue larger and riskier projects at a later stage. 

As with many complex organisational endeavours, the final step is scaling and professionalising the new environment. “At this point, you know what challenges are valuable to solve, which talent gaps you have, what tools you need, and which processes are required to manage these new ways of working,” says De Laat. “This is a far better position to be in when it is time to professionalise, rather than spending big earlier but without a clear picture of the role of the new data analytics team.”


More news on

AI can improve operational risk management in banking

17 April 2019

Risk management is an integral part of banking. By taking financial risks, banks are able to generate the profits that are necessary to survive. Risk management aims to control this process by making potential losses more predictable. This makes the bank more robust to external fluctuations.

Whereas profits can be made by accepting certain financial risks, operational risk is intrinsically different. Operational risks only cause losses – financially in terms of bottom-line impact and non-financially in the form of for instance reputational damage. The consequences of operational risk events can have a large impact on an organisation and the financial system as a whole as experienced during the last financial crisis. It is not surprising therefore that operational risk is receiving more attention within the financial sector, with banks trying to minimise the operational risks they take, given the resources available while keeping in mind the strategic goals of the organisation.

In the past decade, there has been major progress in the development of artificial intelligence (AI). AI algorithms excel at data analysis and have evolved to the point where they surpass human performance for a wide variety of tasks. More and more businesses exploit these technological advances to optimise different kinds of processes such as marketing, sales and e-commerce, manufacturing and logistics. In today’s growing data-driven world, this trend is expected to continue on the back of widening opportunities for use cases, including in the area of operational risk management.AI can improve operational risk management in banking

Challenges in operational risk management

In June 2011 the Basel committee published the Principles for the Sound Management of Operational Risk (BCBS), which provides a framework for the development of proper operational risk management. Three years later, a survey was conducted to measure to what extent banks complied with these principles. One of these principles states that banks should write a risk appetite and tolerance statement. Banks reported that this is more challenging for operational risk than for other risk categories and attributed this to the nature and pervasiveness of operational risk. The banks that did comply with this principle often reported the use of backward-looking metrics of operational risk, such as operational losses as a percentage of gross revenue.

The above example underlines the challenges which banks face in the management and measurement of operational risk. Compared to financial risk, operational risk is a more qualitative field of study. Whereas financial risk management has been the main priority of banking for a longer time, operational risk management is much younger resulting in less extensive historical data. Predictive modeling becomes more of a challenge in this situation.

On top of that, the events in operational risk are much more diverse in scope. The Basel committee defines operational risk as the “risk of loss resulting from inadequate or failed internal processes, people and systems or from external events”. Internal fraud, data leakage and reputational damage are very different problems, yet can be very closely related as well. A lot of the data in operational risk consists of textual input which contain qualitative information. The qualitative nature of operational risk is reflected in the Basel framework, which encompasses guidelines for organisational structures, culture and awareness, and qualitative reporting.

The computer as a reader

Artificial intelligence could play a valuable role in upgrading operational risk practices, with in particular machine learning – the field of self-learning computer algorithms – showing promise. Machine learning algorithms can make predictions based on data fed to the algorithm. Recently, major progress was made in the field of natural language processing (NLP). NLP focuses on using textual data for predictions. As an example, an algorithm can learn to rate hotel reviews. By processing large amounts of reviews together with their given ratings, the algorithm can learn to give a rating to a new review it has never seen before.

“There are many opportunities for operational risk management to exploit AI and other related technological advances.”
– Lars de Ruiter and Matthias Geerse, Solid Professionals

In operational risk, many textual reports are written regarding specific risks or risk appetite of the organisation as a whole. These reports are generally written by risk managers – experts in their field. Assuming that risk reports hold information which is absent in historical loss data, this information could be extracted and used for predictive purposes. So why should computers perform this task instead of risk experts? The answer lies within the physical limitations of human being – as well as the natural biases they have to cope with. While humans may struggle to remember a piece of text, a computer algorithm easily processes thousands of books and finds structure within using statistical methods with near-perfect precision.

Towards prediction of operational risk

The performance of NLP algorithms using word embeddings has increased tremendously of late. Using the embedding as a starting point, these algorithms learn the meaning of full sentences and use this information for further predictions. This is more generally called sentiment analysis. A popular application in the financial setting is predicting stock price movement from news articles, or twitter feeds. A recent paper by Denmark’s national bank predicts corporate distress of firms by analysing their annual reports. Taking textual data under scrutiny can make a contribution to quantitative predictive modelling, even in the financial sector where the use of words is very different to day-to-day language.

In summary, there are many opportunities for operational risk management to exploit AI and other related technological advances. One possibility is the classification of risk events. As an example, a computer algorithm can read risk descriptions written by risk managers and classify them according to their impact and frequency. Combining loss data with risk reports, improved prediction of risk events might lead to more accurate prediction of future losses. A different application is the measurement of more abstract concepts such as the financial health or maybe the cultural aspects of a company.

An article by Lars de Ruiter and Matthias Geerse. Both are finance & risk consultants at Solid Professionals, a consultancy from the Netherlands.