Sophie Dionnet, vice-president of strategy at enterprise AI company Dataiku, oversees the development of analytics and data science throughout financial services. Having previously spent 14 years in management roles at AXA, Dionnet has long specialised in the financial services sector.
In this Q&A, we explore how AI has driven value for financial institutions, the challenges that they face when it comes to successful deployment, and the plans that Dionnet has in place for 2021.
What are the biggest challenges that financial institutions face when it comes to using analytics, and how can they be overcome?
Financial institutions are data-driven by essence, as all their core processes depend on appropriate access to data, fueling models used to make the best risk-adjusted business decisions. In theory, given this, integrating analytics and AI in business processes should be easier for financial institutions. However, the reality is quite different, and the majority of organisations face a litany of issues.
Access to, and the quality of, data explains why banks and insurance companies face an uphill battle. For example, they are typically dealing with data silos linked to a historical “product-only” approach to data, fragmented information systems resulting from long histories of mergers and acquisitions, and a lack of strong internal data culture. Technology will support end-to-end data perspectives, and building a strong data culture with distributed ownership on quality is also essential to break the “user-only” disempowered approach, supported by strong data governance programs.
Likewise, organisations must recognise that scaling with analytics starts with broadening access to data. Banks and insurance companies can be extremely reluctant to do so for a variety of reasons including regulation (eg. GDPR), absence of central warehouses, perceived risks on infrastructure resilience, and more. However, there is a need for strong enablement and governance programs to educate staff on the rights and wrongs of analytics and build referential analytics.
Thirdly, another main barrier to analytics development is linked to upskilling and trust. With financial services being so tightly connected to quality of data and modeling, opening analytics can be seen as a Pandora’s box that could lead to people making the wrong decisions. A typical example is how to deal with absent data. Let’s say we don’t have prices and characteristics for all the instruments traded today. There are times when making an estimate can make a lot of sense (e.g. to estimate margin calls, risks, etc.), but in some cases, “guessing” vacant data can have a significant impact on decision making. Companies willing to embrace analytics have to invest in the upskilling of their employees and build a suitable collaboration environment to organise exchange and controls between risk experts, business professionals and data scientists to develop well-controlled initiatives.
Rebuilding your data analytics capabilities in a post-Covid world
What have been the most valuable ways in which AI has disrupted the financial services sector?
The first to embrace the AI journey were the investment teams, who — in their constant search for unique market insights and investment models — have seen in AI a unique opportunity to innovate. While it has been very successful for a few, it has also led to many unfruitful initiatives and has, to a certain extent, led to the misconception that AI is all about innovation and cracking highly advanced market topics. The financial companies that have been most successful with AI are those who dare focus their AI initiatives on the “day one solving topics”: operational processes optimisation, customer analytics and customer journey enhancement, risk management across all dimensions, and more.
After more than 10 years of deep regulatory transformation, all financial players have significantly enhanced their risk frameworks. But much remains to be done across all dimensions. The successful integration of AI in risk management has played an essential role in supporting reinforced robustness of the banking system, including agility and impact in investigations, development of new internal controls, and enhancement of financial crime monitoring through analytics, to name a few examples.
AI is also a real revolution within risk assessment, notably through the enhanced use of alternative data. This is true both for traditional risks and emerging risks such as climate change, helping all financial players — banks and insurers alike — to reconsider how they price risks. Those who have developed a strong expertise in leveraging alternative data and agile modeling have been able to truly benefit from their investment during the ongoing health crisis, which has deeply challenged traditional models (notably on scoring for corporates).
Lastly, the positive impact of AI on customers should not be underestimated. Financial services are confronted with an aggressive competitive landscape as well as demand from customers for improved personalisation, driving improved customer orientation in these organisations. The capacity to build 360° customer views and optimise customer journeys, notably on claims management, are two examples of areas where AI has significantly supported deep transformation within banks and insurance companies, with yet much more to be delivered.
Neo CTO discusses delivering a tech strategy in SaaS corporate banking
How can AI best be scaled to meet the needs of the enterprise and its customers?
Successfully scaling AI has risen to the top of strategic agendas as more and more companies prove AI’s ability to impact growth trajectories. However, while organisations claim that embedding AI in critical business processes is an objective, most have not yet cracked the recipe for success.
There is no universal solution to such a complex business transformation topic. Each industry — and each company — has to find its own path forward that’s well adapted to its own DNA and business. But as companies strive to define their approaches, all will have to contemplate the following:
1. Aligning AI efforts to strategic objectives. Scaling with AI means getting past experimentation and building beyond the first few successes. Most companies tend to crack the “usual suspects” as a starting point — and they are right to do so; they build trust on AI impact and test their initial operational frameworks at the same time. The real transformation comes with the second step, which is to build beyond these. Ultimately, the objective is not to consider AI as an isolated topic, but rather as a driver of the development of core business.
2. The right framework to scale. Scaling with AI also means embedding AI in business processes. Doing so has to come from the business lines themselves and thus needs to be supported by an organisational model that strikes the right balance between bottom-up empowerment and top-down support. For example, it must include at a minimum upskilling of existing business professionals, access to tooling and data, identification of priorities, and strategic roadmap setting. Cross-functional collaboration — both between types of business profiles, including central operations and data teams — are instrumental to accelerate usage of AI in organisations.
3. Capitalisation and reuse. The initial effort to implement AI is costly. The aim is to ensure that each new piece of analytics or model is not considered as a stand-alone data product, and that AI is developed as an interconnected stack, with existing data products fueling the development of many other AI implementations. This should be core to any AI scaling effort to deliver AI at the right level of return on investment (ROI).
4. Management and control. AI brings new types of risks, the most critical of which is the risk of unintentional misalignment between outputs and intentions. Let’s take a churn model developed by a telco company, driving proactive customer engagement and aggressive price reduction. When developing such a model, there is probably no active conscious decision to favour “risky customers” vs. faithful ones. And yet it could, if not put into the right perspective. Collaboration of all involved stakeholders and having the capacity to fully document, explain, and control models end-to-end are among the must-haves to controlled AI scaling, with the ultimate objective of managing all impacts towards end customers.
Co-managed IT: shifting to a new IT model
What plans do you have in place at Dataiku in 2021?
At Dataiku we are convinced that having an end-to-end, inclusive approach to AI is a must-have to successfully accelerate usage of AI in organisations. We are continuously reinforcing our offer to be able to embark all profiles in the AI journey, tackling both the needs of upskilling business teams through strong visual tools and providing advanced coding environment and tooling to data scientists.
Beyond tooling, there is also a question of moving from vision to implementation. We are heavily engaged in providing the right type of support to all the stakeholders of AI initiatives. This includes offering strong initial training to the tool, and business transformation advice to AI leads. We believe in fostering cross-learning opportunities between this unique community of organisations all sharing the same ambition of successfully embedding AI in their business models.
With AI growing in importance and impact, we also see new topics gaining in materiality, such as the need to fully govern AI, encompassing trends such as MLOps, developing explainable and fair AI, and having oversight and controls of all AI initiatives. We are building on the very strong foundation of the Dataiku platform to further strengthen the support we will bring to organisations on this front.