In the media and at industry conferences, artificial intelligence conjugates the financial sector through all cases. However, when we look behind the scenes, the picture is not so crystal clear. A new report by Cloudera and Finextra Research acts as a cold shower: almost half of financial institutions are stuck between innovative experimentation and real business value. What’s blocking finance giants from hitting the ‘to produce’ button?
Until recently, there was a perception in the industry that the main barrier to AI adoption was a lack of suitable talent or insufficient computing power. Today, we know that the problem lies elsewhere. It is not the algorithms that are the bottleneck, but the archaic data architecture on which they are trying to be built.
According to a recent survey of more than 150 business leaders around the world, the financial sector has split into two speeds. We have the leaders and we have those who are still stuck in – what is termed behind the scenes in IT – ‘PoC hell’ (Proof of Concept).
The 48% trap, or AI in the waiting room
The statistics are unforgiving. As many as 48% of financial organisations admit that, although they have passed the experimentation phase, they still have not integrated AI technology into their core operational activities. This is a worrying sign. It means that almost half of the market is investing time and budgets in solutions that work in a lab setting, but fail to deliver an enterprise-wide return.
In comparison, only 26% of companies have achieved full AI adoption across the organisation. This disparity creates a dangerous gap. Companies in the first group are treating AI as an R&D curiosity, while those in the second are building real competitive advantage on it – from hyper-personalisation of offers to advanced real-time fraud detection.
Why is it so difficult to move from the pilot phase to implementation? The answer is: silos.
Data silos – the silent killer of innovation
Artificial intelligence, and machine learning (ML) models in particular, are only as good as the data we feed them. Meanwhile, the report points to a crushing statistic: 97% of financial organisations say that data silos hinder their ability to develop and implement effective AI models.
In practice, it looks like this: transactional data lies in the Core Banking system on the mainframe, customer behaviour data in the CRM cloud and risk history in yet another on-premise database. Without the free flow of information between these ‘islands’, AI models are malnourished – they only see a slice of reality, which drastically reduces their effectiveness and reliability.
It is the lack of a coherent data management strategy that causes ambitious AI projects to end up as slides in presentations, rather than as working functionalities in banking applications.
Hybrid reality: The end of the war between cloud and ‘on-prem’
The study also brings an important conclusion for system architects: the ‘public cloud or in-house data centre’ debate is no longer valid. Pragmatism has won out. 62% of financial organisations use a hybrid model, hosting data both in the cloud and in local centres. What’s more, up to 91% rate this approach very positively.
Why has the hybrid become the standard?
- Legacy requirements: banks cannot move trading systems to the cloud overnight.
- Scalability: flexible public cloud computing power is needed to train AI models.
- Regulation: Sensitive data often needs to remain under strict physical control.
- The hybrid model, while necessary, nevertheless introduces a new level of complexity to security management. And here we come to the crux of the implementation problem.
Safety and Governance: Brake or steering wheel?
For 25% of companies, security is an absolute priority when evaluating AI platforms. However, in a hybrid environment, where data circulates between the cloud and the server room, traditional security methods fail. Fear of data leakage or regulatory violations (compliance) means that security departments often apply the handbrake on AI projects.
Experts point out that the solution to this impasse is not to give up on innovation, but to change the approach to governance. The key to success is Unified Data Governance – unified data management.
Financial institutions need a platform that enforces the same security, quality and access policies regardless of whether the data resides on AWS, Azure or a local server. Only such a layer of abstraction allows data to be shared securely with AI models without worrying about where those bits physically reside.
Time to clean up the basement
The Cloudera and Finextra report is an important signal to the market. Innovation in AI can only succeed if it is based on a foundation of robust, unified data governance. This is confirmed by 84% of the organisations surveyed, who consider a unified governance framework critical for further development.
