No, seriously. Your bank really isn't ready to implement AI.

Amgun - stock.adobe.com

Artificial intelligence may not yet run the world, but it looms large in every conversation about how the world is going to run from here on out. Nowhere is this more evident than in the financial sector, where the sense of urgency around AI capabilities — and threats— has reached near frenzy over the past year or so.

Powerful new assistive and generative tools suddenly seem to be everywhere, and they really do open enormous opportunities for the holy trinity of successful financial services: making new money, saving old money and reducing risk.

The speed and scale and adaptability of new AI technologies will rewire how institutions competitively pursue all three. But they do not erase the reality that most firms are in no position to use them effectively at present. All of these technologies are fueled by data. And, as noted in MIT Technology Review late last year, "Many financial services companies, especially large banks and insurers, still have substantial, aging information technology and data structures, potentially unfit for the use of modern applications."

Before diving headlong into the AI revolution, financial institutions need to take a true accounting of whether their data infrastructure is actually fit for the fight. Banking and insurance leaders look at ChatGPT or Gemini or Claude, and they see what Microsoft or Google or Amazon are doing with these technologies. And then they think that's what they have to do: "We've got all this information. We have all these documents. Let's just turn them into our own AI!"

But that's not how it works in real life. 

The problem isn't that they don't understand what's at stake or know what they want to do. The problem is they have vague or unrealistic expectations of what can be done, by whom and how. Hype tends to encourage executives (and their boards and their shareholders) to shoot too far, too soon and too high. If organizations actually benchmark and ground where AI plays in reality, and where their processes and IT systems can and cannot accommodate the flow and management of their data in support of such technology, harsh truths emerge. AI prowess is not beyond reach, but it can't be attained by aspiration alone.

Artificial intelligence

Senators propose measures to counter AI manipulation

Mark Warren and Thom Tillis have introduced the Secure Artificial Intelligence Act of 2024 to address the unique risks of AI.

By Miriam Cross

May 1

From a purely technical perspective, the average bank has around 45 different systems that have information about one individual customer, supported by probably 15 to 20 different business lines inside the institution. Two people could share a wall inside the office, and could both be looking at that customer's information and have absolutely no idea that it is the same person. And yet this is an industry that was built around highly informed, trusted and individualized customer service. 

This fracture of systems and fracture of businesses and fracture of relationships is a data architecture problem — and it is anathema in a market that is demanding ever greater speed, personalization, centralized awareness and individualized service. In finance, to effectively leverage AI's benefits for efficiency, risk reduction or even monetization, the industry and the markets themselves require fundamental synthesis of the information occurring across an institution — data compliance, governance, definition, consistency, transparency — are all musts. They are not optional when it comes to actioning financial information, and they are generally not integrated fluidly in the existing data stacks of many, many firms.

Applicable AI technology is based on the very basic principle of data warehousing business intelligence. But like everything else in IT, it's subject to the law of "garbage in, garbage out." So even if your teams know exactly how to program AI, and your organization is completely aligned on what to do with AI, guess what's going to happen with your AI proofs-of-concept if your data is a mess? There is fundamental work involved in laying a modern data foundation and adopting a flexible, data-driven, cloud-based IT architecture suitable for AI implementation. And this kind of inherent technical debt can't be shunted off (per usual) to some inflated GSI no matter how hefty the price tag or pretty the promises. If your digital core is outmoded and cobbled together, it isn't going to be as simple as planning and deploying a mature data architecture overnight.

This is not to say that firms cannot benefit from assistance in modernizing their processes and systems for the age of AI, or that they can't achieve serious improvements fast by updating and refining their data stacks. They can and should draw on modern resources strategically. They can and should engage third-party experts and managed service providers with sector-specific competence. They can and should start working on change right now. 

But all of these pursuits must be grounded in clear-eyed assessment of your existing data infrastructure and real commitment to methodical remediation. And they must be framed directionally for your particular business. There is no point in paying a professional services giant to spend six months "facilitating" your modernization if that modernization doesn't translate into real improvement to everyday operational efficiencies and tangible returns on your investment. And there is no point in investing in technology, no matter how powerful, if you cannot implement it or if it will not work for your business. 

Mastering AI requires pragmatism, not just transformation for transformation's sake. Road maps guiding the steady, distinct internal steps necessary to streamline data architecture and extend capability are what drives true-to-form innovation journeys. That's the tangible work that will meaningfully transform financial services firms and equip them to compete into the future.