If We Were Building It Today

Earlier in my career at BNY Mellon, I worked in Enterprise Business Architecture, helping business line CEOs think about how technology could change their businesses. Some of our most interesting conversations started with a question about a system or a process and quickly became conversations about entirely new products, services, and opportunities. What could we offer clients if we approached the technology differently? What capabilities were becoming possible that hadn’t been practical before? And what would the organization need to look like to take advantage of them?

It was fascinating work because technology had traditionally been treated as something that supported the business. Business leaders determined what they wanted to accomplish, and technology teams figured out how to deliver it. We were trying to bring those conversations together much earlier, allowing emerging capabilities to influence business strategy itself.

I also spent time working on the consolidation of onboarding operations from across the United States into a single team. On paper, the objective was straightforward. In practice, bringing together processes developed by different teams over many years required understanding far more than the steps documented in a procedure. There were questions about how decisions were made, where responsibilities belonged, and how the organization could preserve the knowledge and judgment that made those processes work.

Anyone who has been through a significant organizational restructuring will recognize the challenge. Processes develop histories of their own. Some steps exist because of regulatory requirements or legitimate operational risks. Others reflect decisions made years earlier, when the technology was different or the organization was smaller. Over time, the reasons for those decisions become less visible, even as the processes themselves become deeply embedded in how people work.

I’ve been thinking about those experiences quite a bit as artificial intelligence moves from experimentation into daily operations. Much of the conversation today reminds me of earlier periods of technological change, although the speed and breadth of AI’s development are unlike anything I’ve experienced.

Organizations are asking how AI can help employees write reports, review documents, process transactions, and respond to customers. These are sensible applications, and many will deliver meaningful improvements. But I worry that we’re approaching AI with the same assumptions that shaped the systems and processes we already have.

Consider how a bank handles customer onboarding. Information moves between teams responsible for collecting documents, verifying identities, evaluating risk, and establishing accounts. Each step has developed around particular responsibilities, systems, and regulatory expectations. AI can certainly make individual tasks faster. It can review documents, identify discrepancies, and assemble information for employees to evaluate.

But what happens when we reconsider how the entire process should work? Perhaps information no longer needs to move through the same sequence of departments. Perhaps employees can spend more time resolving complex cases while routine information gathering happens automatically. Some controls will remain essential, but the way we organize the work around them could change considerably.

The same thinking applies to local government. Preparing a municipal budget involves gathering information from departments, evaluating priorities, developing recommendations, and presenting them to elected officials and residents. Much of that process developed when collecting and analyzing financial information was labor-intensive. AI may help prepare the budget document more efficiently, but imagine how the conversation changes when officials can explore financial scenarios throughout the year, understand the implications of competing investments, and evaluate decisions with information that previously took weeks to assemble.

We have built entire organizational structures around the cost of collecting, moving, and interpreting information. Specialized departments, layers of review, reporting cycles, and management practices have evolved around those limitations. AI is beginning to change what is possible, and some of the assumptions underlying those arrangements deserve another look.

Of course, recognizing an opportunity and implementing it are very different things. Anyone who has tried to change how a large organization operates knows that the technology is only one part of the challenge. People have responsibilities, relationships, professional judgment, and an understanding of their work that rarely appears in a process diagram. Changing those arrangements requires bringing people into the conversation and understanding what makes the institution function.

There is also a tendency to underestimate the importance of institutional knowledge. The employee who has spent twenty years handling exceptions may understand risks that aren’t captured in any system. The department manager who knows why a particular approval exists may be protecting the organization from a problem nobody else remembers. Successful implementation requires learning from those people while being willing to challenge the processes they’ve inherited.

This is where I think organizations need to invest more of their attention. A successful AI demonstration tells us something about what the technology can accomplish. Turning that capability into lasting value requires understanding the business, reconsidering how work is organized, and building the confidence necessary for people to operate differently. The objective is an organization that can accomplish more, serve people better, and make sound decisions with the resources available to it.

Through my work in municipal finance, and more recently with companies building these tools, I’ve had the opportunity to approach these questions from several perspectives. I’ve worked with large financial institutions, local governments, and organizations trying to bring new technologies into established operations. The environments are different, but the challenge of translating technological capability into organizational change is remarkably familiar.

We have spent decades designing organizations around the time and effort required to gather, move, and interpret information. AI is beginning to change those limitations in ways we are only starting to understand. Before we automate another familiar process, we should ask a more fundamental question: if we were building it today, would we organize the work this way?