You Can’t Outrun Debt by Spending More!
Very early in my career I was working in credit and lending. I went to meet with the regional VP to share some disturbing news: our delinquency rate was going up. This was not a surprise. We lacked standards and were trying to gain market share by making questionable decisions. What was surprising was my boss’s solution. He told me not to worry, that he had a plan. We were going to get more aggressive and add more loans to our book, which would drive down the delinquency rate.
While mathematically accurate, this approach was obviously problematic. We were losing real dollars today in the hope that future income would offset it, and we were doubling down on bad processes, expecting a different outcome.
As I look at AI implementations today, and the broader view of IT management, I’m seeing a similar strategy. Many teams are throwing whatever they can at the wall, hoping something sticks. This is problematic for many reasons. If we consider IT organizations’ capabilities and maturity levels, we can spread them across a typical bell curve. A few are at the top of their game, some are failing outright, and most are somewhere in the middle. The majority in the middle are good at some things and struggling with others. They likely have transformation initiatives underway, aiming to improve. All of these organizations are wrestling with AI.
It’s no secret that we are in an AI hype cycle. Anytime the business comes to you asking for a specific technology instead of helping to solve a business problem, that’s usually a sign the hype is in full swing. AI is dominating the conversation with CEOs, and pretty much everyone else, saying they’re using it in their personal lives and not only asking how AI can help but directing teams to “use AI first”. This pressure is pushing IT leaders, consultants, and individual contributors to chase quick wins, often without a solid foundation.
Historically, rapid technology adoption has created problems that may not be visible on day one but show up later in the form of cost overruns, increased cyber risk, poor data fidelity, and ultimately bad business decisions. After aggressive implementation cycles, teams often look up with the equivalent of a hangover from a long bender and try to figure out how to clean up the mess.
Now more than ever, sound implementation, architecture, and change management practices are needed. As IT leaders run headfirst into AI implementations, it’s time for a moment of self-reflection. Ask yourself, your team, and your business partners some hard questions. How confident are we in our ability to do this well? Look back at your history of new technology rollouts. Are there things you wish you’d done differently? Are you still cleaning up from the last big thing? Given the outsized potential that AI has for doing both good and harm, can you afford to approach it the same way you always have?
Let’s look at this challenge through the lens of the four primary domains of Enterprise Architecture.
Business Architecture
How well is your technology mapped back to business capabilities? What is your business really good at, and where does it need to improve? What makes you different from your competitors?
Are your business processes well documented? Are they optimized?
Presumably this is one of the main areas where we expect AI to provide value and productivity. But is the plan just to throw AI at your current process and hope it works? Is the goal to replicate what a person does and reduce headcount? If the process is not efficient or well understood, can we realistically expect to hand it off to AI and be successful?
Data Architecture
How confident are you in your data? Is it accurate? Do you have traceability to your books of record? Who in the organization owns each key data element? How would you rate your metadata? Is your data sensitivity properly tagged? What about PPI and other regulated forms of data? Not to mention data access controls.
Data is the lifeblood of AI. When LLMs first emerged, I was already hearing from development teams that we needed to train the model on our data. We had barely scratched the surface of what it could do, and we were already looking to give it all of our data, the good, the bad, and the ugly, without a true understanding or appreciation for what we would be sharing.
Application Architecture
Grade your application portfolio. Do you have more applications than you need? Are your governance processes and architecture practices solid when building or acquiring new apps? Can you accurately tie application costs to the value they provide the business?
Application portfolio management is something many IT organizations struggle with. The introduction of AI-based applications, AI-generated code, and AI agents will only make portfolio management more complex. What is the plan to track and govern AI applications?
Technology Architecture
How would you rate your IT inventory today? Do you have a clear picture of all IT assets and how they interact? Any orphaned assets? Do you have a vendor strategy and a leadership endorsed 3-to-5-year roadmap that is tracked and governed?
New technology has a way of creeping in. Even the best-run IT shops struggle to keep up. With the surge in new tech adoption and more people in your organization experimenting with AI, how will you effectively manage and govern your growing technology footprint?
Now more than ever, the discipline and practice of Enterprise Architecture is essential. EA isn’t about saying no to innovation. It’s about making sure new technology is adopted in a way that actually supports the business using time tested frameworks and practices to ensure you stay in control.
AI is not a magic box or a shortcut to transformation. If your last major tech initiative left behind unfinished projects, bloated apps, or governance headaches, what makes this time different?
Instead of asking, “How can we use AI?”, reframe the question:
Given our current maturity and gaps in ‘X’, is there a focused, justifiable way to use AI to improve ‘Y’?
The hype will fade. The risks won’t. You can’t outrun your technical debt by throwing more technology at it, even if that technology is called AI.
Comments