"We'll invest in AI once the core is ready."  

What if waiting is the most expensive decision you're making right now? 

Slowed by legacy systems, organisations face more than just inefficiency. Outdated infrastructures eat budgets, hinder strategic decisions, and delay the return on AI investments. 

The debt keeps compounding 

Most organisations already know their legacy infrastructure is a problem. What they might be underestimating is how fast the cost of inaction grows. 

Technical debt is a significant burden on an enterprise's technology assets, and it only compounds over time. Every new project pays an additional premium just to work around existing, outdated systems. 

Forrester predicts 75% of technology decision-makers will be dealing with moderate to high technical debt severity by 2026. And the longer they wait to fix it, the more expensive it's going to become. 

Beyond maintenance spend and stalled projects, there's another aspect that is not tracked by any dashboard: the cost of being slow. 

Slow systems mean policies and credit models are calibrated to outdated data, so by the time a new rule is live, the market has already moved. Moreover, product launches can take months instead of weeks because teams are stuck fixing integrations instead of shipping. And the data that could inform a better decision usually exists but isn't available at the right moment. 

Legacy systems are rarely blamed for these problems. When an outdated risk policy hurts portfolio performance, the blame goes to market conditions, or when a product launch takes four months longer than it should, it's an execution issue. These costs show up in NPA, in lost market share, in deals that go to a competitor — and they keep piling up until the infrastructure catches up. 

Why sequencing AI after modernisation doesn't work 

There's a common assumption that modernisation and AI adoption are sequential: fix the infrastructure first, then layer in AI. But modernisation projects are slow, expensive, and frequently delayed. Waiting for the core to be ‘ready’ can mean waiting indefinitely. 

On the other hand, the math gets worse for organisations that try to bolt AI on top of unchanged legacy infrastructure. Keeping old platforms running while piloting new AI applications does not reduce technical debt. Run costs stay high and any productivity gains from AI get offset by the cost of maintaining what's underneath it. 

In fact, 85% of leaders are concerned about legacy systems limiting meaningful AI adoption. In financial services, 68% of CTOs cite legacy infrastructure as the single biggest obstacle, and AI initiatives in the sector experience average delays of 12–18 months specifically because of compatibility issues with existing systems. 

Adopting a dual-track strategy  

Complete core replacement isn't realistic for most enterprises. These systems hold borrower records, drive workflows, process compliance, and more. Replacing them in a massive overhaul means betting your operations on a single migration succeeding, a gamble that rarely pays off.  

Full replacements routinely run over time, over budget, and often get sidetracked halfway through. These inefficiencies come at a high cost: an average global enterprise wastes over $370 million annually from failed modernisation efforts. 

The more effective path is running two tracks simultaneously: modernise the core incrementally while deploying AI-native capabilities now, without waiting for the core rehaul to complete. 

Modern API layers can sit on top of legacy systems, pulling data into new tools, dashboards, and workflows without touching the underlying infrastructure. The legacy system doesn't need to be rebuilt to be useful, it just needs to be accessible. With the API layer serving as a bridge, new capabilities replace old ones gradually, piece by piece, rather than through a single high-stakes overhaul. 

More importantly, AI doesn't need clean, perfectly unified data to generate value. It can work across fragmented legacy sources to surface insights and automate processes. The two tracks are designed to run together: the API and AI layer starts producing returns early while the deeper modernisation continues in the background. Enterprises that take this approach keep run costs at least 20% lower. 

ROI requires capable infrastructure 

The AI opportunity is not in question; many financial institutions are running AI pilots and seeing early wins. 

It’s whether the underlying infrastructure can act on what AI delivers. For instance, a model that identifies the right offer is worthless if the system can't share it in real time, just as a fraud signal that fires 20 seconds too late won't prevent the transaction. Ultimately, AI generates value at the moment of execution, and effective execution is an infrastructure problem.  

Every quarter that passes without a modernisation strategy with a clear deadline, the gap between what your infrastructure can do and what the market demands gets wider. That gap has a price — in bad debt, in missed segments, in competitors who shipped their third product iteration while you were still finishing your first. With these growing costs and competitive pressures, can you still afford to put AI on hold for a full-scale modernisation? 

Until next time,

Srijan
Co-founder
FinBox