Why AI Can’t Modernise What It Doesn’t Understand

Somewhere in your bank, a core system built fifteen, twenty or even forty years ago is approving a transaction right now. It works. It has worked for years. And every year it quietly costs more to keep it working than anyone has written down.

That is the legacy tax. Most institutions pay it without ever seeing the invoice. In the AI era, that invoice is about to grow. The good news is that stopping it no longer takes a five-year rewrite. Intellect’s MSOCK makes a bank’s knowledge of its own systems computable, so AI understands before it acts and modernisation can move at pace. This piece itemises the tax, explains why AI raises the stakes, and shows how MSOCK changes the maths.

The current state: running to stand still

Banks invest more in technology than any other major industry: typically 6% to 12% of revenue, against 3.75% to 5% in telecoms, media and technology.¹ Yet much of that spend never reaches a customer. McKinsey estimates that technical debt accounts for about 40% of IT balance sheets.³

Legacy doesn’t look the same everywhere. In the US and Europe it often means mainframes written in the 1980s. In India, the Middle East, Southeast Asia and Africa, most cores date from the 2000s: younger, but heavily customised and built before mobile-first banking, real-time payments and open APIs. Across both, the pattern is familiar. Patches, customisations and M&A integrations pile up, while the people who understood them retire or move on. The system drifts out of understanding, and a system nobody fully understands is expensive to run and frightening to change.

That expense is the legacy tax.

The legacy tax, itemised

The true bill is larger than any budget line. Deloitte urges banks to judge legacy systems on total cost of ownership rather than purchase price, and warns that the cost and risk of running outdated cores only grow the longer banks wait.⁶ Most published research comes from US and European banks, and exact figures vary by market, but the five places the tax hides are universal.

1. The run cost. Keeping the lights on consumes most of the budget. In a McKinsey survey of CIOs at financial-services and technology firms, 10% to 20% of the budget meant for new products was diverted to dealing with tech debt, and 30% of CIOs put the figure above 20%.^2,3
2. The speed cost. On legacy architecture, new solutions take nine to eighteen months to reach production, against three to four on modern, modular stacks.⁴
3. The customer cost. In 2018, a quarter of new US current-account (checking) customers stayed with their existing bank without shopping around. By 2025 that had fallen to 4%.⁵ Frustrated customers rarely complain; they leave. Where instant payments and digital challengers make switching effortless, the pressure is sharper still.
4. The compliance cost. Regulators increasingly treat IT resilience as a supervisory matter, not a technical one. The Reserve Bank of India barred one large private bank from launching new digital products in 2020 after repeated outages, and in 2024 barred another from onboarding customers online, citing serious deficiencies in IT risk management.⁷ In Europe, the Digital Operational Resilience Act has applied since January 2025, placing similar obligations on banks.
5. The people cost. Many institutions run critical, high-volume transactions on systems that only a handful of engineers understand. Every resignation becomes an operational risk event.

The hidden tax: what legacy stops you doing

Those five costs at least show up somewhere, in a budget, a regulator’s letter or a deposit report. The larger share of the tax never appears on any line. It hides in slower change, repeated discovery, integration work, rework, and the engineering capacity consumed simply to understand how something works before it can be changed.

A change that should take months takes much longer. A new product needs work across systems that were never designed to work together. A modernisation programme spends months reconstructing dependencies before engineering can begin. And every significant change carries the risk of discovering something nobody knew was there.

This is where the legacy tax becomes a knowledge tax. Over decades, banks accumulate knowledge in business rules, product logic, processes, APIs, integrations, data relationships, exceptions, regulatory requirements and the decisions made during earlier transformations. That knowledge is spread across code, documentation, architecture diagrams, databases, interfaces and people.

The problem is rarely that no documentation exists. It is that enterprise knowledge is fragmented, and the relationships between its parts are hard to see as a whole. That is what makes change expensive.

The compounding cost of “next year”

Visible or hidden, the tax grows the longer a bank waits. Deferring modernisation rests on a flawed assumption: that next year’s operating environment will be more favourable. In reality, the cost compounds.

• The debt keeps growing. Six in ten CIOs in McKinsey’s survey said their organisation’s tech debt had risen perceptibly over the previous three years.²
• The complexity spiral. The number of applications banks run keeps rising.¹ Each one layered onto an ageing core makes it more entangled, harder to understand, and more expensive to modernise tomorrow than today.

That makes “can we wait until next year?” the most expensive question in the boardroom.

Why faster code hasn’t meant faster modernisation

Many banks hope AI will be the way out, and the prize is certainly large: McKinsey estimates generative AI could add $200 billion to $340 billion a year in value to global banking.⁸ Yet in several industry studies including PwC’s 2026 survey of more than 4,400 CEOs, 56% said AI had delivered no significant financial benefit so far, and those with strong technology foundations were three times more likely to see meaningful returns.⁹ For banks, that foundation is the core. Modernisation is no longer just about cutting run costs; it is the price of admission to the AI era.

AI has made software production dramatically faster. Developers can generate code, refactor applications, create tests and automate repetitive engineering at unprecedented speed. That should have transformed modernisation. Yet many banks that have added copilots and coding agents find that large transformations take about as long as before.

Consider a seemingly simple request: change this business rule. The code may be straightforward. But before touching it, the bank needs to know:

• Which applications use the rule, and which APIs expose it?
• Which processes and customer journeys depend on it?
• Which downstream systems consume its output?
• What regulatory or operational obligations are attached to it?
• What happens if its behaviour changes in one system but not another?

This is the knowledge tax at work. The answers exist, but they are scattered across code, documents and people.

AI can help answer these questions, but only with the right enterprise context. Without it, the bank simply produces changes faster than it can assess their consequences. Traditional programmes fail from the other direction, freezing an interview-based snapshot into a plan that ages while the system keeps moving.

AI has accelerated the act of engineering. The next challenge is accelerating the understanding that must come before it.

MSOCK: understanding before action

That is what MSOCK, the Multidimensional Multilayer System of Connected Knowledge, provides. It makes enterprise knowledge connected, computable and visual. It brings four capabilities together:

• A 21-dimensional Enterprise Spatial Graph that links business rules, processes, APIs and code into Connected Knowledge Units across business, operations, compliance and technology.
• Precise Blast Radius, which shows everything a change will touch before a line of code is written.
• Four Engineering Machines for BUILD, RUN, CHANGE and MODERNISE across the software life cycle.
• A Command Centre that gives architects and leaders one governed view of the estate.

Go back to that business rule. With MSOCK, Precise Blast Radius shows every application, API, process and obligation the change will touch, before a line of code is written. The questions that once took weeks of workshops become a query.

MSOCK is not another copilot. It is the knowledge layer beneath the copilots and coding agents a bank already owns, feeding them the enterprise context they lack, so existing investments stay in place and become accurate. And because every change is traced to the rules and obligations it touches, the audit trail regulators expect comes built in.

That matters wherever a bank sits on the legacy curve. For a bank running mainframe systems, MSOCK makes decades-old code legible again. For a bank in India or another growth market, it untangles the customisations layered onto a younger core before they harden into the same problem. And for any bank moving to a modern platform, it carries the knowledge across so nothing is lost in migration.

What pace looks like

Intellect has already run MSOCK internally across more than 100 mission-critical projects. On validated engagements, the difference is hard to ignore:

Measure	Before MSOCK	With MSOCK
Project cycle time	11 months	6 months
Engineering effort	170 person-months	50 person-months
AI token cost	Baseline	Up to 70% lower
Rework	Typical	Near zero

Figures reported by Intellect from internal deployments.

Now map those gains back onto the legacy tax. Shorter cycles attack the speed cost. Lower effort frees run budget for growth. Precise blast radius cuts the incidents and control failures that invite regulators in. And a living map of the estate turns the knowledge of retiring experts into an institutional asset rather than a resignation risk.

Making the case to the board

Boards don’t fund architecture. They fund outcomes. The strongest modernisation cases fit on one page, written in currency rather than technical debt, with one line for each part of the legacy tax:

Line item	What to put a number on
Run cost	Maintenance and infrastructure spend released, with a timeline
Speed	Revenue from launching products in months rather than quarters
Customers	Deposits and relationships retained through better digital journeys
Compliance	Probability-weighted cost of a regulatory action or major outage, with and without modernisation
People	Critical systems that depend on a handful of engineers, and the cost if they leave
AI readiness	AI use cases the current core blocks, and the value they would unlock

Then close with one clear, bounded ask. Not a five-year programme, but a short, measurable first step that proves the numbers on your own estate.

Blueprint for action: the Three-Week Open Challenge

The first step is deliberately small. Intellect’s Three-Week Open Challenge invites institutions to bring one complex application module. MSOCK maps its dependencies and change impact before any code is touched. Three weeks later, you see your estate as it really is, and what it would take to modernise it at pace.

The decision in front of you

For years, the case for waiting was that the tools would get better. They have. The decision now rests on a simple comparison. If tech debt already absorbs 10% to 20% of every new-product budget, and context-first modernisation cuts engineering effort by around 70%, delay has become the expensive option. And every year spent on a core that AI cannot understand is a year competitors spend compounding their lead.

The legacy tax is optional. It just doesn’t feel that way until you can see what you’re paying.