AI has moved quickly from an emerging technology discussion to a strategic priority for financial services firms. The opportunity is significant, from improving fraud detection and customer service to increasing productivity and helping people make better use of information across complex organisations.
The question is increasingly how firms move beyond individual use cases and pilots to make AI work reliably at scale. For organisations that have grown through repeated acquisition, I think that raises a more fundamental question.
What happens when you introduce technology that depends on consistent data, clear processes and well-understood decisions into an operating environment built from years of inherited systems, processes and ways of working?
AI may be one of the things that makes that accumulated complexity much harder to ignore.
Scaling AI means dealing with the organisation as it really operates
HM Treasury’s 2026 Financial Services AI Adoption Plan describes the opportunity for the sector to move beyond isolated pilots towards AI being adopted at greater scale, with potential benefits across fraud detection, operations, risk management and customer outcomes.
The ambition makes sense, but scaling technology across an organisation is very different from proving it can work within a controlled use case.
A pilot can operate within one team, draw on a defined dataset and focus on a relatively contained problem. Once AI begins interacting with more of the organisation, it also encounters more of the organisation’s history.
In a business that has grown through repeated acquisition, that history can include different systems holding similar information, several versions of the same process, inconsistent definitions and teams that have developed their own ways of bridging the gaps between them.
As we explored in how acquisitions create complexity in financial services, much of this can accumulate gradually through perfectly rational decisions about what to integrate, what to retain and what simply isn’t worth changing at the time. The problem comes when those decisions begin to interact.
People compensate for complexity in ways AI cannot easily replicate
One reason acquisition complexity can remain hidden for so long is that people become extremely good at working around it.
An experienced employee may know that one system is technically the system of record, but another contains the information they actually need. They may understand that one legacy business handles a particular exception differently or know who to speak to when a process reaches an unusual point.
That knowledge often sits outside the documented process.
AI changes the dynamic because those assumptions need to become more explicit. If information is recorded differently across acquired businesses, which definition should the technology use? If similar customer journeys operate under different rules, is that variation intentional? If ownership becomes unclear when work crosses legacy boundaries, where should an AI-enabled decision sit?
Those are not really technology questions, but questions about how the organisation operates.
The same applies to data. If different parts of a business capture the same information in different ways because they inherited different processes, cleaning or integrating the data may solve the immediate issue without addressing why the inconsistency exists.
AI can therefore act as a useful diagnostic lens, exposing operational differences that people have previously been able to absorb.
There is a risk of making complexity more efficient
It’s entirely possible to build technology around a complicated operating model. If several versions of a process exist, a sufficiently sophisticated solution may be able to accommodate all of them. Integrations can connect systems, automation can reproduce workarounds and AI can help people navigate information spread across different parts of the organisation.
All of those interventions can create value, so the question is whether they have made the organisation simpler or simply made it better at carrying the complexity it already had.
This is somethin we explore further in Untangling acquisition complexity in financial services. Technology introduced before unnecessary variation is understood can end up embedding that variation into the future operating model.
That doesn’t mean organisations should delay AI until every legacy process has disappeared. For most large financial services firms, that would be unrealistic. It does mean there’s value in deciding which complexity should be carried forward before building new capabilities around it.
The aim isn’t to remove every point of friction
There is also an important distinction between simplifying an operation and removing all friction from it.
Financial services organisations operate in environments where additional controls, checks and interventions can serve a deliberate purpose. A customer journey may need friction to identify vulnerability, reduce fraud, manage risk or prevent harm.
Equally, not every inherited step or control exists for an equally important reason. Years of acquisition can leave controls layered on top of controls, exceptions created to deal with other exceptions and additional handoffs introduced because systems or teams never fully joined together.
The challenge is therefore to distinguish purposeful complexity from accumulated complexity.
That matters when deciding what AI should optimise. Before asking how technology can make a journey faster or more efficient, organisations need to understand whether the journey itself still reflects the outcome they actually want.
Use AI as a reason to understand the organisation you inherited
Trying to remove every legacy system, process variation and dependency before progressing with AI would create an enormous transformation programme in its own right. A more practical approach is to look at the particular journey or capability where AI is being introduced and understand the complexity around it.
- Where does the information come from?
- Where does it change meaning?
- Which variations are necessary?
- Where do employees intervene manually?
- Who owns the outcome when something doesn’t follow the expected path?
Sometimes the answer will be to simplify. In other cases, there will be good reasons to retain a particular control or variation. The important thing is that those decisions are being made deliberately, rather than simply carrying forward ways of working inherited from previous acquisitions.
That's really the wider argument behind our whitepaper, The Hidden Operational Complexity of Acquisition. It looks at how operational complexity builds through acquisition, how to identify where it is creating friction and how organisations can start creating simpler foundations for future transformation.
At Reinvigoration, much of our work starts with understanding that operational reality: how work genuinely happens today, where people are compensating for complexity and what needs to become simpler before organisations build what comes next.
For financial services firms looking to scale AI, I think that's an increasingly important place to start.
