Legacy modernisation
Understand your legacy before committing to change
Five lessons from Mitra AI’s UK CDIO Advisory Board for CIOs navigating legacy modernisation, AI and investment decisions.
9 min read

A transformation programme can deliver its technical objectives but still fall short of the business expectations that justified and paid for it. Even when the systems work, the cost, disruption and human energy consumed can leave leaders wondering “Was it worth it?”
That tension framed Mitra AI’s first UK CDIO Advisory Board in early November 2026. Technology leaders from private, public and regulated organisations sat around a table to discuss the realities of legacy modernisation and how AI opens up a whole range of new options to take on the challenge.
One case offered a salutary lesson: a programme to consolidate more than 30 core legacy systems into SAP, with approximately 200 surrounding dependencies, achieved its technical objectives but finished five months late and £18 million over budget.
A recurring problem that emerged was the story of organisations committing to change before they sufficiently understand the systems and processes involved – where important business rule questions and dependencies surface after budgets, objectives and expectations have been set.
AI offers faster ways to investigate these “legacy environments” while increasing pressure to deliver reliable transformation sooner. For CIOs, the challenge is to establish enough understanding to choose a credible path and explain its risks.
Mitra has drawn together five themes from the discussion in this white paper, with recommendations for leaders facing similar decisions.
1. Factor uncertainty into the investment decision
Every modernisation programme begins with incomplete information. Leaders need to recognise the uncertainty, decide how much they will accept and establish whether further discovery would materially change the investment decision.
Members described undocumented processes and business logic buried in old platforms. Teams sometimes had to reverse engineer systems during delivery, consuming the expertise earmarked to build and test the eventual replacement. A late dependency discovery can change migration sequencing, testing scope or the feasibility of a proposed release.
Before approving the programme, ask which processes are understood end to end, which dependencies have been verified and where assumptions remain. Where uncertainty is material, fund discovery before committing to detailed delivery plans.
Members also discussed presenting cost and timing as ranges. Link those ranges to identifiable uncertainties and revise them as evidence improves. Better knowledge can narrow exposure without eliminating it.
Markerstudy assessed six million lines of code in four weeks
AI-assisted analysis produced a modernisation roadmap for the leadership of the Markerstudy insurance group that identified a path to more than £10 million in annual savings; these were potential savings, not confirmed realised benefits. The example illustrates how focused discovery can give leaders evidence before committing to change. [1]
2. The business must help explain what the systems do
Legacy systems contain an organisation’s accumulated decision history: acquisitions, customer commitments, policy changes and workarounds. Some decisions remain essential and endure but others will have outlived the circumstances that created them.
Examining the code and tracing how data moves through and between systems helps reveal how they work and depend on one another. Business involvement is needed to establish why those rules and relationships persist, whether they remain fit for purpose and how they should evolve.
A seemingly unnecessary rule may support a contractual obligation or downstream service and an inherited process (from an M&A transaction for example) may deserve redesign to follow best practice for the new organisation. Digitising a 50-field application form achieves little if existing data and reliable lookups could reduce the information a customer must provide. Understanding the process should help identify what can safely be simplified to improve the user experience.
Members stressed the importance of understanding processes end to end and of addressing the fact that critical knowledge is often concentrated in people who are busy keeping operations running. Confront it now because, when those people leave the company, recovering the rationale behind system behaviour becomes much harder.
You must ensure someone in the business is responsible for each process being changed and has time to help the transformation team understand it and validate proposed changes. Fund and protect that participation, so discovery and testing do not depend on “spare moments between incidents”.
Retain the knowledge recovered and assign responsibility for keeping it current. This reduces reliance on individual memory and heroic efforts. In regulated organisations, losing that knowledge can also make failures harder to explain and controls harder to demonstrate.
NatWest reduced idea to value time from 40 to 16 weeks
In 2022, AWS and NatWest reported that a shared machine-learning platform and standardised delivery processes cut idea-to-value time for data and analytics use cases from 40 to 16 weeks. Environment creation fell from 35–40 days to one or two days, including required validations. The lesson is to improve the workflow around the technology as well as the technology itself. [2]
3. Executive sponsorship has to survive difficult decisions
The strongest consensus concerned people, culture and sponsorship because modernisation depends on sustained leadership commitment, especially when a programme uncovers an inconvenient truth.
Business teams may need to change established processes or release experienced people. Suppliers’ assumptions may need challenging. Sponsors must resolve conflicts and accept the consequences of decisions. A constructive challenge should be assessed on its evidence, whoever raises it.
Leadership turnover makes this harder and members described how repeated changes at the top weaken accountability, confidence and collective memory. Teams become less willing to invest energy in a direction they expect a new leader to reverse or abandon.
Before delivery begins, agree the business outcome, the disruption the organisation will accept and the conditions under which scope, timing or approach should change. Name who can make those decisions and record the reasoning behind them so it is retained through leadership transitions.
Technical completion must also be distinguished from business success. Going live does not establish that the organisation has achieved the intended improvement in cost, service, control or adaptability. At each major review, ask whether the programme is still on course to deliver the outcome that justified the investment.
TSB incurred £48.65 million in regulatory fines
Following TSB’s 2018 migration, the FCA and PRA imposed £48.65 million in fines in 2022 for operational risk management and governance failures, including outsourcing risk. The FCA found that data migrated successfully, but the platform immediately suffered technical failures. This illustrates why executive oversight must test service readiness and supplier assurance alongside migration completion. [3]
4. AI creates more options but demands better judgement
Two approaches emerged in the discussion: some leaders are building AI capabilities over existing systems to deliver business outcomes sooner while others are exploring AI coding tools to interpret legacy logic and support redevelopment.
Building over the estate can deliver useful capability without waiting for replacement, but underlying knowledge gaps may remain. Understand which rules, data and controls the new capability depends on and assess the risk of retaining those dependencies (and quantify it where practical).
AI-assisted redevelopment can recover knowledge while supporting change, but the organisation must validate the AI’s interpretation of existing business logic, preserve essential behaviour and detect errors. Faster code generation creates value only when it produces reliable capability at lower cost or in less time. Your organisation must remain accountable for how the resulting software operates.
Some members were clearly leaning towards building more software internally and they shared examples of replacing purchased capabilities and questioned future dependence on packaged CRM. However, we consider these to be useful examples of individual experiences and views, rather than a collective conclusion of the event.
The implication is to reopen build-versus-buy decisions where AI has materially changed the economics. We recommend comparing the continuing cost and responsibility of owning a capability with the cost and constraints of buying it. You must include testing, security, integration, support and future change. Name who will remain accountable once the software becomes operationally important.
Enthusiasm was tempered by questions about measurable returns. Members described point solutions outside core processes and unresolved concerns about governance and compliance. Any investment case should distinguish demonstrated value from expected gains and specify how the business will test the difference.
Mitra’s view is that AI should widen the options while strengthening the demand for validated evidence. The decision needs sufficient understanding of the business requirement, the relevant estate and the obligations the chosen approach creates.
NEXT revised its estimated project cost from £50 million to under £10 million
In its September 2026 half-year report, NEXT said AI had changed its estimate for interpreting, documenting and rewriting legacy mainframe applications. It now expects the project to cost under £10 million and take less than 2.5 years.
These are forecasts, not completed savings but they demonstrate why previously unattractive modernisation investments deserve reassessment. [4]
5. Plan for changing options while protecting the outcome
Members described a mismatch between transformation timescales and AI development. Programmes that previously would run over years now need to be re-evaluated (as with the NEXT example) because assumptions about tools and capabilities can change within months.
Large, irreversible commitments therefore need careful scrutiny before being made, but indefinite delay also carries costs as operational pressures and knowledge loss continue. CIOs must be careful to distinguish the business outcome that must endure from the implementation choices that will likely evolve.
Better customer service, stronger controls or lower operating costs can remain the objective while the technology changes. Mitra’s approach to progressive modernisation is to deliver change in manageable increments, with explicit review points and criteria for continuing, adapting or stopping. Each stage should produce useful capability or evidence for the next decision.
Members also raised concerns about vendor platforms, traceability, compliance and data sovereignty. A capability that is quick to adopt may create dependencies that are expensive to unwind. Assess that exposure before committing, including how the organisation will retain its data and recovered business knowledge if a tool or supplier changes.
Clear ownership and maintained process understanding help preserve future options. Discovery has continuing value when it leaves the organisation better equipped to make the next decision.
DVLA delivered 20 major releases over two years
Kerv reports that DVLA modernised its Drivers Medical and Drivers Casework services through 20 major releases over two years. Reported results in the five months from January 2026 included more than 28,500 automated decisions and 51,000 online licence renewals. The case illustrates how a sequence of releases can modernise critical services while delivering measurable operational results. [5]
Five questions to answer before the next commitment
Use these questions as a reality check with stakeholders before the next investment or programme review.
What do we know and what are we assuming?
Identify validated business rules and dependencies. Make material gaps visible, with a plan to investigate them.Whose knowledge does delivery depend on?
Name the people and teams needed to explain the operation. Protect their participation and retain their knowledge.Which decisions require executive ownership?
Agree who will resolve conflicts over processes, resources, risk and timing (including through leadership changes).Why is this the right modernisation path?
Compare retaining, extending, wrapping, rebuilding and replacing against business outcomes, risk and continuing ownership costs.What evidence would make us change direction?
Define what would trigger a review and give the programme authority to respond when the evidence changes.
For Mitra, the central lesson is that understanding the estate is part of transformation itself. It informs what to change, how much to commit and which risks to accept. AI can accelerate that understanding. The CIO’s responsibility is to turn it into better decisions.
Sources for the external examples
1 Ensono - Six Million Lines of Code. Four Weeks. £10M in Annual Savings.
2 AWS and NatWest - How NatWest Group built a scalable, secure, and sustainable MLOps platform. 26 April 2022.
3 FCA - TSB fined £48.65m for operational resilience failings. 20 December 2022.
4 NEXT plc - Half Year Results 2026/27. 17 September 2026. Report reproduced by FinancialFilings.
5 Kerv - Migrating Drivers Medical & Drivers Casework Services to Microsoft Dynamics 365.