A leadership team approves an AI-enabled service. The demonstration is convincing. It can interpret a request, find relevant information, draft a response and recommend the next action in seconds.
Then the team tries to place it inside the real business.
The customer record is split between the CRM, a case-management system and individual files. Prices sit in an ERP that was never designed to expose them in real time. Permissions differ across departments. A key step still depends on a spreadsheet and an email. Nobody is entirely sure which system holds the definitive status of the case.
The model is not the immediate constraint. The digital core is.
This is where many AI strategies become detached from delivery. The strategy describes high-value opportunities, but it treats the organisation’s existing technology and operating environment as a later implementation detail. When delivery begins, hidden dependencies appear. Teams compensate with manual work, brittle integrations and additional tools. The prototype still looks intelligent; the service around it becomes harder to operate.
An effective AI strategy therefore needs two connected views. The first sets the ambition: where AI could improve growth, service, productivity or decision-making. The second explains what must be true across the digital core for those outcomes to be dependable.
That does not mean replacing every legacy system before doing anything with AI. It means identifying the smallest set of changes that will unlock priority services, reduce avoidable risk and create reusable foundations for what comes next.
AI exposes the architecture behind the ambition
Traditional digital services can conceal fragmentation. A modern interface may sit over manual checks, re-keyed data and disconnected systems. Staff absorb the complexity by switching between applications, reconciling records and filling gaps from experience.
AI changes the economics of that arrangement. A person can notice that two records conflict, question an unusual result or remember an undocumented exception. An AI-enabled workflow needs reliable access to the right information, explicit rules for what it may do and a clear route when confidence is low. If those foundations are weak, automation can make the underlying problem faster and less visible.
The UK Government’s 2025 State of Digital Government Review provides a useful public-sector illustration. Based on engagement with more than 500 people across 120 public organisations, it describes fragmented technology, point-to-point integrations and modern front ends masking manual processes and data re-entry. It also reports that around 28% of central government technology estates were classified as legacy. These figures should not be generalised to every UK business, but the delivery pattern will be familiar to many leaders: a better customer-facing layer does not, by itself, resolve weaknesses underneath it.
AI can expose those weaknesses because it crosses boundaries. A useful assistant, agent or automated decision flow may need to interpret unstructured information, retrieve a verified record, apply a business rule, trigger an action in another system, maintain access controls and leave an auditable trace. Each dependency becomes part of the service.
This is why AI readiness cannot be reduced to model choice or data quality. It is a property of the wider organisation.
What is the digital core?
The digital core is the set of capabilities on which the organisation’s services and operations depend. It includes:
- systems of record and the applications people use to perform work;
- data definitions, ownership, quality and access;
- integrations, APIs and event flows between systems;
- identity, permissions, security and resilience;
- infrastructure, monitoring and support arrangements;
- the teams, suppliers and decision rights needed to change and operate them.
This definition matters because “modernisation” is often treated as a technology replacement programme. In practice, the most valuable intervention may be to clarify a process, establish a reliable source of truth, expose a controlled interface, improve identity management or give a neglected service an accountable owner.
The digital core is therefore both technical and organisational. Architecture determines what is possible. Operating choices determine whether it remains dependable.
Five foundations an AI-enabled service needs
Leaders do not need to assess the whole estate at the same level of detail. Start with the priority services and use cases in the AI strategy, then test the foundations they depend on.
1. A service and process that make sense
AI cannot repair a service whose purpose, users or outcomes remain unclear. Nor should it automate every step in an inherited process simply because that process already exists.
Map the service from the user’s point of view. Identify where value is created, where judgement is important, where information is duplicated and where delays or failure demand manual recovery. Decide what should be removed, simplified or redesigned before deciding what AI should do.
This shifts the question from “Where can we add AI?” to “How should this service work now that new capabilities are available?” The distinction prevents a faster model from being attached to a process that should have changed.
2. Information that can be trusted in context
AI needs more than access to a large volume of data. It needs information that is current enough for the task, understood well enough to interpret and governed well enough to use.
For each priority service, identify the authoritative records, the data required at each decision point and the conditions under which it may be used. Resolve critical definitions. Make gaps and confidence visible. Establish ownership for correcting problems rather than expecting the AI delivery team to clean everything indefinitely.
Not every dataset needs to be perfect. The standard should reflect the consequence of error. A low-risk internal drafting tool can tolerate different limitations from a workflow that changes a customer’s entitlement, price or contractual position.
3. Connections that support the real workflow
An AI prototype can work with uploaded files and copied data. A live service usually needs controlled access to existing systems and the ability to pass work between them.
That demands more than a one-off connector. Teams need to understand what happens when a source system is unavailable, a record changes midway through a task, an integration returns only part of the required data or a supplier modifies its interface. They also need to decide where an action is recorded and how it can be reversed.
The Public Accounts Committee’s 2025 report on the use of AI in government identified legacy technology and poor data sharing as risks to AI adoption in government. The lesson for business leaders is not that every old system must go. It is that integration constraints belong in the strategy and investment decision, not in a technical appendix discovered after approval.
4. Identity, control and resilience across the service
An AI-enabled service should not gain broader access merely because the technology makes information easier to retrieve or actions easier to automate.
Permissions need to follow the user, task and context. Sensitive information should remain protected across integrations and suppliers. Higher-impact actions should require appropriate approval, and the service should fail safely when a dependency is unavailable or an output falls outside agreed boundaries.
These controls need to work across the full journey. Securing the AI component while relying on shared accounts, uncontrolled exports or informal manual workarounds elsewhere creates an incomplete control environment.
5. The ability to observe, change and own the outcome
AI-enabled services will change after launch. Models, prompts, source data, connected systems, user behaviour and supplier products can all alter performance.
The organisation needs to see what the service is doing: whether people use it, where it fails, how much it costs, what humans override and whether the intended outcome is improving. It also needs a safe route to release changes and a named business owner who can decide whether to improve, expand, pause or retire the service.
This is where technical debt becomes an operating constraint. In Kyndryl’s global 2025 Readiness Report, based on a survey of 3,700 business and technology leaders, 35% of organisations not yet seeing positive AI return on investment cited integration as a leading challenge, while 22% cited technical debt. The survey is global rather than UK-specific, but it reinforces the practical connection between AI value and the ability to change the surrounding estate.
Do not turn AI readiness into a wholesale replacement programme
Once leaders see the dependencies, there is a temptation to conclude that the organisation must complete a major transformation before it can scale AI. That can create a different kind of paralysis.
Most organisations contain a mixture of systems. Some are fragile and expensive. Some are old but stable. Some perform their core job well but are difficult to connect. Others duplicate capabilities or support processes the organisation no longer needs.
Age alone is not a sufficient reason to replace a system. The relevant question is whether it can support the outcomes, level of change and control the strategy now requires.
For each important component, make one of four deliberate choices:
- Retain: keep it where it remains dependable, supportable and fit for the required use.
- Expose: create a secure, governed way for other services to access the required data or capability.
- Modernise: improve the component, integration or operating arrangement where a targeted change removes a material constraint.
- Replace or retire: act where the system creates unacceptable risk, cost or inflexibility, or where the underlying service is no longer needed.
This prevents two common mistakes. The first is the “digital veneer”: placing an intelligent interface over fragmented work while leaving staff to reconcile the consequences. The second is an estate-wide replacement programme so broad that business outcomes disappear beneath technical scope.
The objective is a coherent path from use case to dependency to investment.
Knowing what needs to change is only the start. Organisations also need a practical way to turn those decisions into a funded delivery plan, while keeping each AI use case connected to the systems, information and teams it depends on.
Build a minimum viable digital core around priority outcomes
A minimum viable digital core is the smallest reusable set of foundations needed to deliver the next group of valuable AI-enabled services safely and well. It is not a shortcut around architecture or governance. It is a way to focus them.
Begin with two or three strategic outcomes rather than a catalogue of technologies. For example: reduce the time required to resolve complex customer requests; improve the quality and speed of commercial proposals; or identify operational exceptions before they cause loss.
For each outcome:
- map the service and the decisions it contains;
- identify the systems, information, integrations, permissions and teams it depends on;
- find the constraint most likely to prevent reliable delivery;
- decide which capability should be retained, exposed, modernised or retired;
- design the foundation so it can support more than one isolated use case;
- deliver the enabling change and the AI-enabled service as one joined programme.
This approach keeps the investment connected to visible value. It also creates evidence. Leaders can see whether an integration is reusable, whether a source of truth is genuinely reliable, whether a new control works in practice and whether the operating team can support the service.
The sequence matters. Modernising a system without a clear service outcome can produce a technically cleaner estate with little business change. Building an AI experience without strengthening a critical dependency can produce early excitement followed by operational workarounds. The two need to move together.
Build the digital foundations your AI strategy needs
Calls9 starts with your business priorities, not a pre-selected AI tool. Through focused interviews and workshops, we identify the outcomes that matter and where AI could add real value. We then map the processes, systems, data, integrations, permissions and controls behind the strongest opportunities. Together, we decide which use cases to pursue, what could prevent them from working and what should be retained, connected, modernised or replaced. You leave this stage with clear priorities, ownership and a practical roadmap linking expected value to the technology, governance and delivery work required.
From there, Calls9 can help turn the roadmap into a working service. Depending on what the evidence shows, we can select and integrate an existing product, build a tailored solution or modernise the parts of your digital core holding it back. We test with real users and representative data, measure accuracy and business impact, put the right controls and support in place, and prepare your teams through practical training. We can work as an extension of your team or lead the delivery, keeping strategy, technology and adoption connected from the first decision through to release and ongoing improvement.
* This articles' cover image is generated by AI
