The AI tools everyone has tried, chat assistants and AI features in office software, work on whatever you paste into them. The AI that really changes how a business runs works on your own data and your own processes: orders, customers, cases, stock, documents. To do that, it has to connect to the systems where that data lives.
For a lot of UK businesses, that’s where AI plans stall. The government’s AI Adoption Survey 2026 found that 54% of businesses planning AI investment intend to embed AI into their existing systems. Yet research reported by IT Brief in December 2025 found that 61% of organisations say technical debt from legacy platforms now blocks their efforts to integrate AI. The same research found that while 88% of organisations had a two-year roadmap to modernise their IT, only 36% said those plans were fully funded.
The good news is that you rarely need to replace everything to get started.
Signs Your Systems Are in the Way
You don’t need a full audit to know whether your systems will hold up. These are the warning signs we look for:
- No way in. The system has no API, so the only way to get data out is a person exporting a spreadsheet.
- Data in too many places. The real picture is spread across an old database, a few Access files and someone’s Excel workbook.
- Unsupported technology. The application runs on a framework or operating system that no longer gets security updates.
- Nobody fully understands it. The original developer has moved on, the documentation is thin, and changes feel risky.
- Inconsistent data. Duplicate customers, free-text fields used for everything, and codes that mean different things to different teams.
- No record of changes. You can’t tell who changed what, or when, which matters as soon as AI starts acting on the data.
Any one of these makes AI harder. Several together mean that an AI project will spend most of its budget on plumbing.
Why a Big-Bang Rewrite Is Usually the Wrong Answer
Faced with an ageing system, the tempting answer is to replace it. Sometimes that’s right. More often, a full rewrite takes longer than planned, costs more than budgeted, and stalls the business while it happens. We have written before about why IT projects fail, and rewrites carry many of the same risks.
The alternative is to modernise in steps, delivering something useful at each stage and leaving the core system running until each part is safely replaced.
A Step-by-Step Path to AI-Ready Systems
1. Understand what you have
Map the systems, where the data lives, how it flows and who depends on it. AI tools now make this faster than it used to be: they can read a large, unfamiliar codebase and summarise what it does, which shortens the discovery stage considerably.
2. Open a door with an API
Put a modern, secure API in front of the existing system. It can read from and write to the old database without changing the application itself. This one step makes the data available to AI, to other systems and to future replacements. We build these regularly as part of our API development work.
3. Clean and consolidate the data
AI is only as good as the data it works with. Merge the duplicates, bring scattered spreadsheets into a proper database, and agree what each field means. It’s unglamorous work, but it pays off far beyond AI. Our database development team spends a lot of time here.
4. Add AI where the value is clearest
With data available through an API, you can build AI features on top without touching the core system: reading incoming documents, answering staff questions from your records, drafting responses, flagging exceptions. Using the Model Context Protocol (MCP), you can also give AI agents controlled access to your systems, with clear limits on what they can see and do.
5. Replace the old parts gradually
Once the API is in place, individual parts of the old system can be rebuilt one at a time behind it, moving to supported technology such as .NET 10 without users noticing the switch. This is often called the strangler fig approach: the new system gradually grows around the old one until the old one can be retired.
What It Looks Like in Practice
| Stage | What you get | Typical timescale |
|---|---|---|
| Discovery | A clear map of systems, data and risks | Days to a few weeks |
| API layer | Secure access to existing data for AI and integrations | Weeks |
| Data clean-up | Reliable, consistent data in one place | Weeks to months, often in parallel |
| First AI feature | A working AI process on your real data | Weeks once data is accessible |
| Gradual replacement | Old components retired one by one | Ongoing, at your pace |
Each stage delivers value on its own, so you can pause, reprioritise or stop without having wasted the investment.
How We Can Help
We have spent more than 25 years building and supporting long-lived business systems for UK organisations, including keeping critical platforms running while they are modernised around their users. Today that work increasingly ends with AI: once the data is accessible and trustworthy, we build AI features and agents on top.
If you have an older system that is holding back your plans, whether for AI or simply for growth, get in touch. We will help you work out the smallest sensible first step.