A full system stood up in six weeks, not a capital programme

Why implementations overrun

Implementing a system is usually a big project, and it almost always takes longer and costs more than expected. Three stages carry most of the overrun.

Requirements collection. Months of workshops documenting what people say they do, which is rarely what they actually do. The gap surfaces after go-live, as workarounds.

System selection. Whether a system can genuinely do what the operation needs is guesswork until you have paid to find out. Demos show the happy path.

Migration. Moving the data and the processes across is consistently the most underestimated part of the plan.

Meanwhile the teams build manual workarounds in Excel, the promised benefit never fully arrives, and the next upgrade becomes another major project. Customer opportunities are lost while the programme runs.

What we did instead

We augment the systems a client already runs with AI agents, or implement new ones with AI doing the heavy lifting. Either way the stages change shape.

Requirements are read, not collected. AI analyses what people actually do, independently, from the work itself and from the existing code. The output is a requirements set grounded in observed behaviour rather than workshop recollection.

Functionality is tested rather than guessed. Requirements are mapped against what candidate systems can genuinely do, and AI exercises them to find the gaps before a contract is signed.

Implementation is faster. AI supports the setup, configuration and customisation, and carries the migration. At this client a full system was stood up in six weeks.

What it changes for the business

The cost moves from a capital programme to an operating budget. That is not only a finance point: it puts the decision with the people who feel the problem, and lets the system keep evolving after go-live instead of freezing until the next programme is approved.