• Mark Williams, Chief Operating Officer at Sharp UK


• 26.06.2026 09:30 am
#ArtificialIntelligence #FinanceTransformation

Finance teams are not short of AI pilots. What they are short of is AI that actually sticks. Tools get trialled, results look promising, and then the rollout stalls, adoption slows, and the team reverts to the spreadsheet they have always used. Technology is rarely the problem; getting its implementation right is.

For finance functions, where accuracy is non-negotiable and the consequences of a data breach are serious, the hesitation is understandable. But caution without a clear path forward is just delay.

Start with the workflows, not the technology

The most effective AI deployments in finance do not start with a tool. They start with a process review: where is time being lost, where do errors creep in, and where are skilled people doing tasks that do not require their expertise?

Month-end reporting, account reconciliation, compliance processes, tax and treasury operations: these are the areas where AI consistently earns its keep. Not by replacing the judgement of finance professionals, but by handling the repetitive, high-volume tasks that create pressure at exactly the wrong moments. Getting clear on those use cases before selecting any technology is what makes the difference between a deployment that delivers and one that drifts.

Why finance teams are moving away from open LLMs

One shift worth paying attention to is the move toward private AI models trained on an organisation's own data, rather than open large language models. This approach, known as Retrieval Augmented Generation or RAG, keeps sensitive financial data within a controlled environment rather than feeding it into a public model. That matters both for data protection and output quality. The AI is working from your data, not generic training sets, which means responses are more relevant to the decisions you are actually making.

More organisations are building and training their own AI agents this way, giving teams the ability to adapt and refine them over time without depending on external platforms for every update.

Make prompt engineering a finance skill now

The most valuable AI skill a finance team can develop right now is prompt engineering, knowing how to instruct an AI tool clearly and effectively. It sounds technical, but it is closer to clear thinking and precise communication than anything else. Teams that treat it as a core competency, rather than something that sits with IT, consistently get better results. That means building it into training and making space for people to practice in real use cases.

The cultural piece most organisations underestimate

Getting finance teams genuinely confident with AI, rather than wary of it, is what separates a successful rollout from one that stalls. That requires more than a training session. It requires clarity on what AI is being used for, why, and what good looks like in practice. Involve finance teams from the start, set clear boundaries around AI-assisted versus human-led tasks, and build in time to get comfortable before scaling.

Done right, AI stops being something the finance team is asked to adopt and starts being something they want to use.