At Kirkstall Precision Engineering, innovation doesn’t only happen on the machine floor. Sometimes it happens in the systems, processes and operational thinking that keep a precision medical manufacturing business moving efficiently, compliantly and with confidence.
Impactful use of AI
For Sarah Wood, Operational Excellence Manager, that mindset recently led to a practical and highly impactful use of artificial intelligence, transforming a complex, resource-heavy compliance task into a streamlined digital workflow that saved weeks of manual effort.The challenge centred around training matrices and competency records, a vital requirement for maintaining quality standards and supporting ISO 13485 compliance. While externally delivered training naturally came with certificates, many of the most valuable skills within the business were developed on the job: internal systems, process handovers, works order issuing, shop floor procedures and role-specific operational knowledge.
Bringing records up to date
These competencies were already captured in Kirkstall’s matrices, but the supporting evidence trail needed to bring the records fully up to date represented a significant administrative burden.
“There are a number of administrative tasks within a business like KPE that are both very resource-intensive and burdensome,” Sarah explains. “But at the same time, they’re very high value, because we need them in order to meet our requirements under ISO.”
With hundreds of discrete skills spread across multiple teams and employees, the scale was substantial.
“We identified that we had a lot of these skilled on-the-job trainings recorded in our competency matrix,” she says. “Across all employees and all these discrete skills, we were looking at well over 800.”
An opportunity
ParagrapRather than accept the task as a month of manual quality administration, Sarah saw an opportunity.
Having spent the last two years developing her understanding of different AI platforms, from Copilot and Gemini to Claude and ChatGPT, she had built a strong sense of where AI could add genuine value and where it couldn’t.
“One of the things I’ve learned over two years is that I now have an understanding of which tasks are AI-able, and which ones aren’t,” she says. “You start to understand where you get real bang for buck.”h
The breakthrough
The breakthrough came from the structure of the existing data. Sarah first reviewed the six competency matrices to establish whether the information was consistent enough for AI to process reliably.
“To my absolute delight, they were all the same,” she says. “They were all written in Excel, all the specific data was in the same rows and columns. That meant I could teach the AI exactly where it would find the information every single time.”
From there, Sarah used ChatGPT to build a stepwise workflow. Rather than jumping straight into output, she began by testing whether the model truly understood the task.
“I generally explore with the AI first, does it understand the task?” she explains. “Because if it doesn’t, it will go off and do it wrong.”
Feedback loop
That deliberate feedback loop became central to the project’s success. Sarah created a memorandum template aligned to Kaleidex branding, mapped the relevant data fields from the matrices, and progressively refined the prompts through repeated testing until the outputs were accurate.
“It’s really a conversation,” she says. “It tells me what it thinks the task is, I amend it, then we run it once, run it twice, and improve it each time.”
After four or five iterations, the system began producing reliable outputs at scale. The result was striking. A task that would likely have required around a month of manual work from one person in the quality team was completed in approximately seven hours.
“The time it took in total to get good data and produce one good set, about 300 memos that were right, was around seven hours of my work,” Sarah says. “The manual alternative would probably have been around one month’s work.”
The final outcome
The gains didn’t stop there. Using Microsoft Power Automate, Sarah then built a secondary workflow to batch-convert the generated Word documents into PDFs, automatically naming and sorting them in a way that made filing into personnel folders fast and intuitive.
The final outcome was a fully digital, paper-free process that closed compliance gaps without drawing heavily on departmental resource.
“Now we’re up to date with our compliance, we have no gaps, we didn’t need to draw anybody’s resource, and we didn’t generate any paper,” she says.
For Sarah, the project also points to something much bigger: the role AI can play in making operational excellence more accessible across the business.
The next step
The next step is exploring how a dedicated business AI environment could be structured around departmental functions, allowing quality, operations and management teams to build knowledge over time while retaining appropriate controls.
“I think the value really is about rolling it out into the wider workforce,” she says. “AI is actually more accessible for people who struggle with normal systems, because essentially it’s just a chat function.”
Importantly, Sarah is clear that AI works best when paired with human judgment.
“There is a part of the loop which needs ‘belts & braces’ review before you actually go live with any data that it’s given you, you do need to sanity check it with a human eye.”
That blend of structured systems thinking, curiosity and pragmatic experimentation is exactly what operational excellence looks like in a modern medtech manufacturing environment.
At Kirkstall, it’s another example of how the right people, asking the right questions, continue to drive smarter ways of working.
