When you walk into one of JENKI’s vibrant matcha bars across London, it’s easy to assume things are just as simple behind the scenes as they appear at the counter. But beneath the surface of every matcha latte is a finance operation quietly powered by automation, AI, and a healthy dose of self-taught Python.
In this week’s AI in Action feature, we spoke to Luke Simpson, Finance Director at JENKI, about how he’s using AI and code to reclaim hours from his finance team’s week, tackle reconciliation headaches, and explore the opportunities of machine learning in hospitality forecasting.
Matcha bars, margins and python
Luke joined JENKI in December, stepping into a multi-channel business selling matcha drinks in-store, online, and wholesale.
“The goal is always to keep the customer experience simple,” Luke explains, “but behind the scenes, things are far more complex: multiple revenue streams, various data sources, changing costs.”
Having taught himself Python years ago, Luke was well-positioned to automate the types of recurring tasks that often sap time in lean finance teams. And now, AI is helping him go even further.
From manual matching to automated journals
One standout use case? Automating the time consuming sales reconciliation process for Selfridges, one of JENKI’s bars.
Every week, the finance team receives a remittance PDF and a separate Excel sales report. The goal is to turn that into a clean, reconciled journal entry for Xero. Previously, this would have involved line-by-line checks, manual lookups, and more than a few hours of someone’s time.
Now, thanks to a system Luke built using Python and OCR (optical character recognition), aided by Cursor, the process is almost entirely automated.
- The script pulls product data from the Excel sheet and categorises
- It uses OCR to extract values from the remittance PDF
- It compares the remitted amount, sales report, and what’s recorded in Xero
- If everything matches, it auto-generates a clean journal entry
“What used to be a tedious manual check is now just me dropping the files into the right folder,” Luke says. “The script does the rest.”
It’s a perfect example of where AI plays a supporting role, eliminating the repetitive friction.
Using Cursor to speed up development (without sacrificing control)
A key part of Luke’s approach is leveraging AI tools like Cursor effectively.
““I used to spend hours on Stack Overflow, trying to stitch together snippets of code,” Luke says. “Now, I use Cursor to build things line by line. It helps me move fast, but I still do the thinking.”
That philosophy matters. Instead of throwing vague prompts at an AI tool and hoping for the best, Luke uses his existing knowledge to drive the process, validating code as he goes and debugging as needed.
“You can code now without understanding the language,” he explains, “but I’ve found that unless you check it properly, it just doesn’t work. Cursor speeds me up, but it doesn’t build it all for me.”
Forecasting habits (not just numbers)
While Luke is already using AI for reconciliation and reporting, he’s also exploring forecasting, an area he sees too often overcomplicated in hospitality.
“I think we underestimate how habitual our customers are,” Luke says. “Sales patterns are more logical than we realise. People will grab a matcha at set times of the day.”
Luke has built a basic model using past sales data to forecast daily revenue. While it’s not yet in production, it’s already revealing key trends and helping validate his belief that off-the-shelf tools often miss business-specific nuance.
A final word: Use the tools, but think for yourself
Luke’s story is a powerful reminder that AI works best when paired with human insight. Whether it’s building a reconciliation tool, forecasting sales, or speeding up repetitive reporting, it’s his finance experience, not just the tech, that makes the systems work.
“AI helps me work faster,” Luke says, “but I still do the thinking. That’s the bit that matters.”
Want to learn from others using AI in finance?
Join The Stack Exchange: Our free Slack forum where in-house finance teams share tools, test ideas, and compare notes on what’s working (and what isn’t) when it comes to AI in the finance function.
.png?upscale=true&width=1196&upscale=true&name=Untitled%20design%20(42).png)
