3, September 2026
by: Grace McCarthy
“AI becomes another garbage in, garbage out tool if you don’t get your data right first.”
Jason Conway has spent almost his entire career at Marston’s, and nearly all of it in data, even though that’s not where he started. Now leading Applications, AI & Innovation at the pub company, he’s spent the last several years building the single source of truth that Marston’s runs on today, and more recently, working out what AI actually looks like once that foundation is in place. In this episode of the Arreoblue podcast, he spoke about how that foundation came together, why adoption was always harder than the technology, and why he’s in no rush to let AI move faster than the data underneath it.
Marston’s is a pub operator with around 1,300 sites across the UK, many of them tucked into housing estates rather than town centres, a long way from its old life as a brewery. Jason started out as an accountant there and moved into management accounting before realising the two disciplines had more in common with data than he’d expected. “Management accountancy, at least my understanding of it, was effectively to analyse data to help the business make good decisions for the future,” he said, “which is pretty much also the definition of business intelligence,” and that overlap is what pulled him across.
He went on to spend five years running Marston’s BI function before taking on applications and AI earlier this year, a shift he describes as building on top of the work already done rather than starting something separate.
The biggest change Jason has seen in his time at Marston’s isn’t a piece of technology, it’s how data driven the business has become. Finance ran largely on spreadsheets when he started, and while some of that still happens, the shift toward a single source of truth has gone far enough that BI is now where people instinctively go to find their numbers.
It wasn’t really a technical problem so much as an adoption one, and weaning people off the spreadsheets they’d built their working habits around took real effort. Jason’s boss described the early stages as “trying to push water uphill,” and for a while that’s exactly what it felt like, until a point came where people started seeing the value and began going there naturally. These days, he said, “if something’s wrong in BI, people get pretty annoyed about it,” which he takes as a good sign even though it doesn’t always feel like one at the time.
Winning over the last of the holdouts eventually meant a harder call, stopping spreadsheets from going out at all. “It was in the end a little bit forced,” Jason admitted, but it’s what pushed the final bit of adoption over the line.
Marston’s is now looking at putting an AI layer on top of its BI reporting, something that can take a mature set of dashboards and turn them into a short, prioritised digest of what someone should actually focus on that morning, rather than leaving people to work it out for themselves. It’s a natural next step, but only because the groundwork was already there, which is a point Jason kept returning to throughout the conversation.
He’s said as much himself, and he’s glad Marston’s spent the time getting BI right before trying to layer AI on top of it. Bringing in third party data reliably, transforming it properly and understanding what it actually means is harder than it looks from the outside, and skipping that step tends to catch up with you later.
Part of the conversation turned to a problem that sounds small until it isn’t, since an AI model asked about ROI has no way of knowing whether someone means return on investment or the Republic of Ireland unless the organisation has done the work to describe what its own data actually means. That semantic layer, unglamorous as it sounds, is what stops an AI tool from confidently giving the wrong answer.
As Jason put it, “nothing kills adoption like someone who might already be cynical getting a clearly wrong answer,” which is why he’s cautious about releasing AI tools before they’re ready. “Go slow to go fast,” he said, borrowing the well-worn phrase, adding that “I think it does actually ring true here.” In practice that means putting guardrails in place and encouraging people to stay a little sceptical while the tools are still finding their feet, while accepting there’s only so much any guardrail can catch.
When asked what great data teams will be doing in five years that today’s teams aren’t, Jason expects “much less people sitting down writing code,” with a lot more time spent producing good specs for agents to work from, assuming the economics hold up. On advice for a business starting its data transformation, he kept it simple: “get the foundations right, don’t rush into it,” since shortcuts tend to come back and bite later, a point that carried straight through to the next question. Asked to choose between data quality and AI, he didn’t hesitate: “obviously quality.”
The habit he thinks every data leader should build is patience, particularly around foundational decisions that are hard to unpick once they’re made, and the best career advice he’s ever received followed a similar logic. Rather than chasing the right path, he suggests ruling out the wrong ones: “it’s easier to rule things out than it is to choose the right path from the outset.”
Sound familiar? It’s a conversation we’re having with a lot of teams at the moment. We’re always happy to talk through where you are and what the next step looks like.
Have a question or want to make an enquiry?
Contact us
