Olympic Glass, the Sheerness-based IGU manufacturer, rolled out Lumeso’s AI-driven order capture in five months and has not looked back. Glass Times editor Luke Wood sits down with Gary Jenkins, MD of Olympic Glass, Adam Roda, operations director, and Georg Katzlinger-Söllradl, CEO of Austrian software firm Lumeso, to talk about what AI in the order office actually looks like once the slides come down and the work starts.

Luke Wood (LW): Gary, Olympic Glass has been running a successful processing operation in Queenborough – why bring an AI tool into the order office at all?
Gary Jenkins (GJ) – Because the maths stopped working. We run a very broad product mix, thousands of orders a week across IGUs, toughened, laminates, the lot. Every one of those orders has to be keyed in correctly or the factory pays for it downstream.

We were spending a lot of evenings and weekends on data entry, frankly, no one wants to do that at seven o’clock at night. We have always tried to be early on the technology side – that is how we have grown – so when we started seeing what AI could actually do with a messy input data, we went looking.

LW: Adam, was the operations team on board from the start?
Adam Roda (AR): The order processors had been saying for at least a year that the volume was not sustainable without either more headcount or a different way of working. And qualified order-entry people are not easy to find – it is a skilled job, you need to understand glass.

So when we showed them a tool that promised to do the routine 80% and leave them the interesting 20%, the reaction was, frankly, “when can we start?” That mattered. A lot of software projects die because the people who have to use it were not asked.

LW: Georg, five months from first conversation to running in production is fast for industrial software. How was it possible?
Georg Katzlinger-Söllradl (GK): Two things. First, Olympic Glass arrived prepared. Their data was in good shape, their team was committed, and Gary and Adam made the decisions quickly when decisions needed to be made. You would be surprised how often a project stalls because nobody is willing to own it.

Second, our integration with LiSEC GPS Order is now well-trodden. The first ERP connections we did anywhere were genuinely painful – I will admit that openly. But by the time Olympic Glass came on board, we knew the GPS interface, and LiSEC knew us. That saved months.

Luke Wood: Adam, what did the five months actually look like from your side?
AR: Roughly, the first four weeks were the technical connection to GPS and getting the two systems talk to each other. Then I ran it in shadow mode: I tested everything myself, processed incoming orders alongside the team, and evaluated the results.

That phase showed me how the system really learns your customers’ quirks. One customer writes “4/16/4” and means one thing, another uses different words to describe the exact same products. I quickly saw the benefits and gave it to the team. By month five, the majority of our incoming order lines were going through with the team checking rather than typing.

LW: Gary, the headline number people will want is hours saved. Can you provide a figure for that?
GJ: I will give you the one that matters to me: we have eliminated regular overtime in order processing. That is not a small thing. It used to be a structural feature of how we ran – evenings, Saturdays, end-of-month crunch – and it is gone.

The team take their days off. They take their holiday when they want to take it. One of them told me, unprompted, that this is the first year he has not dreaded Mondays. You cannot put that on a spreadsheet but it is the thing I notice most.

LW: And on the spreadsheet?
GJ: Return on investment inside six months. That includes the integration cost, the internal time we put in, the lot. After that the system pays for itself every month, and because Lumeso charges per order line we process, the cost scales with the work.

Quiet month, smaller bill. Busy month, bigger bill, but we also did more business that month. That model suits me much better than a big upfront licence.

LW: Adam, is the team honestly any they better off?
AR: Honestly, yes. The job changed. They are not typing the same dimensions into a screen all day, they are checking the AI’s work and handling the unusual orders. That is a more interesting job, and frankly a more skilled one.

Nobody has left the team since we went live, which in this industry tells you something. We have also been able to take on more volume without hiring on the order desk – which, given how hard recruitment is right now, is worth a great deal.

LW: Was there anything that did not go to plan?
AR: Honestly, the main thing is what the system does not do yet. Georgian bars and shaped units – it does not handle those today, and a fair slice of our mix involves one or the other. We knew that going in, Lumeso were upfront about it, and the update is on the roadmap.

So for now, those orders still go through the team the old way, and everything else runs through Lumeso. We are eagerly waiting for that release, because once it lands we close the last real gap in coverage.

GK: Adam is right, and we are deliberately open about it. Georgian bars and shaped units are genuinely harder – the geometry and the way they are specified vary far more from customer to customer than a rectangular IGU, and we would rather ship support that works reliably than something that looks clever in a demo and falls over in production.

Both are on the roadmap, shapes first, and Olympic Glass is one of the customers we are building against. That is how we prefer to work: tell the customer honestly what the system does today, what it does not, and when the gap closes.

LW: A question I imagine you get a lot, Georg – a CEO reads about AI, looks at the bill for specialised software, and thinks his IT person could knock this together over a few weekends with one of the AI coding tools. What is your answer?
GK: My answer is: build the small things yourselves, absolutely. A dashboard, a report, a little connector between two systems – today an in-house IT person with a tool like Claude Code can produce that in an afternoon, and that is genuinely useful.

But there is a difference between a prototype and a product that runs your order desk five days a week, integrates with your ERP, handles edge cases sensibly, and is still being maintained in three years’ time. The prototype is the easy part. Living with it is the hard part. I would ask the CEO: is the person who built it still going to be at your company in five years, and still excited about maintaining it?

LW: Gary, did you have that conversation internally?
GJ: We did. And we concluded that we are a glass processor, not a software house. I would rather pay people who do this all day, every day, for hundreds of customers, than have one of my people try to keep up with where AI is going on their own. The technology is moving fast – too fast for a part-time effort.

LW: Georg, would this work as well for glass processors on other systems than LiSEC?
GK: We are deliberately ERP-independent. We have customers running on GPS, on A+W Business, on Hanic Optiplus, on Synerglass Symbiose, and on others. Our specification is open – any ERP vendor who wants to connect to Lumeso can do so, and we align with VDMA 24124 (OPC UA), the industry standard.

But the wider message I would give any processor reading this is: ask your ERP supplier how open their system is. If the answer gets vague, that is a warning sign. We are past the point where one system can do it all, and the next five years in this industry will reward the processors whose core systems let them plug new tools in.

LW: Final question. Where does the data sit, and is Olympic Glass’s data effectively training a tool that helps its competitors?
GK: European servers, full GDPR compliance. And every customer has their own isolated learning environment. What Olympic Glass teaches Lumeso about Olympic Glass’s customers, article numbers and conventions stays with Olympic Glass. It would not be usable by anyone else, because the article IDs and the customer language are specific to them.

AR: That mattered to us. We checked it carefully before we signed, and we are satisfied with how it works in practice.

LW: A question for each of you: what should UK glass processors be doing in the next 12 months?
GJ: Stay on the front foot with technology. The tools are moving quickly, and the processors who keep experimenting will be in a stronger position than those who wait. We have more in the pipeline ourselves – this was the start, not the finish.

AR: Get your master data in order. Whichever tools you bring in next, they will only be as good as the data you give them. Tidy article numbers, clean customer records. That is the work nobody wants to do but everyone benefits from – and it is what lets you move quickly when the next thing comes along.

GK: Free someone in your organisation up to drive digitalisation. Not as a side project – properly. And experiment with AI yourself, on small internal problems first. It demystifies the technology and you quickly learn the difference between what is worth building and what is worth buying.