Your Factory Already Runs on AI — You Just Don’t Call It That – Rob Jolliffe of Sabre Limited, Part 3

Rob Jolliffe, Sabre Limited — Business Central Extensions & MRP as the Original AI
The Integrate Intelligently Podcast — with Scott Brickler (CADTALK) and Rob Jolliffe (Sabre Limited)

Rob Jolliffe, CEO of Sabre Limited, joins Scott Brickler to talk about how Business Central extensions coexist without breaking core functionality, why MRP and finite capacity planning were arguably manufacturing's original AI, and where AI in manufacturing actually has room to add value beyond what ERPs already do.

Speaker note: lines are attributed to Scott and Rob based on context — Scott opens the conversation and repeatedly references his own past ERP consulting work (the slitting manufacturer, wine label slitting, Sightline); Rob carries the deep technical walkthrough of the Real Steel product and Business Central's extension model, and later references having worked in manufacturing three decades ago. A few exchanges in the middle move quickly enough that the attribution is a best guess — worth a quick check on your end before this goes out publicly.

Introduction and the Business Central Extension Model
Scott: How do you do these integrations? Just selfishly, on our part, these integrations tend to work together pretty well because there's a crossover where they kind of run —

Rob: As a general rule, I think they do, because what I've seen from most add-on vendors for Business Central is that they're trying not to interfere with anything Microsoft did. What we've done is try to act like macros on top of the standard Business Central functionality — streamlining it, making it easier to use, or adding custom fields so you can collect extra data. Like a planning line in a project not having a drawing ID field — so we added a drawing ID field. That's a pretty logical thing to add if you're loading a bill of materials.

If you're not trying to twist what Microsoft did and make it work differently, everything works great together, because you're all going into that same base, respecting the basic functionality Microsoft built in. If I write something expecting it to work the way Microsoft intended, and you wrote something that leaves it working the way Microsoft intended, we'll work together just fine.

There are a couple of exceptions. There's a product we sell for the print manufacturing industry — a really unique, strange industry. If you've ever worked in print manufacturing, they sit somewhere between craftspeople and manufacturing companies — it's actually very sophisticated, and I've learned a ton about it over the last few years. Because that industry is so unique, this vendor's product made some changes that don't exactly align with how Microsoft does things. We're actually one of the only resellers in North America who sell it, and because of that, I now have to go back to the vendor and ask, "how do you do that, what's your approach," because I know what our stuff can do, but if I try to plug something else into theirs, it might not work — they haven't followed the standard Microsoft path.

Scott: They're exceptional though — almost nobody does that.

Rob: Well, yeah — because if you're going to build like that... I used to do a lot of ERP customization myself, back when I was doing sideline consulting years ago. You're essentially building better interfaces over the standard functionality. We worked with a slitting manufacturer that had a lot of lot-tracking requirements, and we ended up building an alternate order entry interface for handling slit rolls and the associated tracking.

Rob: Can I bring up something else — do you mind if I mention a product I don't actually sell?

Scott: Sure, sure, go ahead.

Rob: So I'm going to pitch somebody else's product here. There's a product called Real Steel, built into Business Central, for things like steel slitting — I know it well because we shared a mutual customer. What they did was take the standard sales order screen and add an alternate order entry section right onto it, just like what you described.

How Add-Ons Coexist Without Breaking Core Functionality
Rob: When you were taking an order for a customer's slit coil, you'd fill it in on this extra section, click a "transfer to order" button, and it would push that data into the standard sales order lines. It didn't replace those standard lines — they were still there, working exactly the way they would in any standard Business Central sales order. But a little configurator popped up to let you configure the slit coils, and if you configured a coil in a way that hadn't been done before, it would create that item on the fly — because there are literally millions of variations of how a slit coil can be specified.

It even tracked things at the sales order level using custom user fields — things like "eye to the sky" versus "eye to the horizon," meaning whether you want the coil sitting with the circular hole facing up or lying so the hole is horizontal. That distinction matters because some companies' cranes or forklifts need the coil oriented a specific way to pick it up properly.

This company enhanced nearly every screen of Business Central it could, without changing the underlying functionality, and built a full steel processing solution on top of it. The company's called Walcott Group, and Kevin owes me a beer for mentioning them. It's a great example of an add-on that fundamentally expanded what Business Central can do and which industries it can serve, without overriding any of Microsoft's default behavior — which means it still works fine with a credit card processor, or a FedEx-style rate-shopping app doing LTL, FTL, or carrier rate comparisons, because none of that got rewritten.

Scott: We call that a pipeline integration approach — exactly what you're describing. We did something similar for wine label slitting: built all the screens on top of the ERP feeding into the normal data tables, so you're really just adding a step ahead in the pipeline.

The Real Steel Example: Extensions in Practice
Rob: Right — and it's worth remembering, when you're editing inside Business Central, you're not running a separate application sitting as a layer above it. You're filling in information that flows into Business Central from beneath or alongside it — you're inside Business Central, with an extra section embedded right into the app. I don't know if that's what you did with yours.

Scott: Actually, the same thing — though you could also do that with Sightline at the time. You could build custom screens that feed into the standard data.

Rob: That model of integration or augmentation is a beautiful way to do it, because you're feeding into the same pipeline rather than changing the core functionality. Everything lives in one database.

The other thing about the Business Central model: if you installed Real Steel, used it for a while, and decided you didn't like it, you could pull it out of Business Central and everything would revert to the original standard Business Central. The data you created — the items, the records — would still be there, but they'd be missing all the custom user fields that were part of Real Steel. Maybe those matter, maybe they don't. But the beauty of Microsoft's extension model is that it allows that kind of customization without touching the original, so you can plug it in and take it out like a Chrome browser plugin — install a password manager, decide you don't want it anymore, uninstall it, and Chrome reverts to normal. Business Central works the same way.

Scott: That's beautiful — because really what they're doing is building functionality above the data. If you're doing slitting on a normal order entry screen, it would be horribly inefficient — a nightmare. So you build a layer above it to automate what would otherwise be a manual mess, but the underlying data is still there. If you removed the add-on, you'd still have all the coils that were ordered on that order, sitting there as they would have been — you'd just lose the friendly configurator you used to create those lines. You could still hand-build a BOM in an Excel sheet and paste it into Business Central if you had to. So really, like I said, it's macros layered on top to make things easier — taking something that would take 20 steps and bringing it down to three, because you've automated the process.

MRP as the Original AI: Why Manufacturing Already Has Expert Systems
Rob: And then you add AI on top of that, and we don't even fully know yet what it's going to be able to do — but it's certainly going to get integrated. One last thing before we run out of time, on AI and ERP and manufacturing specifically: technically, if MRP counts as an expert system — which I'd argue it does — and finite capacity planning is an expert system too, which I'd also argue is true, then manufacturing has had AI, if "expert system" and "AI" are roughly synonymous, longer than almost any other industry.

Scott: I'd agree with that — maybe airlines too, given how long automated scheduling, automated planning, and that kind of system-driven output has existed there.

Rob: And even things like visual inspection — I remember working for a company 30 years ago experimenting with visual inspection systems that could pick defective parts off a line and let the good ones flow through. Part of what causes confusion for people following the AI conversation in manufacturing is that outsiders come in saying, "we could help automate your purchasing," and manufacturers are like, yeah, it already does that — or "we could help optimize your machine setups," and again, yeah, we already have that, as long as you feed it the data. But AI needs data too.

I think the real opportunities in manufacturing are elsewhere. One example: manufacturers commonly set purchase order approval thresholds — say, $10,000 — because approving every single PO manually would be unworkable. Someone in a management position checks anything above that threshold, usually asking things like: is there a better vendor, did we actually research who we should be buying from, could this part go obsolete soon. I think AI could handle that, and you could push that threshold down to $5 purchases, because AI can be trained to check things like whether a part has been dropped from a newer engineering revision in product lifecycle management, whether the part is essentially a commodity whose market price has dropped even though the purchased price hasn't caught up — things that would be far too onerous for a human to check on every low-dollar purchase order, but that's exactly the kind of thing AI should be doing in manufacturing.

Scott: Right — not the stuff CADTALK already handles, like pulling engineering drawings into the bill of materials, or running MRP or production scheduling. It's these other things we haven't fully thought about yet. That's a very Eli Goldratt kind of point — he talks a lot about how the obvious use case for a new technology usually isn't the actual use case that ends up mattering.

Why Current AI Pitches Miss the Mark with Manufacturers
Scott: No, I agree completely — we should have a longer conversation, because there are things we're working on that I think you'd find interesting. I think it's more about the workflows than the obvious use cases — where humans are still directly involved is where all the value is. People think the opportunity is in the algorithms, but manufacturing is a keystone industry — there's been more work done on production planning algorithms than almost anywhere else, and that's been true forever.

Rob: Right, these are all expert systems — everybody thinks of LLMs as "the thing" now, but expert systems have been the original AI all along. The part that's actually new is things like purchasing workflows and supply chain decisions, where the volume makes it prohibitively hard for a human to do consistently, but trivially easy for an ERP paired with AI — sourcing vendors, switching materials on the fly, suggesting alternates to a shop floor worker when inventory runs low and updating the bill of materials to account for the swap before it's even used. Those are the things humans still do today — time-consuming and important, but we limit how much of it we do because of that time cost. Throw some of that at AI. Let humans keep doing the $10,000 approvals if that still feels necessary, but let AI take on the things you've never even bothered doing.

The Real Opportunity: Tasks Too Onerous for Humans
Scott: I saw a genuinely interesting AI solution that emails vendors directly and asks, "is your PO still on time, and if not, what date do you expect delivery?" When the vendor replies, that flows straight into the ERP — it tells the buyer the vendor's going to be late, and updates the ERP's due date on the purchase order, which then informs planning and everyone else downstream. That's a really handy use of it.

Rob: Oh yeah — you can go really deep with that kind of thing.

Scott: I'd love to swap notes on some of what we're working on.

Rob: You'll have to invite me back, Scott.

Scott: Absolutely, absolutely. I'm glad we got to meet on this call — similar backgrounds, both excited about the same things, both big Eli Goldratt fans. This is a good match, I'd love to talk more.

Rob: That'd be awesome.

Scott: I know we're well over time — I could talk forever, but I don't want to burn up yours.

Rob: I do have to run, unfortunately.

Scott: I'm sure you do. We'll let you go, but I really appreciate the time today, Rob — this has been awesome. Where are you located, by the way? I know you're in Canada.

Rob: I'm in Guelph, Ontario — about 35 miles west of the Toronto airport.

Scott: Not far at all — short flight for us, we're in Cincinnati.

Rob: You guys aren't far at all. I'll come visit sometime — I want to catch a ball game there someday.

Scott: Maybe not this summer, but a couple summers out we're planning a tour — Detroit on the way, then Cincinnati, then somewhere else on the way back. We've done Cleveland and Boston so far, then COVID hit and we stopped traveling, but the goal was to hit every MLB park.

Rob: You should come out to Great American Ball Park — I can actually see it from my new house.

Scott: We'll get you out for a game — you'll have to buy me a drink first, though.

Rob: Oh, hell yeah, we'll do that. I'm a huge Reds fan from way back — I've got Pete Rose stuff on my walls, so we'll get along fine.

Scott: All right, Rob — thanks so much, this was a real pleasure.

Rob: Same here — thank you so much.

Scott: All right, thanks, Rob.