RSB Contracts fabricates and erects structural steel. We have been building their software alongside the people doing the work since November: automating BOQ inputs, building and revising tenders, generating fabrication packs, running monthly claims, tracking production across every live project, and matching purchase orders to supplier invoices. The same job data carries through each step, so the commercial, production, and finance teams share one view of what is happening.
The quotes below come from recordings of the weekly build calls with Richard Spencer (Managing Director), Ruan Swart (who built and ran RSB's production master workbook), and Demi Van Der Merwe (who runs the monthly claims). They are lightly edited for clarity.
The spreadsheet that ran the workshop
RSB's production lived in a master Excel list that Ruan built — and had to actively defend:
"I actually did a little macro to collapse columns within my Excel master. If Alex or any clever oke in the shop accidentally unlocked the sheet for editing, as soon as you click one of those, it locks them as well."
— Ruan Swart
Excel could hold today's numbers, but not the history behind them:
"The biggest trick in this — similar to what I faced in Excel — is storing the historic data. In Excel, that was a battle. Once you change it, whatever was in that cell is missing."
— Ruan Swart
And the data itself could only go so deep:
"Within my Excel, an employee is just a name that's being used multiple times."
— Ruan Swart
What the workbook was hiding
The most expensive problems were the ones the spreadsheet couldn't show. Steel gets re-fabricated when drawings are revised — and that rework was invisible to the people doing the invoicing:
"On our current master list there are a lot of items that are booked and unbooked, booked and canceled, where no one knows about it. Demi doesn't really see this document — so she'll claim for the beam once, where in reality we might have manufactured the beam twice because of revisions."
— Ruan Swart
"It actually happens more often than what we are aware of — we don't actually get to see it. Someone needs to be able to see it who's actually working with the financial side of things, not just the guys doing the bookings."
— Richard Spencer
Even the daily production numbers could drift:
"Today I pull tonnage and the tonnage is 20 tons. Tomorrow they just go and change their date — moving the produced mass to the following day because they might have had a bad day, just to even out the numbers."
— Ruan Swart
And keeping parallel systems in sync by hand was a losing game:
"To have to capture the same information twice is not going to be effective — the chances of it ever being exactly the same are super, super slim. You have to keep the two sets of data identical, otherwise you're always going to have a disaster."
— Richard Spencer
He tried to fix it himself first
Richard is not new to AI — he spent 100+ hours trying to improve the estimating spreadsheet with ChatGPT before finding LlamaPress, and kept hitting the same wall:
"I've got a few graphs that I wanted to be able to generate for me — I've had to go into ChatGPT and build things to do it."
— Richard Spencer
Their industry software couldn't do it either. When drawing-revision tracking came up, Ruan noted it was "the one item that I think even our PLM software struggles with." Richard's answer:
"We have absolute faith in you guys."
— Richard Spencer
What we built — and what happened when they saw it
Tenders, fabrication packs, monthly claims, and a live production tracker, all on the same job data. When the system compared two 17,000-line drawing-revision imports and flagged exactly what changed:
"Okay — on the second one the phase has changed, the third one has been removed... that's amazing."
— Ruan Swart
"That's cool, because it shows you: it was 705, now it's 702."
— Richard Spencer, watching an assembly's mass change between revisions
The locked-down web app ended the spreadsheet's fragility:
"It'll be amazing — now no one can screw around with the filters."
— Ruan Swart
And the workflow logic matched how the shop actually runs, stage by stage:
"That's exactly — that's brilliant."
— Ruan Swart, seeing the prep → fit-up → weld stages
The claims side got the same treatment. Demi described how the app mirrors their real tender-lock discipline: once a tender is awarded, the items lock, "and you can only overwrite it if you create a revision." Richard put it simply when RSB updated their ISO process documents to match the new system:
"The system is pretty much what we do — it's just better and more controlled."
— Richard Spencer
Where it's going
With hard date-stamps, revision history, and hours per welder in one place, RSB can finally ask questions the workbook never could:
"Just to have a cost per ton, and hours per ton, per boilermaker — because you've got cheap boilermakers that outperform the more expensive ones."
— Ruan Swart
"Then you'll start getting real feedback every day — and that's when the momentum is going to start to pick up."
— Richard Spencer
His verdict on the engagement, closing out one of the build calls:
"I'm just excited. The progress from November till now is massive."
— Richard Spencer
