On the surface this looked like a standard manufacturing situation, and an off-the-shelf cloud system handles the standard version well enough. The real complexity sat elsewhere. Refractories are not standard goods. Each of the manufacturer’s hundreds of product grades corresponds to a specific composition, application, and customer specification. Those specifications are not metadata. They drive raw material procurement, they drive production planning, and above all they drive pricing, because every customer bid is built from the specific properties of the grade being quoted. The legacy system captured none of this. Its catalog was effectively flat, and pricing was being worked out entirely outside the system.

The bidding workflow turned out to be the single most complex thing in the build. We redesigned and rewrote its specification several times, including once after the system had already gone live, iterating with the client until it was right. A more conventional vendor would have called that scope creep. We treated it as the work, because the workflow could only be understood by building it against the real operation. That is exactly what the previous vendor had not done, and why their system handled billing but not the business. Most software fails before the build begins, in a discovery that never went deep enough.