AI Operations Co-Pilot for a Singapore Wholesale Distributor
100,000+ unstructured SKUs in SAP Business One. Sales reps quoting from memory. We're building a three-layer AI Operations Co-Pilot that parses the SKU catalogue, surfaces the right product in seconds, and writes orders back to SAP.
A business that had outgrown its own knowledge.
A second-generation Singapore wholesale distributor with decades of operating history. Over 100,000 SKUs in SAP Business One, accumulated across years of supplier catalogues, price changes, and product renamings. No consistent naming convention. The same physical product might exist three times in the system under three different SKU codes, with different units, different suppliers, and different price histories.
Sales reps quoted from memory. Senior staff carried the catalogue in their heads. New hires took six months before they could quote a customer without supervision. Stock decisions were made on instinct because the data wasn't trustworthy enough to drive them.
The owner was clear about the problem: the business had outgrown its own knowledge. The next generation couldn't inherit what wasn't written down anywhere a system could read.
Off-the-shelf ERP add-ons had been evaluated and rejected. None could parse the existing SKU mess without a year-long data cleanup project that the operations team didn't have capacity to run.
Don't clean the mess. Build a system that reads it as it is.
We proposed a three-layer AI Operations Co-Pilot rather than a data cleanup project. The thesis: don't force the team to clean 100,000 SKUs into a tidy taxonomy. Build a system that reads the mess as it is.
Layer 1 — SKU Intelligence
An AI parsing engine that ingests the SAP Business One product master and builds a semantic understanding of what each SKU actually is. A sales rep types "12mm clear glass tempered" and gets the right SKU — even though the SAP record reads "GLS-TMP-12C-001 · Asahi 12mm CL TMP." The system learns the language the team actually uses and bridges it to the language SAP stores.
Layer 2 — Sales workflow
A clean interface for reps to search, quote, and convert opportunities. Quotes pull live pricing, check stock, and surface margin. The rep never touches SAP directly during a sales conversation.
Layer 3 — Write-back to SAP B1
Confirmed orders flow back into SAP Business One as proper sales orders, preserving the existing accounting and inventory workflow. The ERP remains the system of record. The Co-Pilot is the system of work.
A working SKU parsing engine, in trial.
A working proof of concept for Layer 1 is in trial. The SKU parsing engine handles the full 100,000+ SKU catalogue and returns relevant matches in under a second for natural-language queries. Initial accuracy on a tested sample is high enough that the client agreed to fund a paid pilot before committing to the full build.
The trial is structured deliberately: a fixed-scope, fixed-fee paid pilot lets the team validate the technology against their real catalogue and their real sales workflows before the larger build is greenlit. This is how we recommend every AI engagement of this size starts.
The shape of most AI-for-SME opportunities in Singapore.
This is the shape of most "AI for SMEs" opportunities we see in Singapore. The data isn't clean. The processes aren't documented. The institutional knowledge lives in two or three senior staff. Traditional ERP consultants will quote you a two-year cleanup project. We think the better answer is to build software that works with the mess — and then lets the mess get tidier over time as a byproduct of usage.
The engagement is ongoing. We'll update this case study as the system enters full deployment.
Next case study
Operations ERP for a Glass & Façade Contractor
Construction · Custom ERP · In design
Start with a workflow audit.
Two weeks. We study your operations, map where software and agents create the most leverage, and show you exactly what a custom system would look like for your business — before you commit to building anything.
You walk away with
- A workflow map
- A system architecture proposal
- A phased build plan
- A fixed-price quote
- You don't need another SaaS subscription that almost fits.
- You don't need an ERP rollout that takes two years.
- You don't need to become a software builder yourself.
- And you shouldn't have to buy custom software from a static PDF.
Any workflow. Any system. any.software.