A client came to me needing multiple wholesale suppliers connected to their retailers — orders, stock, couriers, payments, the whole operational chain. The kind of project that, with a development team, is usually a one-to-two-year build. Not because developers are slow, but because a bespoke system like this needs constant iteration — off-the-shelf CRMs simply don't bend to how a real wholesale business actually runs.
It shipped in four to six weeks. Fully audited for compliance along the way, not bolted on afterwards. I want to be upfront about what I can and can't tell you here — by agreement with the buyer, this stays at outcome and process level. No product name, no code, no architecture. What I can talk about is the thinking, and that's genuinely the more useful part anyway.
Here's the bit that surprised me most, and it's the bit I keep coming back to with every client since: the hardest problem on this build wasn't the software. It was that the people who'd actually use it — suppliers, retailers, sales agents — are notoriously tech-averse. Ask any of them to upload a spreadsheet, or fill in a form to create an account, and they'll quietly go straight back to phone, pen, and paper. Every time. No exceptions.
So the entire design brief became: remove friction, don't add features. Accounts got created automatically. Login links went out that just worked — no password to remember or reset. The retailer view stayed deliberately, almost stubbornly simple. The supplier view, simpler still. Anything with real complexity — data insights, order tracking, courier integration, the full ledger — lives entirely in the admin layer, out of sight of the people who just want to place an order and get on with their day.
Courier tracking plugged straight in, so nobody's chasing a parcel by phone. Gap analysis flags a missing order before it becomes an angry call. A full ledger and communications trail means nothing quietly falls through the cracks three months later. None of this is exotic — it's the unglamorous, plumbing-level work that actually saves someone's Tuesday.
The AI-assisted part wasn't "type a prompt, get a CRM." It was iterative, disciplined, and documented at every stage — the same practice I bring to my own work now, and the same one I'm building toward automating (with a human always checking before anything ships, same rule as everywhere else on this site).
The platform's since been sold. The buyer's onboarding onto it now. That's not the point I want to leave you with, though — the point is that "agentic AI" isn't a slogan when you can point at something real, sold, and in use, and say honestly what it took to build it.
This isn't the limit of what's possible — it's a way of showing you how I think. The pattern (audit the real bottleneck, design out the friction, keep a human on anything that matters) is the same one I'd apply to your business.
Read the full case study