Naledi Freight, a mid-sized forwarder operating out of Durban, had a problem that didn’t show up in its pricing: speed. Compiling a multi-leg freight quote — rates, surcharges, currency, routing — was a manual job that could take up to three days. By the time a quote went out, faster rivals had often already won the tender.
What they built
Rather than buy an off-the-shelf product, the team assembled a workflow: a cleaned, shared rate database as the single source of truth, with GPT-5.5 used to assemble draft quotes from a structured brief. A human still checks and signs every quote — the model drafts, the operator approves.
We weren’t losing on price. We were losing because we were slow. That was the whole problem.
Thandi Nkosi, Operations Director
What worked, what didn’t
The first attempts failed because the rate data was inconsistent across spreadsheets. Only after a fortnight of cleaning records did the drafts become reliable. The lesson echoed across the business: the AI wasn’t the hard part — the data was.
South African context
The estimate workflow runs in the cloud, so a local power outage at the office doesn’t stop quoting. Rate data with client details stays in the firm’s own systems, never pasted into a public model.
