Ordering materials from a quote with an AI agent, keeping the human in the loop
Between the quote a client accepts and the materials delivered on site, there is a thankless step every tradesperson knows: going through the quote line by line and re-typing it into a supplier order. Time, in the evening, and a risk of error at every reference. We helped a construction tradesperson with exactly this. The result: an AI agent that reads the quote, places the order, and only disturbs him when a real decision is needed. Here is how, and why the “human in the loop” changes everything.
The problem: from quote to order, by hand
A partition job starts from a quote: so many plasterboard sheets, tracks, studs, screws, filler. Once the quote is approved, that list has to become a supplier order. It is repetitive, often done late, and that is where errors creep in: a forgotten quantity, a wrong reference, an incomplete delivery that stalls the install.
The tradesperson did not want one more piece of software to fill in. He wanted the quote he already produces to trigger the order directly — with no re-entry.
Our approach: an agent that reads, orders and flags
An AI agent is not a magic button. It is an assistant you hand a clear task to, that acts on rules decided with you and knows when to stop and ask. We built it around three moments.
Reading the quote. The agent takes the quote as it is — the tradesperson’s usual format — and pulls out what matters: products, quantities, references. Plasterboard, tracks, screws: each line becomes an order line, with no manual copying.
Placing the order. From that list, the agent prepares the order with the right supplier and sends it. What used to eat an evening now happens in the background, from a document the tradesperson had already written anyway.
Flagging a mismatch. This is the heart of the system. If the price found at the supplier does not match the one in the quote — a rise, a swapped reference, a discount that vanished — the agent does not order in silence. It stops and warns.
The “human in the loop,” where it counts
When a price drifts, the agent sends a WhatsApp message to the tradesperson: here is the line, the expected price, the actual price, the gap. And a simple question: order anyway, look for another supplier, or wait?
That is the “human in the loop”: the machine does all the grunt work — reading, comparing, preparing — but the decision that commits money goes back to the human, at the exact moment it arises. The tradesperson answers in a word from the job site, and the agent carries it out.
The detail that changes everything: he is only asked about the exception. As long as prices match the quote, the order goes out on its own. The moment there is a doubt that costs money, he is the one who decides. Neither blind autopilot nor manual labor — the right balance between the two.
What stays fully human
The agent does not invent a tolerance threshold, does not “negotiate” a price and never decides on extra cost alone. Those rules — from what gap to warn, which suppliers are acceptable — are set with the tradesperson, and the final call is always his. The model executes fast and tirelessly; it does not sign the purchase order in the boss’s place.
That distinction matters on a job site, where an ordering mistake is paid in days of waiting. An agent that asks beats an agent that gets it wrong with confidence.
What it changes in practice
The benefit is not a gadget, it is less load and fewer errors. The move from quote to order, which used to nibble away evenings, becomes nearly invisible. And the risk of a nasty surprise at delivery — a price that doubled without anyone noticing — disappears, because the gap surfaces before the order, not after the invoice.
In all honesty, we will not quote a time-saved figure: it would be made up. What we can say is that the logic has changed. The tradesperson no longer re-enters anything and no longer discovers price gaps too late; he decides, at the right moment, only on the cases that deserve it.
Why it works for a tradesperson
Many automations fail because they try to do everything alone, or because they demand too much data entry. The right approach is in between: hand the repetitive work to the agent, keep the committing decision for the human. WhatsApp, already in the tradesperson’s pocket, is enough as a control point.
The takeaway: if a task starts from a document you already produce, follows clear rules and includes a point where you sometimes have to decide, it is an ideal candidate for a “human in the loop” agent. Ordering materials from a quote ticks every box. If this sounds like you, let’s talk: the right automation is not the one that replaces you, it is the one that gives you back time and calls you at the right moment.