From flyer to listing: an AI agent that reads property flyers

Property flyers are designed for people, not software. The price is in a corner, the bedroom count is an icon, the address is split across two lines, and the photos are part of the layout. For one of our clients, that meant every listing started with someone reading a flyer and typing it into the CRM by hand.

We built an AI agent that does that job.

What it does

The agent takes a flyer — a PDF, a scan or a photo — and returns a structured listing ready for the CRM: address, price, property type, bedrooms, bathrooms, size, key features and description, plus the property photos cut out of the flyer as separate images.

It reads the flyer the way a person would, using a vision-language model that understands both the text and the layout, then maps what it finds onto the client’s listing fields. Every listing is reviewed by a person before it goes live, so the agent does the typing and people keep the judgement.

Why a local model

The agent runs on Qwen, an open-weight model, hosted on the client’s own infrastructure rather than a public AI API. That choice did three things:

• The data stays in-house. Flyers carry addresses, prices and contact details. None of it is sent to a third-party AI service.
• The cost is flat. There is no per-page API bill, so processing a thousand flyers costs the same as processing ten.
• Nobody else can change it. The model does not get silently updated underneath the client, so the output stays consistent.

What changed

Listing entry went from a manual typing job to a review job. The team no longer re-types flyers; they check what the agent produced and approve it.

The pattern is not just for property

Any business that receives documents designed for humans — invoices, delivery notes, application forms, supplier catalogues — and re-types them into a system has the same problem. The same approach works: a vision model to read, an agent to structure, a person to approve.

Bring a document your team re-types every day to stand B1350 and we will show you what the agent would pull out of it.