An AI Sales Agent That Runs an Entire Publishing Network's Inbound Revenue
An AI email agent that handles every inbound sales conversation across 60+ web properties, from first enquiry through pricing, objections, fulfilment, invoicing, and upsell, with no human in the loop for the standard deal.
Where they started.
The client operates a large network of owned news and blog properties, 63 sites in total, and sells editorial placements on them. Every one of those sites has its own dedicated sales inbox, and every inbox receives a steady flow of enquiries. Historically this meant a team of people reading emails, quoting prices, negotiating, handling objections, collecting content, chasing invoices, and trying to upsell, across dozens of sites at once. It was slow, inconsistent, expensive, and impossible to scale without adding headcount linearly.
The goal was blunt: replace the entire inbound sales function with a single AI agent that could represent every site correctly, hold a natural conversation, close on fixed commercial terms, and never lose a thread.
What we did.
The core insight was that this is not one agent, it is one brain wearing 63 faces. A single conversation engine sits behind every inbox. When an email lands, the system detects which site it arrived at and loads that site's identity, pricing tier, and rules before the agent ever composes a reply. The customer always feels like they are talking to the specific publication they contacted.
Pricing was fixed and tier based rather than negotiable, which removed the biggest risk in sales automation: an agent giving away margin. Objections are handled not with discounts but with value, by adding complimentary placements from entry tiers. The whole sales motion was broken into swappable stage modules (first contact, objection handling, fulfilment, payment, upsell) so the agent always knows exactly where a conversation sits and what it is trying to do next.
The architecture is hub and spoke. Each of the 63 inboxes runs a lightweight listener that does one job: catch new mail and hand it to the core. All 63 feed a single main processor that carries the sales logic, the stage flow, and the reasoning. Two supporting workflows run alongside it: a timed chase engine that queries unpaid threads hourly and escalates through a fixed reminder sequence, and a payment webhook that catches confirmation, reinstates any paused placements, updates the client record, and triggers the upsell.
Everything the agent knows lives in a structured data layer: a multi table base holding sites, clients, live sales records, outstanding invoices, bundles, retainers, the stage modules themselves, and a full thread and message history so the agent has perfect recall of every prior exchange. Risky accounts are flagged to invoice before fulfilment rather than after. The conversation engine was chosen for speed and cost at volume, because inbound replies are high frequency and need to feel instant.
What changed.
The system runs the standard deal end to end with no human touch, at a total operating cost of around 105 dollars a month. That figure replaces a sales and admin function that previously cost between 9,000 and 9,500 dollars a month in wages. It manages a book of over 4,300 client records and thousands of historical orders without dropping context, mixing up threads, or misquoting a single site. Because pricing is locked and objections resolve through added value rather than discount, margin is protected on every conversation the agent handles.
The economic shape of it is the headline: a two figure monthly cost doing the work of a four figure monthly team, across 63 storefronts at once, day and night.
Built with n8n, Claude Haiku, Airtable, IMAP / SMTP email, PayPal API.
AI Automation
This is AI Automation applied to revenue: one system that reads every enquiry, quotes, handles objections, invoices, and upsells with no human in the loop. We can build the same inbound engine around your offers, your pricing, and your inboxes.