Auditing 1,100 Avito Chats with AI: Where Leads Slip Away and How to Write a Bot PlaybookArticle · digest for October 22
A desk with an open laptop and a smartphone displaying incoming nighttime chat messages on the screen.
Article3 sections≈ 4 min

Auditing 1,100 Avito Chats with AI: Where Leads Slip Away and How to Write a Bot Playbook

Picture this: your Avito campaign is dialed in, paid promotion packages are running, and views and chats trickle in every day. Your contact budget keeps ticking down, yet sales are stuck in neutral. It's tempting to blame the platform: "Junk traffic, tire-kickers, slow season." But the moment you crack open the actual chat logs, an uncomfortable truth often surfaces: the ads did their job, and the sales team fumbled half the hot leads right inside the inbox.

Classifieds practitioner Alexander Gorbachev put that hunch to the test with hard numbers. He took a lumber supplier (timber, boards, moldings) where the average ticket is high and the sales cycle is short, exported 1,100 real chats, and ran them through Claude for a deep audit.

Anatomy of a Leak: Three Black Holes in the Inbox

The AI filtered out spam and isolated 510 genuinely qualified commercial inquiries: requests for specific wood grades, delivered price quotes, and cubic-meter calculations. That is where the numbers got ugly.

First, in 41% of qualified chats (209 out of 510), the potential buyer had the last word. The customer asked a concrete question about placing an order—and the rep went silent. Even worse, staff had opened 90 of those chats within the first 30 minutes, read them, and promptly forgot about them. The lead stalled out, and the buyer moved on to the next seller.

Second, 35% of incoming traffic landed outside business hours—after 7:00 PM. For evening messages, the median time to first reply hit 10 hours. Someone planning a build in the evening is ready to order right then, yet they don't hear back until the next morning, long after they've already made a deal elsewhere.

Finally, the audit uncovered a glaring gap with repeat customers: out of 87 returning buyers, 61 still had an unresolved question sitting in their past chat history. Instead of building long-term loyalty, the business was leaving repeat orders on the table.

Integration Architecture: Wiring Avito to AI via API

You can't force human reps to babysit chats 24/7. Alexander took the engineering route: he connected the Avito business account to an AI assistant through the official Avito Messenger API.

Here is how the pipeline works:

  1. On an "Extended" or "Maximum" plan, you generate a client_id and client_secret pair in your dashboard so the app can obtain an OAuth token.
  2. You register a Webhook on your server: whenever a message arrives, Avito sends a JSON payload with the chat ID, message text, and an x-avito-messenger-signature header.
  3. The model checks current stock sheets and pricing tables on Yandex Disk and drafts a reply.
Anti-fraud and reply speed

Never configure an instant 0.1-second reply. Avito's spam filters quickly flag unnatural bot-like speed and restrict the account for spamming. Always build in a 5–15 second delay with typing-indicator simulation.

Four Rules for the AI Sales Assistant's System Prompt

To keep the assistant from firing off useless boilerplate, Alexander baked four rules into the system prompt:

  • After-hours capture and spec gathering: outside business hours, the bot keeps the buyer engaged, greets them, and collects order parameters (wood species, grade, dimensions, volume, delivery address). For standard orders it gives a quick estimate; for custom ones it states the exact time the morning rep comes on shift.
  • Telling a closed chat apart from a dropped lead: the model distinguishes a wrapped-up conversation ("Thanks, got it") from an unanswered question ("Can you deliver to Serpukhov?"). Open threads get tagged and routed to a human rep.
  • Conversation memory: when a buyer returns, the AI reads the chat history and recalls their previous order.
  • Stock alternatives: if a requested item is out of stock, the bot suggests the closest matching grade or dimension with an updated price.

Automating the routine doesn't replace a skilled human salesperson—a neural network isn't going to negotiate custom wholesale discounts. But it seals the leak on hot after-hours leads for good. For more on designing chat-bot logic, read our article on building an AI salesperson for Avito chats.

Want a deep diveVotes help us choose what to break down in detail. This material has 0 votes.
Sources · 2