A Chatbot That Places the Order Itself: How AI Order Creation Works (Flow + Checklist)

A chatbot that places the order itself is an AI sales agent that doesn't just answer questions — it finishes the sale: it identifies the product, checks stock, builds the cart, collects the phone number and delivery details, creates a real order in the CRM, sends a payment link, and generates the shipping label. The defining difference from a regular chatbot is the outcome of the conversation: not a "lead" someone processes by hand later, but a finished order record in the CRM with line items, total, and delivery data.

For a store selling through Instagram Direct and Telegram, that means orders get placed at 11:40 pm and on Sundays — no "we'll reply shortly." The manager opens the CRM in the morning and sees ready orders marked "awaiting payment" instead of a wall of unread chats. We covered what an AI sales agent is in general in a separate article; this one is specifically about order creation: the full flow, step by step, plus a buyer's checklist.

The full flow: 9 steps from "Hi" to a tracking number

Here is what an order's journey looks like when an AI agent runs the conversation. We built exactly this chain at Mercon for product e-commerce, so this is from our own kitchen:

1. The customer writes in a messenger. Instagram Direct, Telegram, or Facebook Messenger: "Hi, is this serum in stock?"
2. The AI detects intent and identifies the product. The agent distinguishes browsing from comparing, objections from a clear buying signal. It finds the product via semantic catalog search — even when the customer writes "that retinol cream from your last reel," with no SKU or exact name.
3. It checks stock and price. Before promising anything, the agent verifies inventory. If the item is out of stock, it suggests an alternative instead of "confirming" an empty shelf.
4. It builds the cart. Adds the product, suggests a relevant companion item where it makes sense — a toner to go with the serum, not random spam — and calculates the total.
5. It collects delivery details. Phone number, city, pickup point or address — in Ukraine, Mercon's home market, that's a Nova Poshta branch. One question at a time, like a human manager, not a seven-field form dumped in one message.
6. It creates the order in the CRM. The moment of truth: a full order record appears in the CRM — line items, total, customer data, delivery method, pipeline status. Not an email notification, not a spreadsheet row.
7. It sends a payment link. In Mercon's case, Monobank or LiqPay — straight into the same chat. The customer pays without leaving the messenger.
8. It creates the shipment. After checkout, a shipping label is generated and the tracking number lands in the customer's chat.
9. It follows through to delivery. Automatic status updates ("your parcel has arrived at the pickup point"), unpaid-order reminders, and abandoned-cart recovery.

The key steps, unpacked

Steps 2–3: understanding the product, not matching keywords

Classic bot builders run on buttons and exact matches: the customer taps "Catalog," then "Creams," then scrolls a list. This breaks on the first natural-language request — "looking for something for dry skin under $20." Semantic search matches the meaning of the request against product descriptions, so the agent answers concretely: two or three relevant products with prices and stock status. And the reverse holds too: if inventory is zero, it says so and offers a substitute — the stock check is built into product selection, not left for a manager to discover in the morning.

Step 5: collecting details without a form

This is where buyers are easiest to lose: ask them to fill out a seven-field form and some will simply close the chat. So the agent gathers data the way a person would: phone first, then city, then the pickup branch. If the customer answers "same as last time" or mixes up the branch number, the AI asks again and validates the data instead of writing garbage into the CRM.

Step 6: the CRM record — the real test

"Our bot takes orders" in an ad can mean anything: a message to a manager, an email, a spreadsheet row. Real AI order creation means the system creates the order in the CRM via API — with line items, total, customer, and a pipeline status. In the Mercon + KeyCRM setup, the order also syncs to KeyCRM, so the store keeps working in its usual system of record. If this step is missing, you don't have sales automation — you have lead capture.

Steps 7–9: money, tracking, and the people who stall

An order created but not paid is not yet a sale. So the payment link follows the order immediately, and then the automation takes over: reminders for unpaid orders, abandoned-cart recovery, shipment creation, and delivery status updates. The buyer never has to ask "where's my parcel" — the bot messages them when it arrives.

Buyer's checklist: what "the bot takes orders" must actually include

When evaluating a platform, test mechanics, not promises. Here's the filter table:

The simplest verification is a test conversation on a demo: place an order as a customer, then ask to see what appeared in the CRM. If the "order" turns out to be a notification in a manager's chat, that's not AI order creation.

Why most bots can't do this

An honest market picture, as of June 2026:

• ManyChat — strong at broadcasts, comment automation, and audience growth, but it does not create CRM orders and bills per active contact. Paid plans start around $14/mo, and Manychat AI is a separate +$29/mo add-on. Its free relief program for Ukraine ended on January 31, 2026.
• SendPulse — a flow builder; its ChatGPT step requires your own OpenAI API key, and the funnels stay scripted: the bot walks branches but doesn't autonomously create a CRM order.
• Helpdesk AI agents (Intercom Fin at $0.99 per resolution, Gorgias AI Agent at $1.00 per resolved conversation, as of June 2026) — excellent at resolving support tickets, but they're priced and built around support resolutions, not around walking a buyer from "is this in stock?" to a paid order with a shipping label.
• A custom agency-built bot — setup starts around $200 with a Ukrainian agency (more elsewhere), and most often it's a hard-coded script for one scenario that needs rewriting every time the catalog changes.

When a human should step in

AI order creation is not a reason to pull managers out of chats entirely. Some conversations a human simply handles better, and an honest platform should say so:

• Custom requests: made-to-order items, engraving, wholesale with special terms — anything outside the catalog and knowledge base.
• Complaints and returns: an angry customer wants a person, and the reputational cost of a mistake is highest here.
• Off-policy discounts and B2B negotiations: decisions that touch your margin belong to the owner, not the model.

The right mechanic isn't "AI or human" — it's a handoff: the agent recognizes when a dialog goes beyond its limits, calls the operator, and hands over the full conversation context. In Mercon, the manager sees the entire thread in an omnichannel inbox and can switch the AI off in a specific chat with one toggle — and back on the same way.

What it costs

As of June 2026, the AI-conversation market sits in the $0.5–1.4 per dialog band: Intercom Fin charges $0.99 per resolution, Gorgias $1.00 per resolved conversation, Tidio Lyro from about $0.65. We price it differently at Mercon: Growth at $139/mo includes roughly 2,500 dialogs — about $0.06 per dialog — and with your own OpenAI or Gemini key (BYO) the same plan is $79/mo, with no markup on AI and no message caps. The full market math is in our cost-per-dialog breakdown, and current plans are on the pricing page.

FAQ

Does the chatbot really create a CRM order, or just collect contacts?

It depends on the platform — and this is the single most important pre-purchase question. Builders like ManyChat or SendPulse collect data and walk a script, but they don't create the order record in a CRM themselves (as of June 2026). An AI agent like Mercon creates the order via API automatically: line items, total, delivery data, and a pipeline status.

What does the AI do when a customer asks something non-standard?

A properly configured agent answers only from the catalog and knowledge base, and when a request goes beyond that, it calls a human operator and hands over the dialog context. Complaints, custom orders, and off-policy discounts should go to a person — that's a healthy, correct boundary for automation.

Which channels and integrations does this work with?

Mercon's AI agent works in Telegram, Instagram Direct, and Facebook Messenger. Orders are created in the built-in CRM and sync to KeyCRM; shipments are created with Nova Poshta (Ukraine's main carrier) with tracking numbers and status updates sent into the chat; payment links go through Monobank or LiqPay.

What counts as one AI dialog?

On average, one dialog is about 8 AI messages: greeting, product selection, delivery details, order confirmation. That's why comparing plans per dialog is more honest than per "message" or "reply" — converted to the same unit, the numbers differ severalfold.

Want to see this flow running on your own catalog? Connect Mercon: CSV/XLSX product import, Telegram and Instagram Direct, CRM order creation, payments, and shipping labels out of the box. Start a 14-day free trial, no credit card — one evening is enough to set up a working agent and run your first test order.