Where beauty brand DMs break down
If you sell skincare or cosmetics, Instagram DMs are probably your largest inbound channel and your tightest bottleneck at the same time. The same questions arrive all day: is this in stock, what is good for oily skin, can I order straight from the DM, do you ship to my city. Each one is easy on its own. Together they are more conversations than a small social team can hold.
The loss is invisible because it never shows up as a cancelled order. A DM you answer the next morning is a customer who has already bought somewhere else. Before you change anything, look at what share of yesterday’s DMs got a same-day reply, that number is usually the honest description of the problem.
What the agents actually do
A workable setup on Instagram is three MyChatBot agents rather than one catch-all bot. A Sales Agent answers product questions and carries the conversation towards an order. A Catalog Search Agent, connected to your full range through the Product Feed in the Knowledge Base, resolves requests like "something for sensitive skin under 500₴" from real catalogue data instead of guessing. A Follow-up Agent picks up anyone who goes quiet mid-conversation.
Give all three one instruction set built in the Configuration Wizard. That is what keeps the brand voice identical whether a customer asks about ingredients or about shipping, and it means you edit tone in one place rather than three. If you also book in-store consultations, connect Google Calendar or Altegio so the same conversation that answered a skincare question can offer a slot without leaving the DM.
How to set it up, step by step
Start by connecting the Instagram account and filling the Knowledge Base: Business Domain docs for policy, returns and delivery, the Product Feed or Product Spreadsheet for the catalogue. Then spend real time on tuning, teaching the agent your product language, the quirks of your categories, and exactly where its authority ends. Battle-test every version of the instruction in the Configuration Wizard before you point live DM traffic at it.
The biggest lever is hand-off discipline. With Hand-off Control and a stop-word, the agent handles routine questions end to end but routes anything sensitive, reactions, complaints, bespoke gift orders, straight to your human team, who get a Telegram ping via Flight Control.
What changes in practice
The mechanical change is simple: no DM waits. Answers arrive in seconds at any hour, which removes the whole category of loss where a question goes cold overnight. Your social team does not shrink, it stops doing data entry and starts handling the conversations the agent passes over, the ones where a person genuinely helps.
The second change is visibility. Every closed conversation lands in your CRM through create_crm_deal with the chat link attached, so you can finally see what Instagram DMs are actually producing instead of guessing. Add Follow-ups from Lead Processing and the people who asked about a product and drifted away get one more nudge instead of silence.
What makes it work
Three things, none of them magic. The agent answers from real catalogue data, not vibes. It sounds like your brand, not like software. And it knows when to step aside. All three are configuration choices you control, and they are the difference between an assistant your customers trust and one they route around.
If you run Instagram at volume, the setup above is repeatable: connect the channel, load the catalogue, tune the voice, set the hand-off rules, then let Follow-ups do the rest.
A real customer story on this: DMs and comments answered in seconds across 4 networks.