One business ran 238,000 Instagram DM conversations in six months. 4,200 became sales. 80% of revenue traced back to those messages. No setter team. No commission structure. Never offline.
The difference was not a better chatbot. It was three layers of AI running behind every message.
This skill builds those three layers. It works on Instagram, WhatsApp, or any platform with DM API access. Give the skill folder to Claude or Codex and say: "Set up the DM analyst agent for my [platform] account."
The Problem: Your DM Bot Replies. That Is All It Does.
Most DM automations stop at the response. A lead messages you. The bot qualifies them with two questions. The lead books or drops off. The system resets and does it again tomorrow.
Three things never happen.
The bot never figures out which conversations became sales. It sends the same replies whether they work or not.
Nobody reads those DM logs and tells your website, email, or content team what to change. The intelligence inside every conversation evaporates by the following Monday.
Nobody runs outbound on its own. If leads stop coming in, the pipeline empties and waits.
A chatbot does one thing. This skill does three.
The DM Analyst Agent: Three Layers
Layer 1: The DM Agent
Connects to your DM platform via API or webhook. Runs continuously. When a message arrives, it reads the conversation history and your offer context, then qualifies the lead with three questions: what they do, what problem they are trying to fix, and whether they have tried to fix it before.
Every conversation logs a structured summary: message content, question asked, objection raised, outcome (replied / interested / not interested / booked).
Layer 2: The Business Analyst
Runs once a week. Reads every conversation log from the last seven days. Finds the top three questions people asked, the top two objections raised, and which approaches correlated with positive outcomes.
Then it writes a brief with three action items. One for your website agent. One for your email sequence. One for your content calendar. The format is specific: "Update X to address Y because Z of conversations mentioned it before dropping off."
The brief saves as a markdown file your other agents read on their next run. Your entire front end adjusts based on what the DMs revealed, without you touching anything.
Layer 3: The Outbound Researcher
Runs once a week. Takes the profile of your best-converting conversations (industry, role, language patterns, objections they overcame) and searches public accounts or profiles that match.
Returns a scored list of 10 to 20 prospects with one buying-signal note per entry. You review the list before any outreach fires.
How to Use It
Download the skill folder. In Claude Code, move the unpacked folder into ~/.claude/skills/dm-analyst-agent/. Then open Claude Code and say: "Set up the DM analyst agent for my Instagram account."
For Claude on the web or desktop, upload the zip under Settings then Capabilities, start a new chat, and give the same instruction.
The skill walks you through each step. It tells you exactly where to add your API credentials. Never paste a token or key into the chat.
Try It: Five Things to Say First
- "Set up the DM analyst agent for my Instagram account."
- "Run the business analyst on my last week of DM logs."
- "What were the top questions people asked before they booked?"
- "Find me 15 prospects that match the profile of my last 10 closed deals."
- "Write the brief for my website agent based on this week's conversations."
What You Get Back
- Every DM handled, qualified, and logged. No lead disappears because it came in at midnight.
- A weekly brief that tells your website, email, and content agents exactly what to update. Conversion data feeding back into your marketing stack on its own.
- A fresh prospect list every week built from the profile of people who actually bought.
- Full conversation logs with outcome data. You can read what worked before shipping any change.
- A setup that runs without a setter, without commission, without time zones.
The Limits You Should Know
The business analyst needs at least five logged conversations with outcome data before it produces a useful brief. Let the DM agent run for a few days first. Launch the analyst on day one and the brief comes back thin.
The outbound researcher returns a list for your review. It does not send outreach automatically. That is by design. A prospect list you have not checked is a deliverability risk. Review before anything sends.
Platform connection requires API or webhook access. Instagram requires a business account with an approved app. WhatsApp requires the Business API. Full setup steps for each case are in the skill's troubleshooting file.
If you do not yet have a website agent or email agent, the Layer 2 brief saves to a Notion page or shared folder until you do. The DM agent layer runs on its own regardless. Building the conversation layer first gives the analyst something to measure from day one.
One Thing to Do Now
Install the DM agent layer. Get the conversation logging running.
The analyst and researcher are already in the same package. They activate once the logs are there to work with. One download, one instruction to Claude, and the first layer is live today.
If it does not fit your setup after a week of real conversations, uninstall in under a minute. That is the whole risk.