Instagram DM automation for agencies: a playbook

Original scope diagram for an illustrative agency package. Agree real commercial terms separately.

Start from the client's task. "Deliver the preparation guide to people who ask under the launch post" tells you what to build and tells the client what to expect. "Build a chatbot" does neither. Want to try the workflow on your own account first? Start free and build one campaign before you price anything.

Split the launch package from the operating service. Launch work creates and verifies one campaign. Maintenance keeps agreed parts current. Neither should quietly grow into unlimited campaigns or answering every DM for the client.

Estimate effort from the real tasks, including communication and testing. In an invented example, discovery takes 1.5 hours, copy and mapping 1.5, configuration 2, testing 2, one revision round 1.5 and handover 1.5, so 10 hours in total. Replace every figure with your own records.

List the client inputs with an owner and a due date: account authorization, the final resource, verified offer facts, a working destination and the person who receives enquiries. Running several client profiles? The multiple accounts guide covers account slots and shared usage, and the social media manager page covers day-to-day client work. If you also recommend the tool to clients, the referral programme guide explains how to join and disclose commission.

Build a short review pack in reading order: the post draft, the opening DM, each next step, the resource filename or version date, and the destination URL. Ask the reviewer to follow each route as a customer would. "Does this look good?" invites opinions about colours and misses wrong promises. Keep API keys, contact exports and real conversations out of the pack, and run the security checklist before you give a client any access.

Name one accountable approver for the offer and use a readable version label, such as "North Street beginner guide, review 03". Record each decision as one of three states:

• Accepted: no open conditions, with the version and date recorded.
• Changes required: each change listed with an owner.
• Accepted with conditions: not final until someone shows the conditions closed.

A specialist check, such as a teacher confirming session content, is useful but is not overall approval. Keep two status lines before publishing: content accepted and implementation checked. A client can approve the wording while the setup still needs testing.

A late change to a price, date, resource or destination reopens review for the affected route only. In ReplyMagnet, replacing an asset changes future deliveries but does not rewrite links already sent, so decide whether earlier recipients need a correction.

Followers, raw comments and billable engagements are different quantities. Copy the remaining allowance and reset date from the workspace you will actually use, not from a remembered pricing table. Then build low, central and high cases for every campaign sharing that allowance. The figures below are hypothetical.

Compare each total with the real remaining allowance. If the central case nearly uses it, act before the next promotion post goes out. Over-limit work waits as pending instead of sending, and a request that waits too long can pass its messaging window before it is released.

Name owners for usage checks, the resource, customer exceptions and the decision to pause a promotion. Write the trigger in advance, for example: if observed demand makes the remaining allowance insufficient, the promotion owner reviews the next scheduled post before it goes out. After the period, record actual usage beside each forecast and note why they differed.

Write a tool-independent map of the existing campaign before rebuilding. Similar-looking settings in two products may not behave the same, so mark anything the new tool does not support as a decision, not a silent drop. Contacts, tags and conversation history do not move just because both tools connect to Instagram.

Then work through the switch in order:

1. Preserve the old configuration and your notes.
2. Rebuild one simple, single-resource campaign as a draft.
3. Confirm which connected app is responsible for answering, so two tools never race on the same comment.
4. Switch off the old overlapping trigger at the agreed time.
5. Test from an account you control, plus one unrelated comment that should stay out of scope.
6. Restore the old setup if the wrong resource arrives, replies duplicate or captured details reach the wrong destination.

After acceptance, check older posts whose captions still invite the old request. Links the old tool already sent stay with the old tool, so decide what happens to them before that subscription ends.

Write the jobs first, for example: deliver a requested checklist, route a relevant question to an answer, and let the owner maintain the setup without the original freelancer. Then classify each requirement as a gate, a weighted tradeoff or out of scope.

With ratings of 3, 2 and 1 on the weights below, an invented candidate scores 3×3 + 2×2 + 1×1 = 14. Mark untested items as not verified rather than 0 or full marks, and never fill a result from a sales page. Run the same acceptance task in every trial, performed by the person who will operate the tool.

Compare limits in each vendor's own unit. A contact-based allowance counts something different from an engagement, as this free-plan limit breakdown shows. If a client wants a custom build instead, the n8n versus managed tools guide maps the hosting and maintenance work you would then own.

Write the learning objective first, for example: "The viewer can see that a WORKBOOK comment returns the workbook link." Then prepare the data as carefully as the script:

• Comment from a separate test account, because ReplyMagnet ignores comments from the connected account itself.
• Give demo resources titles that match what the DM promises.
• Close unrelated tabs and silence notifications before recording.
• Keep credentials, client inboxes, emails and private messages out of the shot.
• Say when footage is shortened or reordered.

Read every overlay as a claim. Words like delivered, booked and subscribed need evidence that the recording actually shows. A headline such as "comment to customer in seconds" promises an outcome no demo can establish. For the build itself, follow the comment-to-DM setup guide and record a real test from the second account.

Keep a short production note with the recording date, the mode shown and the demo data used. Recheck the video when the product or the workflow changes.