Monitoring Instagram comments and DMs for every brand you manage

A workflow runs on a schedule for each brand’s Instagram professional account. The Instagram block lists comments on recent media and lists new direct messages, then reads each one. Run it every hour in working time to begin with, and more often during launches.

Each comment or message becomes a row in a mentions table in Tables: brand, where it came from, the text, the author’s handle and a link back to the post. A brand column on every row lets row-level access keep each brand’s team to its own rows.

How are complaints told apart from noise?

An Agent block in the same workflow classifies each row: sentiment, topic, and one class from a list you agree, such as complaint, question, lead, praise or spam. Write the classes down with two or three real examples each. Most of the accuracy comes from good examples, not from a bigger model.

For D2C brands in India, add the topics that drive most complaints: delivery delays, cash on delivery problems, wrong item, refund status. A comment like "COD order cancelled without asking" should be a complaint about delivery, not a general negative comment.

Who writes the replies, and who sends them?

The workflow also drafts a reply for anything that needs one, using the brand’s reply guide in the Knowledge Base: tone, what the team may say, and what it must never say, such as promising a refund in public. For complaints, it drafts two things: a short public reply that moves the conversation to DMs, and a private reply asking for the order number.

Nothing is sent from the drafting step. A second workflow starts with a Human block that shows the community manager the comment and the draft. They approve, edit or reject. Only after approval does the Instagram block post the reply to the comment, send a private reply, or send a DM. If personal details such as phone numbers turn up in messages, a Guardrails block can mask them before text reaches the next step.

When does a surge become an alert?

Put a named measure in Metrics: complaints per brand per hour. Add a monitor with a threshold per brand. A small brand might alert at five complaints in an hour, a large one at forty. Monitors re-check as often as every five minutes and alert by email, in the app and in Slack.

The monitor alert can also start a Relay called Social Insights. It reads the recent complaint rows, groups them by topic, and writes a short note: what people are complaining about, which post it started under, and whether the same issue appears in DMs. The note arrives in the Relay’s chat with a notification, and the Relay can email it to the brand lead after approval.

What goes into the weekly insights?

On a weekly schedule, the same Relay summarises each brand: volume by class, the top three complaint topics, questions the team kept answering, and leads that came in through comments. It drafts the summary as an email to the brand lead. Relays never send email without an approval, and the recipient policy can stay on internal only.

A dashboard on the mentions table in Dashboards shows the same numbers week by week, so the summary has a chart behind it.

What should a team try first?

Start with one brand and comments only. Run classification for a week without drafting replies and check a sample of rows by hand. Then switch on drafted replies for questions, which are low risk, before complaints. Add DMs last, once the reply guide has settled. For the weekly numbers in client reports, see client reporting.