AI-powered feedback summaries replaced 20 manual reviews a day

The Client

STN Digital is a digital media agency specializing in sports and entertainment brands. Their creative teams produce high-volume content on tight timelines, managing production in Wrike and communicating in Slack. When a piece of creative moves to review, the relevant context—prior feedback, revision notes, approval status—needs to follow it. At STN’s volume, that handoff happens dozens of times a day.

‍

The Challenge

STN had already automated the trigger. A Zapier workflow fired when a Wrike task moved to “Submitted for Review” and posted a notification to Slack. But after the notification landed, a team member still had to open the Wrike task, read the full comment thread, identify which comments were creative feedback, and manually summarize them in the Slack thread. Every time. For every task.

At ~20 tasks per day, this was cumulative friction: time lost, inconsistent summaries, and people pulled away from higher-value work. The feedback was in Wrike. It just wasn’t reaching Slack without a human in the middle.

The Solution

Streamline built an n8n workflow on STN Digital’s own cloud instance that replaces the manual step entirely. The automation triggers on the same “Submitted for Review” status change and follows three steps:

• Data Retrieval —  n8n fetches all comments from the Wrike task via the Wrike API.
• LLM Processing —  A custom-engineered prompt isolates creative feedback—revision notes, edit requests, design direction—and formats it as a structured summary. General project comments are excluded.
• Slack Delivery —  The summary posts as a reply in the existing Slack thread tied to that task.

The LLM prompt was the critical piece. Wrike threads mix creative direction with logistics, status updates, and conversation. Streamline iterated on the prompt across multiple rounds using STN’s production data until the output consistently met the team’s accuracy and formatting standards.

The existing Zapier automation stayed active throughout the build. No gap in STN’s notification flow. The n8n workflow cut over only after the team confirmed it met their acceptance criteria.

‍

What We Delivered

• n8n Workflow —  Production-ready automation on STN’s instance. All credentials stored in their environment.
• LLM Prompt —  Custom prompt refined iteratively against production data.
• Documentation —  Architecture overview, node-by-node breakdown, and troubleshooting guide.
• Testing Report —  QA summary confirming accuracy at ~20 triggers/day.
• Post-Launch Support —  14 days of bug fixes and minor adjustments.

‍

Outcome

The manual step is gone. When a task hits “Submitted for Review,” the creative feedback summary is already in the Slack thread by the time a reviewer opens it. Consistent format, every time. No one reads through comment threads or writes manual summaries.

The workflow runs on STN’s own infrastructure with no data leaving their environment. The summarization logic lives in a prompt the team can see and adjust—not a black box, but a tool they own.