# How Midjourney Built a $500M Business by Listening to Discord -- And What Product Teams Can Learn > Discord communities generate 1.1 billion messages per day. The product signals buried inside -- feature requests, bug reports, use case discoveries -- represent the largest untapped source of community-driven intelligence. Here's why most product teams can't keep up. --- [Discord](https://closedloop.sh/tag/discord)[Community Feedback](https://closedloop.sh/tag/community+feedback)[Product Intelligence](https://closedloop.sh/tag/product+intelligence)[Customer Conversations](https://closedloop.sh/tag/customer+conversations)[Community Management](https://closedloop.sh/tag/community+management) # How Midjourney Built a $500M Business by Listening to Discord -- And What Product Teams Can Learn Oct 12, 2025 12 min read ClosedLoop AI Team Discord communities generate 1.1 billion messages per day. The product signals buried inside -- feature requests, bug reports, use case discoveries -- represent the largest untapped source of community-driven intelligence. Here's why most product teams can't keep up. On this page On this page In 2022, a small team of eleven people launched an AI image generation tool with no marketing budget, no venture capital, and no standalone application. Their entire product lived inside a Discord server. By 2025, that company -- Midjourney -- had grown to $500 million in annual revenue, making it one of the most capital-efficient technology companies in modern history. The standard narrative credits Midjourney's success to the quality of its AI model. That is only half the story. The other half is what happened inside their Discord server: every single user interaction became a visible, analyzable product signal. Every `/imagine`command was a public data point. Every complaint, workaround, and feature request unfolded in real time, in front of the entire team. Midjourney did not just build a product on Discord. They built a feedback engine that no traditional product analytics stack could replicate. The lesson extends far beyond Midjourney. Discord now hosts 32.6 million active servers and processes 1.1 billion messages every day. For software companies, game studios, developer tool makers, and an expanding universe of non-gaming businesses, Discord communities have become the primary venue where users talk about products in their most unfiltered, detailed, and honest form. The product intelligence buried inside these conversations is staggering in both volume and quality. Most product teams cannot extract it. ## Discord's Quiet Transformation Into a Business Platform The perception of Discord as a gaming chat app is several years out of date. As of Q2 2025, the platform reports 231 million monthly active users across 689 million registered accounts, with projections exceeding 300 million MAUs by Q4 2026. The server count has surged 380% since 2020, reaching 32.6 million active communities. Revenue hit $561 million in 2025, reflecting 29.2% year-over-year growth. More telling than the raw numbers is the demographic shift. Forty-six to fifty-four percent of Discord's user base now identifies as non-gamers, and 78% of all users report using the platform for activities unrelated to gaming. Developer communities, open-source projects, SaaS companies, educational institutions, and creator economies have all adopted Discord as their primary community hub. Thirty million people use AI-powered tools on Discord monthly -- a number that continues to climb as companies embed bots, integrations, and automated workflows directly into their servers. For product teams, this shift has created a paradox. The richest source of community-driven product feedback now lives on a platform that was never designed for structured data collection. Discord excels at real-time conversation. It was not built for product analytics, feedback categorization, or trend analysis. That mismatch is the central challenge. ## The Midjourney Playbook: When Product and Feedback Share the Same Room To understand why Discord community feedback is so valuable, consider what Midjourney actually built. Their Discord server grew to 21 million registered members, with an additional 7 million servers running the Midjourney bot. Discord was not just a support channel or a community add-on. It was the product itself. Users typed commands, received generated images, and iterated on prompts -- all within public channels where every interaction was observable by the team and the community alike. This architecture created three feedback advantages that traditional product development pipelines struggle to match. **First, usage and feedback were co-located.**In most software companies, product usage happens inside the application while feedback arrives through separate channels: surveys, support tickets, sales calls, NPS scores. The gap between the experience and the report introduces delay, filtering, and loss of context. On Midjourney's Discord, the experience and the reaction to it happened in the same message thread. A user who generated a disappointing image would immediately describe what went wrong, often with the generated image still visible in the conversation. Other users would pile on with similar experiences, alternative prompt strategies, or workarounds. The team could see the problem, the user's interpretation of the problem, and the community's collective response -- all without asking a single survey question. **Second, feedback was social and self-amplifying.**Discord's reaction system -- where users can add emoji responses to messages -- created an organic voting mechanism. When someone posted a feature request or reported a bug, the number of reactions served as a rough demand signal. This was not a formal upvote system designed by a product team. It was emergent behavior from the community itself, making it harder to game and more reflective of genuine sentiment. Midjourney's team could scan channels and immediately gauge which issues had broad resonance versus which were edge cases. **Third, the feedback loop was continuous, not periodic.**Traditional feedback collection happens in cycles: quarterly NPS surveys, post-release feedback forms, scheduled user interviews. Midjourney received feedback every second of every day. When they shipped an update, the community response was immediate and overwhelming in volume. They did not need to wait weeks to understand whether a change was well-received. They knew within hours, sometimes minutes. As one industry analysis noted: "While competitors struggle to gather training data and user feedback, Midjourney's massive user base generates both automatically, every single day." The result speaks for itself. Midjourney scaled from $200 million in 2023 to $300 million in 2024 to $500 million in 2025, entirely self-funded, with zero traditional marketing spend, and a team that grew from 11 to roughly 107-163 employees. The company's ability to iterate rapidly on product quality was inseparable from its ability to listen at scale through Discord. ## Why Community Feedback Carries Signals That Other Channels Miss The Midjourney case is exceptional in its scale, but the pattern it reveals applies broadly. Discord communities generate categories of product intelligence that other feedback channels either miss entirely or capture in diluted form. ### Bug Reports With Built-In Context When users report bugs in Discord, they rarely file terse one-line descriptions. Community members routinely include screenshots, screen recordings, reproduction steps, environment details, and before-and-after comparisons. Other users confirm or contradict the report, add their own observations, and help narrow down the conditions that trigger the issue. Ghost Ship Games, the studio behind Deep Rock Galactic, formalized this pattern by creating a `#jira-bug-reporter`channel with a webhook that pipes community bug reports directly into their issue tracker. The reports arriving through Discord consistently contained more actionable detail than those submitted through traditional bug report forms. ### Feature Requests With Organic Demand Validation A feature request submitted through a feedback form is a single data point from a single user. A feature request posted in a Discord channel with 47 reactions, 12 reply threads, and three users sharing mockups of how they envision the feature working is a fundamentally different kind of signal. The community has not only surfaced the request but has also validated demand, contributed design thinking, and in some cases identified implementation constraints -- all without being asked. ### Workflow Revelations and Use Case Discovery Some of the most valuable product signals emerge from conversations that are not explicitly about the product at all. When users discuss their workflows, share tips with each other, or describe how they have combined a product with other tools, they reveal use cases that the product team never anticipated. These organic workflow discussions are nearly impossible to surface through structured feedback mechanisms because users do not think of them as feedback. They are simply describing how they work. ### Onboarding Pain Points in Real Time Help channels and beginner-focused threads provide a running record of where new users get stuck. Unlike support tickets, which tend to capture only the problems severe enough to motivate a formal request, Discord help channels capture the full spectrum of confusion: the features that are not discoverable, the documentation gaps, the UI elements that mislead, and the mental models that do not transfer from competing products. Supabase, with its 46,000-member Discord community, and Reactiflux, the 220,000-member React and JavaScript ecosystem server, both see this pattern daily -- new users asking questions that collectively map the exact contours of the onboarding experience. ### Competitive Intelligence From Switching Conversations When users join a Discord community, they often arrive from a competing product. Their early messages frequently contain comparisons: what they liked about the previous tool, what drove them to switch, and what they expect from the new one. These switching narratives provide competitive intelligence that is both granular and authentic. Unlike competitive analysis reports assembled from marketing materials and feature lists, these are real users describing real decision-making processes. ### Pricing Sensitivity and Willingness-to-Pay Signals When a company announces a pricing change, the Discord server becomes an instant focus group. Users do not just say "too expensive" or "good value." They contextualize their reactions: what tier they are on, what features justify the price, what would make them upgrade, and what would make them leave. This real-time pricing feedback, complete with behavioral context, is extraordinarily difficult to capture through any other channel. ## The Scale Problem: Why Product Teams Cannot Keep Up If Discord communities are such rich sources of product intelligence, why are most product teams not systematically extracting it? The answer is straightforward: the volume, noise, and lack of structure make manual analysis impossible, and the problem compounds as communities grow. ### Volume That Defies Manual Processing Discord as a platform processes 1.1 billion messages per day across its ecosystem. Even a modestly successful community server generates thousands of messages daily. A server the size of Midjourney's or Reactiflux produces volumes that no human team can read comprehensively, let alone analyze systematically. Academic researchers studying Discord at scale -- such as the team behind the DISCO dataset, which compiled 1.5 million messages from 323,600 users across just four developer communities -- have documented the sheer density of conversational data these environments produce. For a product manager trying to understand what their community is telling them, the task is not finding feedback. It is finding the right feedback among tens of thousands of daily messages. ### The Duplication and Fragmentation Problem The same issue gets reported dozens of times, in different words, across different channels, by different users. A bug affecting image upload might surface as "upload broken" in one message, "can't attach files since the update" in another, and "anyone else having issues with drag and drop?" in a third. Without deduplication and clustering, a product team either counts the same issue three times or misses the pattern entirely because no single report reached critical mass. This fragmentation extends across time as well. A feature request that surfaces in January, gets discussed again in March, and reappears with renewed urgency in June looks like three separate requests rather than one persistent need. Historical feedback in Discord is notoriously difficult to retrieve as channels scroll and search capabilities hit practical limits. ### Missing Business Context Discord messages arrive without the metadata that product teams need for prioritization. There is no account tier attached to a message. There is no ARR value, no usage frequency, no customer health score. A passionate feature request from a free-tier user exploring the product for the first time looks identical to one from an enterprise customer whose contract renewal depends on the feature being built. Without this business context, product teams cannot perform the revenue-weighted analysis that turns raw feedback into defensible roadmap decisions. ### Signal-to-Noise Ratio Community Discord servers are social environments first. General conversation, memes, off-topic discussions, and interpersonal dynamics generate substantial message volume that has no product relevance. In active communities, the ratio of actionable product signals to social noise can be as low as one in fifty or one in a hundred. Manually filtering this noise is not just time-consuming -- it is cognitively exhausting in a way that leads to analysts missing genuine signals that happen to be buried between casual conversations. ### The Temporal Challenge Product feedback on Discord is temporally distributed in a way that defeats snapshot analysis. Important signals do not arrive in neat batches after a release. They trickle in over days and weeks, interspersed with unrelated conversation. A critical usability issue might first appear as a single confused question on Tuesday, get independently reported by two more users on Thursday, trigger a workaround discussion on Saturday, and finally explode into a visible thread the following Monday when a popular community member encounters it. The signal was present from Tuesday. But recognizing it required continuous monitoring that no human team can sustain across multiple channels, multiple time zones, and multiple concurrent conversational threads. ### The Organizational Gap Even companies that recognize the value of Discord feedback often lack a clear owner for extracting it. Community managers focus on engagement and moderation. Product managers focus on roadmap execution. Customer success focuses on retention metrics tied to paying accounts. Discord feedback falls into the gap between these functions. Research from companies that have formalized community feedback processes suggests that integrating customer suggestions into product development correlates with a 65% product launch success rate, and that companies responding actively to community feedback see 25-30% higher retention. But achieving these outcomes requires systematic extraction, not occasional monitoring. ## Closing the Extraction Gap The pattern across the technology industry is clear. Companies that build on Discord -- or maintain significant Discord communities alongside their products -- are sitting on an asset that most of them cannot fully exploit. The feedback is there. The signals are there. The community is doing the work of surfacing, discussing, validating, and contextualizing product insights at a scale that would cost millions to replicate through traditional research methods. What is missing is the infrastructure to extract those signals systematically, deduplicate them, enrich them with business context, and deliver them to product teams in a form that drives decisions. Tencent saw this during the Delta Force launch, when their Discord server was adding 8,000 to 10,000 new members per day at peak -- each one a potential source of feedback, but only if someone could process the firehose. Ghost Ship Games addressed a narrow slice of the problem with their Jira webhook channel, but that only captures bug reports that users choose to file formally. The vast majority of product intelligence -- the casual mentions, the workflow discussions, the workarounds, the competitive comparisons -- remains trapped in conversational threads that scroll past and disappear. The companies that solve this extraction problem will have a structural advantage. They will understand their users better, respond to issues faster, build features that align more closely with actual demand, and make roadmap decisions grounded in the full breadth of community intelligence rather than the narrow slice that happens to reach a product manager's inbox. This is the problem that ClosedLoop AI was built to solve. By connecting to the platforms where customers actually talk about products -- including Discord communities, sales conversations, support interactions, and more -- ClosedLoop AI extracts product signals at scale, deduplicates and clusters related feedback, enriches it with business context, and surfaces the insights that matter most to product teams. The goal is not to replace community engagement. It is to ensure that the intelligence communities generate every day actually reaches the people making product decisions, in a form they can act on. Discord communities are already doing the hard work of discussing, debating, and describing what they need from the products they use. The question for product teams is no longer whether that feedback is valuable. It is whether they have the infrastructure to listen. ![Jiri Kobelka](/assets/images/jiri-kobelka.png)Jiri Kobelka Founder We build tools that turn customer conversations into product decisions. ClosedLoop AI analyzes feedback from 40+ integrations to surface the insights that matter. ### Get insights like this in your inbox Product intelligence insights delivered weekly. No spam, just signal. Subscribe Join product leaders from companies using ClosedLoop AI ## Related Articles More insights you might find useful Integration Sep 22, 2025 ### [The In-Context Advantage: Why Intercom Conversations Capture What No Other Feedback Channel Can](https://closedloop.sh/blog/intercom-in-context-product-feedback) Intercom reaches 800 million monthly active end users with its in-app messenger. Unlike surveys or support tickets filed... 5 min read [Read Article](https://closedloop.sh/blog/intercom-in-context-product-feedback)Integration Oct 8, 2025 ### [2 Billion Messages a Day: Why Slack Is Your Richest (and Most Wasted) Source of Product Feedback](https://closedloop.sh/blog/slack-messages-product-feedback-waste) Slack processes 2 billion messages daily. Your customer-facing shared channels contain unfiltered feature requests, pain... 5 min read [Read Article](https://closedloop.sh/blog/slack-messages-product-feedback-waste)Integration Nov 16, 2025 ### [3.3 Trillion Minutes of Zoom Meetings Contain Product Intelligence Nobody Extracts](https://closedloop.sh/blog/zoom-meeting-recordings-product-intelligence-untapped) 500 million people use Zoom. 3.3 trillion meeting minutes happen annually. Unlike sales-focused tools, Zoom captures eve... 5 min read [Read Article](https://closedloop.sh/blog/zoom-meeting-recordings-product-intelligence-untapped)[Browse All Articles](https://closedloop.sh/blog) --- ## More Information - Website: https://closedloop.sh - Documentation: /docs - Pricing: https://closedloop.sh/pricing - Contact: https://closedloop.sh/contact