How Italy's Largest Marketplace Automated 99.5% of Moderation and Scaled Messaging Without Adding Headcount

Subito rebuilt buyer-seller messaging for more than 10 million users, and moved moderation, fraud prevention, and offer negotiation onto Stream, without standing up a chat infrastructure team to do it.

Messaging Is the Marketplace

Every second, buyers and sellers on Subito negotiate car prices, schedule apartment viewings, ask questions about electronics, and finalize purchases. For Italy's largest re-commerce marketplace, messaging isn't an add-on feature. It's the infrastructure behind trust. "Having instant communication is vital because it allows users to close deals in a very short time and helps them trust the person they are talking with," says Luciano Berardi, Engineering Manager at Subito. For a peer-to-peer marketplace, trust is the product. Every fraudulent message that reaches a buyer raises support costs and risks the relationship. Subito needed to scale that trust without scaling the team behind it. However, supporting real-time communication at scale had become increasingly difficult.

The Build vs. Buy Decision

Subito's legacy in-house messaging system was straining under the cost of supporting millions of users across web, iOS, and Android. Maintaining feature parity, scaling the backend, and evolving the product experience internally were consuming engineering time that competed directly with the core marketplace roadmap. "It was quite difficult to maintain a complex product like a messaging system," Berardi explained. "It was hard because of the infrastructure, the complexity of the product, and all the side issues this kind of tool can bring." Messaging had quietly become one of the most technically demanding systems inside the company. "The difference is huge," Berardi adds. "We understood that messaging is not just a feature, it's a complex system that needs maintenance, evolution, and support." The cost of the build path was never just the system itself. It was the dedicated engineering ownership across web, iOS, and Android SDKs, the reliability and security work, and the moderation tooling, all competing with the features that actually differentiate the marketplace. Partnering with Stream redirected that capacity back to product. The decision: stop maintaining chat infrastructure internally and partner with a provider purpose-built for large-scale real-time messaging.

Why Subito Chose Stream

Subito evaluated Sendbird and Twilio before selecting Stream Chat. The evaluation came down to three things: SDK depth across web, iOS, and Android; the ability to support marketplace-specific message types (offer negotiations, system messages, restricted visibility) without forcing Subito into a generic chat UX; and a partnership model that could accommodate the company's scale. The team needed:

• Flexible APIs and SDKs
• Reliable infrastructure at massive scale
• Enterprise-grade moderation support
• Multi-platform consistency
• Fast implementation timelines
• Room for future innovation "We were looking for a product that provided a set of APIs we could integrate with an affordable effort," Berardi recalls. "Stream provided separate Chat SDKs for different platforms and APIs that suited our needs." But the decision wasn't just about APIs. Subito needed a development partner capable of adapting to the unique complexity of marketplace communication. "From the beginning, Stream showed a strong understanding of both the technical challenges and the product requirements behind our messaging experience," Berardi notes.

Migrating Millions of Conversations Without Disrupting Users

Subito began the migration in early 2025. The stakes were enormous:

• More than 10 million registered users
• More than 2 million daily users
• Millions of active conversations
• Zero tolerance for downtime or degraded user experience Rather than rushing the rollout, Subito executed a carefully staged migration strategy:
• Proof of concept and technical validation
• Frontend SDK integration
• Conversation architecture design
• Dual-write migration with both systems running simultaneously
• Gradual rollout and infrastructure tuning
• Full migration to Stream Chat "We started filling the data on Stream but still had the legacy system," Berardi explained. "As a last macro step, we moved completely to Stream, leaving the legacy system." By the end of 2025, Stream had fully replaced Subito's legacy messaging infrastructure.

Building Marketplace Messaging, Not Generic Chat

Subito's deployment goes far beyond basic real-time messaging. Berardi explains, "Our platform mostly requires one-to-one conversations, but those conversations require many interactions and many actions around transactions." What that looks like in production:

• Structured offer negotiation flows
• Pending messages
• Restricted visibility system messages
• Delivery receipts
• Custom attachments
• Push notifications
• Fraud and spam moderation integrations

Offer Negotiations With Pending Messages

On marketplace platforms, trust and transaction velocity are closely linked. Buyers are far more likely to complete purchases when they can quickly ask questions, negotiate details, and receive immediate responses from sellers. Research on digital marketplaces consistently shows that real-time communication reduces purchase hesitation and increases transaction confidence. One of Subito's most sophisticated implementations is its use of Stream's pending messages functionality to power structured buyer-seller negotiation flows. Instead of relying on simple freeform messaging, Subito designed conversations around the actual mechanics of marketplace transactions. Messages can exist in temporary negotiation states before being committed as finalized offers, directly reflecting how buyers and sellers negotiate in real life.

Scale That Improved Stream's Infrastructure

Subito's implementation didn't just scale successfully. It actively improved Stream's platform. The marketplace relied heavily on restricted visibility messages: system messages visible only to specific participants in a conversation. At Subito's scale, this uncovered backend query inefficiencies that would never appear in smaller applications. Left unresolved, that kind of inefficiency is exactly what forces expensive infrastructure scaling down the line. Instead, Stream's engineering team rebuilt the query path directly. Before optimization:

• 226ms planning time
• 94ms execution time
• 365 gRPC calls After Stream engineering rebuilt the query path:
• 3ms planning time
• 30ms execution time
• 2 gRPC calls That represented:
• 75x faster query planning
• 3x faster execution
• A massive reduction in backend load The optimization shipped platform-wide, so Subito got faster message delivery at lower backend cost without spending a single engineering sprint on it. Other Stream customers using restricted visibility messaging benefited from the same work.

Automating Trust and Safety at Scale

Trust and safety are essential for peer-to-peer marketplaces. Subito's moderation stack combines machine learning, pattern-based rules, blocklists, and automated pipelines, with Stream wired in as the messaging layer on which those systems sit. Today, 99.5% of moderation actions are automated. At Subito's volume, that is the balance that makes safety affordable: automated pipelines handle the bulk of the traffic, and human reviewers are reserved for the edge cases that need judgment. Before Stream, Subito had no scalable path to either. "The architecture we put in place with Stream allowed us to integrate a third-party moderation system. We are now really satisfied with our ability to catch dangerous messages. The fact that we can move forward on identifying these cases without disrupting our flows is a great result," Berardi explains. Because moderation sits as a layer on top of Stream, Subito can evolve its trust and safety systems independently, without rebuilding the messaging infrastructure underneath them.

A Partnership Between Engineering Teams

As the migration accelerated, Subito and Stream established a dedicated Slack channel connecting engineering, infrastructure, and support teams directly. "Every need and every request we posted on Slack was immediately translated into a ticket for Stream," Berardi says. "The support from Stream was very responsive and very available to support us." That channel enabled rapid infrastructure tuning, backend optimizations, SDK troubleshooting, feature requests, hotfix deployments, and query-optimization work driven directly by Subito's use cases. "It's a system with a very good team that demonstrated adaptability to the customer and the needs of the customer," Berardi said. "Technically, it's quite good."

The Outcome

Subito now runs one of the largest marketplace messaging deployments in Europe. It supports tens of millions of messages every month, runs moderation and fraud prevention as an evolving layer rather than a rebuild, and ships new messaging-driven marketplace features (offer flows, delivery receipts, pending messages) without touching the infrastructure underneath. The strategic shift matters as much as the technical one. Subito stopped treating messaging as something to build and started treating it as critical infrastructure: something to partner on, not recreate. The engineering team now spends its time on what Subito actually sells: Italy's best re-commerce experience. "Having Stream as a partner means our teams can spend less time solving infrastructure problems and more time building features that improve transactions and user trust."