Introducing Relay: AI-Powered Email Support with Human-in-the-Loop Review

Today we are publicly launching Relay, an AI-powered email support platform built around a simple idea: AI should draft your support responses, but your team should decide what gets sent.

This is not a chatbot. It is not a generic email assistant. Relay is purpose-built for customer support teams that handle email as their primary channel. It connects to your inbox, reads your knowledge base, and drafts accurate, contextual responses to every incoming customer email. Your agents review those drafts, make any needed edits, and approve them for sending. The result is faster response times, more consistent quality, and a team that spends their time on judgment and empathy instead of typing out routine answers.

This post covers what Relay does, why we built it the way we did, and how it fits into your existing support workflow.

What Relay Does

At its core, Relay automates the most time-consuming part of email support: researching the answer and writing the response. Here is the workflow:

1. A customer sends an email to your support address
2. Relay analyzes the email to understand what the customer is asking
3. It retrieves relevant content from your knowledge base
4. It generates a draft response grounded in that content
5. The draft appears in your team's review queue
6. An agent reviews, optionally edits, and approves the response
7. The approved response is sent from your email address

This loop runs for every incoming email. For routine questions with clear answers in your knowledge base, agents often approve drafts with a single click. For more nuanced questions, agents edit the draft to add context, adjust tone, or provide additional detail. Either way, the agent is starting from a researched, written draft rather than a blank screen.

Why We Built It This Way

The Knowledge Base Problem

Most AI email tools treat the AI model as the source of knowledge. They generate responses from the model's general training data, which means the responses are generic at best and wrong at worst when it comes to your specific products, policies, and processes.

We built Relay around retrieval-augmented generation, where the AI always starts by finding relevant content in your knowledge base before generating a response. This means:

• Responses reflect your actual documentation, not generic AI knowledge
• As you update your knowledge base, responses update automatically
• You can see exactly which sources informed each draft
• The AI declines gracefully when no relevant knowledge base content exists, rather than making something up

The Human-in-the-Loop Imperative

We believe that AI support tools should augment human agents, not replace them. This is not a philosophical position; it is a practical one. AI models are remarkably capable but not perfectly reliable. They occasionally misinterpret questions, miss context, or apply information incorrectly.

The human review step catches these errors before they reach the customer. It also lets your team add the personal touches, contextual adjustments, and judgment calls that make the difference between adequate support and excellent support.

Over time, as you build confidence in the AI's accuracy for specific types of questions, you can enable auto-send for those categories. But you start with full human review, and the decision to expand automation is always yours.

Multi-Provider by Design

The AI model landscape is evolving rapidly. We did not want our customers locked into a single provider's capabilities and pricing. Relay supports multiple AI providers:

• OpenAI (GPT models)
• Anthropic (Claude models)
• Google (Gemini models)

You can switch between providers at any time without changing your knowledge base, configuration, or workflow. This gives you the flexibility to choose the model that works best for your use case and to adapt as new models become available.

Similarly, Relay supports both Gmail and Microsoft Outlook for email integration. Whether your team uses Google Workspace or Microsoft 365, the setup process is the same.

Key Features

Knowledge Base Management

Relay includes built-in knowledge base management. You can create, organize, and update your support content directly within the platform. Content is indexed and optimized for retrieval, so the AI can find the right information quickly and accurately.

You can organize content into collections, assign collections to specific mailboxes, and manage everything from a single interface. When you update an article, the change is reflected in AI-generated responses immediately. There is no manual re-indexing step and no delay between making a content change and seeing it in draft quality.

Intelligent Classification

Every incoming email is automatically classified by topic, intent, and urgency. This classification powers smart routing, priority scoring, and analytics. You can see at a glance what types of questions are coming in, how they are trending, and where your knowledge base has gaps.

Classification happens in the background as emails arrive. By the time an agent sees a conversation in their queue, it has already been categorized, prioritized, and assigned to the appropriate team member. This eliminates the manual triage step that consumes significant time in traditional support workflows.

Team Collaboration

Support is a team effort, and Relay is built for collaboration. Conversations can be assigned to specific agents. Internal notes let team members communicate about a case without the customer seeing. Activity history tracks every action taken on each conversation.

When a conversation requires input from multiple team members, the collaboration happens within the same interface where the agent reviews and sends responses. There is no switching between tools or copying context between systems.

Analytics Dashboard

Understanding how well the AI is performing and where improvements are needed is critical. Relay provides analytics on:

• Draft acceptance rates (how often agents approve drafts without changes)
• Response time metrics broken down by category, team, and individual
• Volume trends by category and time period
• Knowledge base coverage gaps identified from conversations where the AI could not generate a confident response
• Agent performance and workload distribution across the team

These analytics are not just informational. They drive a feedback loop where you can identify specific knowledge base improvements that will have the highest impact on AI draft quality.

Configurable AI Agents

Each mailbox has its own AI agent that you can configure with specific tone preferences, response guidelines, and knowledge base sources. This means your technical support inbox can behave differently from your billing inbox, even though both use the same underlying platform.

Agent configuration includes detailed instructions for how the AI should handle different types of situations: how to respond when the knowledge base does not cover a topic, whether to proactively offer related help, how to handle follow-up questions, and what level of formality to use. These instructions shape every draft the AI generates for that mailbox.

How It Fits Into Your Workflow

Relay is designed to be the center of your email support operation, not an add-on to existing tools. Here is how it fits:

For Small Teams (1-5 Agents)

Relay replaces the need for a separate shared inbox tool and a separate AI writing assistant. Everything is in one place: email management, AI draft generation, knowledge base, and team collaboration. Most small teams are up and running within a day and see meaningful response time improvements within the first week.

For Growing Teams (5-20 Agents)

As your team grows, Relay's classification, routing, and analytics become increasingly valuable. The AI handles the growing volume of routine questions while agents focus on complex issues. The draft-and-review workflow ensures quality remains consistent even as new agents join the team.

For Established Teams (20+ Agents)

For larger teams, Relay serves as the AI layer in your support stack. Its multi-provider flexibility, advanced classification, and detailed analytics support the operational sophistication that larger teams require.

Pricing

We designed our pricing to be straightforward and accessible:

• Starter ($49/month): Everything you need to get started with AI-powered email support
• Pro ($99/month): Advanced features for growing teams
• Ultra ($249/month): Full capabilities for teams that need maximum flexibility and throughput

All plans include knowledge base management, AI draft generation, human review workflow, and multi-provider support. There are no per-email fees or hidden AI usage charges.

What We Are Working On Next

Relay is launching with a strong foundation, but we have an ambitious roadmap. Here is a preview of what is coming:

• Enhanced auto-send controls with granular category-based and confidence-based rules
• Expanded knowledge base integrations connecting to popular help center platforms
• Advanced analytics with trend detection and improvement recommendations
• Workflow automations for common multi-step support processes

Getting Started

If your team handles email support and you want faster response times, more consistent quality, and a better experience for both your agents and your customers, we built Relay for you.

You can sign up for a free trial at userelay.ai and have your first AI-drafted responses within 30 minutes. Connect your inbox, add your knowledge base, and see the difference immediately.

We are excited to see what your team builds with Relay.