This week: The Missing Layer Between High Automation Rates and Quality.
Customer Experience News is a weekly newsletter about the most important news and discussions for Customer Experience and Customer Support Leaders.
This is all the weekly news you need in around 5 minutes.
Our main story today is about Intercom Monitors Monitors Brings AI Agent QA inside the platform.
Today’s 8 headlines are from:
Intercom Monitors Brings QA Inside the Platform
Intercom launched Monitors this week, completing the third pillar of their Analyze suite alongside Insights and Recommendations. The pitch is full observability across every customer conversation, not just aggregate reporting, but a structured, continuous QA system built directly into the platform where those conversations already happen.
π For support teams running Fin at scale and pushing toward high automation rates, that distinction matters. Separate QA tooling creates workflow friction and coverage gaps. Monitors is positioned as the answer to both.
The mechanics are straightforward. Monitors define which conversations get reviewed. You set filter criteria using people, company, or conversation attributes, combined with AND/OR logic, or use natural language flag criteria to surface higher-risk scenarios.
Scorecards define how each conversation gets evaluated β fully configurable criteria, scoring weights, and the option to mark specific criteria as critical, which drops the overall score to zero on failure regardless of other results.
π€The two work together: a Monitor without a Scorecard flags conversations but doesn’t score them. For a complete QA workflow, you need both. Auto-review can automate the entire process when AI scores all criteria confidently, routing only failures and edge cases to human reviewers. The practical effect is that your team stops reviewing conversations that don’t need attention and focuses entirely on the ones that do.
π€ There are meaningful caveats to understand before building out a configuration. Monitors currently evaluate Fin AI Agent conversations only. Applying Monitors and Scorecards to human teammate conversations is on the roadmap but not yet available as of March 2026.
This is a significant gap for teams running blended support operations where quality consistency across both Fin and Specialists matters equally. Monitors also do not have access to your knowledge base, Guidance, Tasks, or Procedures, so a Monitor cannot check whether Fin had access to a specific article or whether a conversation should have been escalated based on a guidance rule.
π Intercom has confirmed real-time alerts are coming, as is evaluation against your knowledge base content. The feature is also gated behind the Pro add-on.
The configuration investment is also real. Scorecards need to reflect how your operation actually works, and that requires deliberate thinking about what quality means for your specific team before you start building criteria.
The broader signal here is Intercom’s trajectory.
π Their own support team is currently at 83% automation with a stated target of 95%. Monitors is the infrastructure that makes high automation rates operationally defensible, a system to verify and maintain.
For any team running Fin at meaningful volume, the question is no longer whether AI is handling conversations, but whether it’s handling them the right way.
That’s what Monitors is designed to answer.
Help Scout AI Answers Now Reads Google Docs
Help Scout, a customer support platform used by SaaS and growing businesses, announced that its AI Agents now support Google Docs and Google Sheets as live knowledge sources for AI Answers. Teams connect a document as a single source, set its sharing to “anyone who has the link,” and the AI Agent pulls updated content automatically as the document changes.
Operational Impact
Support teams that maintain their internal policies, FAQs, or product documentation in Google Workspace can now feed that content directly into Help Scout’s AI without copying it into a separate knowledge base. When a policy changes in a shared Google Doc, the AI Answer reflects the update without manual republishing. This removes a lag point where AI responses drift out of sync with actual company policy, which is a real problem for teams that update policies frequently but lack dedicated knowledge base managers.
Implementation Considerations
The integration requires documents to be shared with “anyone who has the link,” which creates a consideration for teams with strict document access controls. Policy documents covering compensation, legal terms, or unreleased product details should not be connected without confirming that the sharing requirement is acceptable under your organization’s information governance rules. The integration also works only as well as the documents themselves. Poorly structured, outdated, or internally inconsistent Google Docs produce AI Answers with the same problems, just from a different source. Teams that have not audited their Google Workspace content before connecting it will carry that mess directly into customer-facing AI responses.
Pylon Connects AI Agent to GitHub, Linear, Notion
Pylon, a B2B customer support platform built around Slack and Teams-based workflows, announced that its AI agent now supports custom MCP (Model Context Protocol) connections to external tools including GitHub, Linear, Notion, and Composio. The connections give the AI agent live read and write access to data across a company’s tool stack when working inside Pylon issues.
Operational Impact
Support teams on Pylon can now instruct their AI agent to pull live data from connected tools while composing a response or handling an issue. An agent fielding a bug report can reference the current status of a linked GitHub issue without a human switching tabs. An agent handling a product question can search Notion documentation in real time rather than relying on a static knowledge base that was last updated weeks ago. The MCP server also now supports posting internal notes via the AI agent, which means agents can log context for human teammates without sending that information to the customer.
Implementation Considerations
MCP connections require deliberate access scoping. Giving an AI agent live access to GitHub or Notion means it can surface content from those systems, including content that was not intended for customer-facing responses. Teams need to define what data the agent can access and test edge cases before enabling the feature in production. The March 16 release also includes a fix for AI-generated issue titles that previously ignored the customer’s language and defaulted to English, so teams upgrading should review all changes together rather than enabling MCP connections in isolation.
Sierra Benchmarks AI Agents on Knowledge and Voice
Sierra, an enterprise AI agent platform, released ΟΒ³-Bench, expanding its open-source agent evaluation framework to two new domains: Ο-Knowledge, which tests whether agents can navigate large collections of company documents to complete multi-step tasks, and Ο-Voice, which evaluates agents under realistic phone call conditions including background noise, interruptions, and compressed audio. The release also incorporates community-contributed fixes to the original Ο-Bench airline, retail, and telecom tasks, including contributions from Anthropic and Amazon.
Operational Impact
The Ο-Knowledge results give support leaders a data point for evaluating AI agent claims against realistic document-navigation tasks. The benchmark uses Ο-Banking, a fintech support domain built from 698 documents across 21 product categories. Even when the exact relevant documents are provided to the model, the best-performing frontier model (GPT-5.2 with high reasoning) completed only 40% of tasks. Without pre-selected documents, that drops to 25%. For teams that have been told their AI agent handles complex, multi-step knowledge lookups reliably, those numbers are a useful calibration check before setting deflection targets. The Ο-Voice results are similarly grounding: voice agents under realistic call conditions achieved 26-38% task success, compared to 85% for text-based agents with reasoning capability.
Implementation Considerations
Sierra built ΟΒ³-Bench from its own deployment experience, which shapes what it measures. The Ο-Banking domain covers fintech customer support workflows; teams in healthcare, B2B SaaS, or retail will find partial but not complete overlap with their actual environments. The benchmark is open source and accepts community contributions, but extending it to your specific domain requires time and domain expertise that most support operations teams do not have available. The Ο-Voice results also point to a meaningful performance gap between lab demos and production phone calls, which is relevant for any team currently evaluating or piloting voice agents. What performs well in a controlled test environment typically degrades once real background noise, caller accents, and dropped audio frames enter the picture.
LiveAgent Adds Telegram as a Native Support Channel
LiveAgent, an all-in-one support platform serving budget-conscious teams, announced that Telegram Messenger is now a native channel starting in version 5.61. Incoming Telegram messages automatically generate tickets, appear in the shared dashboard alongside email, chat, and phone channels, and support conversation tagging, history search, and team routing.
Operational Impact
Teams supporting customers in markets where Telegram is the primary messaging channel, particularly Central and Eastern Europe, the Middle East, and parts of Southeast Asia, can now route those conversations into the same queue where they handle all other channels. Before this integration, a support team with a Telegram contact point was managing those messages in a separate app, outside of their ticketing workflow, without SLA enforcement, routing rules, or a shared conversation history. Centralizing Telegram into LiveAgent makes those conversations searchable, auditable, and subject to the same SLA and routing logic as every other channel.
Implementation Considerations
The integration requires LiveAgent version 5.61 or later. Teams on earlier versions need to upgrade before connecting Telegram. Telegram’s API structure differs from traditional messaging channels, with meaningful differences between bot-mediated and direct messaging setups, so teams should verify their current Telegram configuration before assuming the integration connects without additional steps. Teams that currently handle Telegram through a separate tool should also map their existing conversation tags and routing rules before going live, since historical Telegram conversations will not automatically import.
Five CX Platforms Launch MCP Connections
Model Context Protocol (MCP) support reached several CX platforms in the same week.
- Missive launched MCP support that lets AI tools connect directly to a team’s shared inbox and take actions such as creating Linear issues or searching Notion documents, alongside improvements to token efficiency and translation memory for multi-language teams.
- Enterpret released a Customer Insights Plugin for Claude that gives AI assistants access to Enterpret’s Customer Graph; the company reported 175 active users across 29 accounts and more than 14,000 Customer Graph queries in the early weeks after launch.
- Chattermill released an MCP server in early access that lets AI tools query NPS changes, sentiment trends, and customer verbatims directly, with compatibility across Claude, Cursor, ChatGPT, Codex, and Gemini CLI.
- Miro launched an MCP server that enables AI coding tools to interact with its collaborative whiteboarding environment.
- Zoho also announced MCP support across its product suite. The density of these launches in a single week signals that MCP is becoming a standard integration requirement, not an optional capability, for platforms that want AI agents to operate across a customer’s full tool stack.
HappyFox Launches All-in-One AI Contact Center
HappyFox, a customer support platform serving SaaS and mid-market companies, announced the launch of HappyFox Contact Center, an omnichannel platform that bundles help desk, native voice, chatbot, AI agents, and web messaging into a single product. The announcement positions it as a unified alternative to running separate tools across support channels.
Operational Impact
Support leaders running a helpdesk alongside a separate voice platform and a standalone chatbot maintain three tools, three contracts, and three integrations to keep working. A single system that handles all channels from one admin panel reduces the overhead of syncing data across tools and gives agents a single view of customer conversation history regardless of channel. For teams already on HappyFox, the contact center expansion adds native voice and expanded AI without requiring a platform migration. For teams on separate tools, the pitch is consolidation, though that consolidation comes with a migration and workflow change for agents, not just an admin swap.
Implementation Considerations
“AI-native” in a product announcement covers a wide range without committing to specifics. Teams evaluating this platform should ask what the AI agent resolves without human involvement, how the voice component handles complex or high-emotion calls, and whether the pricing structure changes as you add channels or volume. Teams migrating from established platforms like Zendesk or Intercom should budget for the workflow changes their agents will need to adjust to, not just the technical migration. The announcement does not include publicly available performance data for the AI components, so evaluation will require a proof-of-concept on your own ticket volume.
Attio App Store Adds 30 New Integrations
Attio, a CRM platform built for modern go-to-market teams, added more than 30 new apps to its App Store, including Amie, Snitcher, Flowla, Ringover, and Synter. The additions expand the range of sales, communication, and revenue tools that can sync data directly into Attio records.
Operational Impact
For support and success teams using Attio as their customer relationship layer, more native integrations mean less manual data entry and fewer custom webhook builds to keep contact records current. A team using Ringover for calls can now log call data to Attio without exporting CSVs or maintaining a separate sync. The broader the integration catalog, the more useful Attio becomes as a system of record for customer interactions across channels, which reduces the time agents spend switching tools to find context before a call or conversation.
Implementation Considerations
Releasing 30 integrations at once means the depth of each one varies. Some will support bidirectional sync; others may only push data in one direction or cover a limited set of fields. Teams should evaluate the specific integration they need rather than assuming a consistent level of functionality across the catalog. Attio is primarily positioned as a CRM for sales and growth teams. Support and customer success teams adopting it to track customer relationships should confirm how their support tool’s ticket and case data integrates before building workflows that depend on that connection.
Maven AGI Launches No-Code Agent Capabilities Builder
Maven AGI, an enterprise AI agent platform for customer support, announced the launch of Agent Capabilities within its Agent Designer tool. The feature lets CX and operations teams define new agent behaviors using two types of building blocks: Actions, which handle mid-conversation tasks triggered by natural language instructions, and Triggers, which fire automatic system responses based on defined conditions. Maven positions the feature as a way for support teams to build, test, and deploy new agent behaviors without filing an engineering ticket.
Operational Impact
The practical difference here is who owns the configuration. When a support team needs an AI agent to handle a new workflow, such as routing escalations to a specific team, auto-tagging certain ticket types, or sending confirmation messages after specific actions, that work typically requires an engineering ticket and a wait through a sprint cycle. Agent Capabilities shifts that configuration into the hands of the support operations team. A team that ships a new refund policy can update their agent’s behavior on the same day rather than scheduling it for the next release. That cycle-time reduction is meaningful for teams running frequent policy changes or seasonal workflow adjustments.
Implementation Considerations
Agent Capabilities that touch live systems, such as order management, account updates, or billing adjustments, need testing that takes longer than a quick configuration. Teams should treat this as a tool that accelerates the build-test-deploy cycle, not one that removes the testing step. The feature also requires clear thinking about scope: vague trigger definitions produce inconsistent agent behavior that is harder to diagnose than a well-defined rule. Teams new to configuring AI agent behaviors should start with low-risk, high-confidence use cases, such as tagging or routing, before moving to actions that update customer records or initiate transactions.
Now this is:
Strategic Support for CX Leaders.
You’ve got ambitious Support targets and new metrics but you’re not sure what to prioritize first.
The list is long, the queue is getting longer, and you don’t have time to step back and think about CX strategically.
What if you could pressure-test your thinking with someone who’s spent 20 years building Customer Support operations?
No pitch, just a conversation.
