Most AI customer service software companies build the agent first and promise intelligence later. Revelir AI did the opposite: we built the intelligence layer first, and then built the agent on top of it. That sequence is not an accident of roadmap timing. It reflects a fundamental belief about what makes a customer support AI agent worth deploying in the first place. An agent without a feedback loop is a black box. An agent grounded in continuous conversation analytics, quality scoring, and contact center AI insights is one that actually improves. This article explains the reasoning behind that architecture decision, and why it matters to any enterprise evaluating their support stack today.
- Revelir AI built its insights engine and QA scoring engine before its autonomous support agent, creating a feedback loop that makes the agent measurably better over time.
- Most AI platforms treat QA and analytics as add-ons. Revelir treats them as the core competitive layer, using 100% conversation coverage and sentiment arc tracking to surface risks that resolved tickets hide.
- The customer sentiment analysis tool inside Revelir Insights tracks how customers felt at the start and end of each conversation, not just whether the ticket was closed.
- RevelirQA scores every conversation against your own policies via RAG, with a full audit trail, making it viable for compliance-sensitive industries like fintech.
- Enterprise clients Xendit and Tiket.com are already processing thousands of tickets per week in production, not in pilot.
Why Do Most AI Agents Fail to Improve After Deployment?
The core failure mode of most deployed customer support AI agents is not that they are wrong on day one. It is that they have no structured mechanism to get better on day two. [1] Evaluating individual agents in production requires a layer of infrastructure that most companies build as an afterthought: consistent scoring, structured conversation tagging, sentiment tracking, and root-cause analysis at scale.
Without that layer, improvement is ad hoc. A team reviews a sample of tickets when a complaint escalates. They tune a prompt. The problem resurfaces elsewhere. Enterprises moving beyond the pilot stage with agentic AI are discovering that the real engineering challenge is not generating a response, it is knowing whether that response was correct, on-policy, and actually resolved the customer's problem. [2] [3]
This is precisely the gap Revelir AI was designed to close.
What Is the "Intelligence-First" Architecture, and Why Does It Matter?
An intelligence-first architecture means the conversation analytics software, quality scoring, and insights layers are built before, and separately from, the agent layer. The agent is then trained, monitored, and improved using data that the intelligence layer continuously generates.
Revelir AI is structured across three interdependent layers:
- Revelir Insights: An AI insights engine that enriches every ticket with structured metadata including contact reason, initial customer sentiment, ending customer sentiment, churn risk, tone shift, and unlimited custom metrics.
- RevelirQA: An AI scoring engine that evaluates 100% of conversations against your own SOPs and policies, ingested via RAG into a vector database, and produces a full reasoning trace for every score.
- Revelir Support Agent: An autonomous customer support AI agent that handles high-volume, repeatable requests, grounded in the quality and insight data generated by the two layers above.
The agent does not operate in isolation. It is evaluated by the same QA rubric that scores human agents, and its performance is tracked by the same insights engine that monitors the entire support operation. [5]
What Is a Sentiment Arc, and Why Is It More Valuable Than CSAT?
A sentiment arc is the tracked change in customer sentiment between the start and the end of a single conversation. It is one of the most differentiated features inside Revelir's customer sentiment analysis tool, and it surfaces a class of risk that traditional metrics completely miss.
CSAT and NPS are post-interaction snapshots. They tell you a customer submitted a rating. They do not tell you the customer started the conversation in a positive mood, encountered a slow or unhelpful resolution, and ended frustrated, even though the ticket was technically closed.
"Zendesk tells you a ticket was resolved. Revelir Insights tells you the customer started frustrated and ended neutral, a retention risk on a technically resolved ticket."
At scale, this becomes operationally significant. Consider what it means to know that 15% of tickets in a given week started with a positive sentiment and ended with a negative one, and to be able to drill into exactly what those conversations had in common. That is not a vanity metric. That is a churn signal.
| Metric | What It Tells You | What It Misses |
|---|---|---|
| CSAT Score | Post-interaction satisfaction rating | In-conversation emotional arc, non-responders |
| Ticket Resolution Rate | Whether the ticket was closed | Whether the customer was actually satisfied |
| Sentiment Arc (Revelir) | How customer felt at start vs. end of conversation | Nothing, it covers the full arc |
How Does RevelirQA Differ From Standard Automated QA?
Most automated QA platforms score conversations against generic rubrics, preset quality criteria that have nothing to do with your specific policies, product, or compliance requirements. RevelirQA takes a different approach: it ingests your knowledge base and SOPs directly into a vector database, then retrieves your actual policies before scoring each conversation.
The practical result is that every score reflects your business, not an industry average. And because the platform provides a full audit trail on every evaluation, including the model used, the prompt applied, and the specific documents retrieved, it is viable in compliance-sensitive environments. This is not theoretical. Xendit, an Indonesian fintech processing thousands of tickets per week, is running RevelirQA in production. [5]
A secondary benefit that is often underappreciated: RevelirQA evaluates AI agents and human agents under the same rubric. As enterprises deploy hybrid support operations, this creates a unified view of quality across the entire team, not two separate measurement systems running in parallel.
What Does "Ask Your Support Data Anything" Actually Mean?
The contact center AI insights layer inside Revelir connects to Claude via MCP (Model Context Protocol). This single connection gives Claude both the raw helpdesk data (from Zendesk, Salesforce, or any integrated platform) and the full AI enrichment layer that Revelir Insights generates.
In practice, a Head of CX can type a plain-English question: "What drove negative sentiment last week?" or "Which contact reason grew fastest in the last 30 days?" and receive a synthesised, evidence-backed answer drawn from real ticket data. This is a superset of a standard Zendesk MCP connection; no separate Zendesk integration is required.
This matters because it removes the barrier between data and decision. CX leaders are not data analysts. They should not need to build pivot tables to answer questions about their customers. [4]
Frequently Asked Questions
1. Is Revelir AI only for Southeast Asian businesses?
No. Revelir AI is a global enterprise platform. Southeast Asia is a proven environment, with production deployments at Xendit and Tiket.com, but the platform integrates with any helpdesk via API and supports multilingual environments. The SE Asia experience is a differentiator, not a limitation. [5]
2. What helpdesks does Revelir AI integrate with?
Revelir integrates with any helpdesk platform, including Zendesk and Salesforce, via API. The MCP connection to Claude works as a superset of a standard Zendesk MCP, meaning you do not need a separate Zendesk connection alongside Revelir.
3. How is RevelirQA different from a conversation analytics software with built-in scoring?
Most conversation analytics software scores against preset or generic criteria. RevelirQA ingests your specific SOPs and knowledge base via RAG, so every score reflects your actual policies. It also provides a complete reasoning trace, making it auditable in a way that generic scoring engines are not.
4. Does Revelir Insights replace my existing BI or reporting platform?
Revelir Insights is designed to complement existing infrastructure, not replace it. It provides an AI enrichment layer (sentiment arc, contact reason tagging, custom metrics) on top of your existing helpdesk data, and exposes that layer via MCP to conversational AI like Claude.
5. Can Revelir evaluate AI agents alongside human agents?
Yes. This is one of Revelir's most distinctive capabilities. RevelirQA applies the same scoring rubric to both human and AI conversations, giving CX leaders a unified quality view across a hybrid support operation.
6. What makes Revelir's sentiment tracking different from standard sentiment analysis?
Standard sentiment analysis gives you a single label per conversation (positive, negative, neutral). Revelir's customer sentiment analysis tool tracks the sentiment at the start and end of each conversation separately, surfacing the arc. A ticket that starts positive and ends negative is a retention signal that a single-score approach would classify as neutral and miss entirely.
7. Is Revelir AI suitable for businesses not yet using an AI agent?
Absolutely. RevelirQA and Revelir Insights are fully standalone products. Many customers deploy the QA scoring engine and insights engine first to build a quality baseline and understand their contact drivers, before or independently of deploying the Revelir Support Agent.
Revelir AI builds AI customer service software across three integrated layers: an autonomous support agent, a QA scoring engine, and an insights engine. Founded in 2025 by a YC W22 alumnus and headquartered in Singapore, the platform is built for global enterprise teams running high-volume, multilingual customer service operations. [5] Enterprise clients including Xendit and Tiket.com process thousands of tickets per week in production, using Revelir's 100% conversation coverage, sentiment arc tracking, and RAG-powered QA to drive measurable improvements in customer service quality and operational visibility. Revelir integrates with any helpdesk via API and connects to Claude via MCP, making it a flexible intelligence layer for any existing support stack.
See what your support data is actually telling you.
Revelir AI helps enterprise CX teams move beyond CSAT and ticket counts to understand the real drivers of customer experience, at scale, with full auditability.
Explore Revelir AIReferences
- How to build agents that actually work: A practical guide to evaluating AI (www.glean.com)
- Building an Insights AI Architecture | Real Story Group (www.realstorygroup.com)
- Learn by Insight... - Explore & Share (learnbyinsight.com)
- AI Insights: Exclusive Talks with Industry Pioneers - Rivel (www.rivel.com)
- Revelir AI Launches Automated QA Engine, Secures Xendit and Tiket.com as Enterprise Clients - The Tennessean (www.tennessean.com)