RevelirQA, the AutoQA engine from Revelir AI, reviews customer service conversations to the standard of a human quality analyst. Revelir calibrates it with each customer until, when that customer's own QA analysts validate a sample of its evaluations, more than 95 out of 100 are judged correct. Most auto QA tools already claim to score 100% of conversations, so coverage is table stakes. What matters for this question is that every ticket RevelirQA scores also gets a contact reason, sentiment and recurring issue type attached, and that is the data Claude needs. The short answer: connect Claude to consistently tagged ticket data through MCP. Ask for growth in absolute tickets, percentage change and share of volume, with a minimum-volume floor. Then ask Claude to explain the top risers with sentiment, sub-issues and quoted tickets before you act.
TL;DR
- RevelirQA performs automated quality assurance at the standard of a human quality analyst. It is calibrated per customer to over 95% accuracy, as validated by the customer's own QA analysts on a sample of evaluations, and it enriches every scored ticket with a contact reason Claude can trend.
- "Fastest-growing" has three readings: absolute, relative and share of volume. Ask for all three.
- Inconsistent helpdesk tags produce growth that reflects how tickets were labelled, not what customers did.
- Verify before acting: check volume floors, align time windows and read real tickets.
About the Author: Revelir AI builds RevelirQA, which runs in production at Xendit and Tiket.com on thousands of tickets a week. It enriches each ticket with contact reason, sentiment and recurring issue type, and connects that data to Claude via MCP.
What does "fastest-growing contact reason" actually mean?
Building on that short answer, the first thing to pin down is the metric, because "fastest-growing" has three defensible readings that rarely agree.
- Absolute growth: the reason that added the most tickets. This shows where workload is heading.
- Relative growth: the biggest percentage increase. It catches emerging problems early but exaggerates tiny categories.
- Share growth: the reason taking a larger slice of total volume. It strips out the effect of overall volume rising during a campaign or peak season.
A reason moving from 5 to 15 tickets has tripled. One moving from 2,000 to 2,300 grew 15% but added 30 times more work. Big buckets also hide problems: a broad category such as billing or payment issues can cover a large share of all contacts. A new failure mode inside a bucket that size barely moves the headline number, so you need sub-issue detail too.
Why does tagging quality decide whether Claude's answer is right?
A related but distinct question is where the contact reasons come from, because Claude can only trend the labels it is given.
Helpdesk categories are often set by agents picking from a dropdown or by macros. When a new macro ships, a new hire tags differently or a contact form changes, counts shift even though customers behaved the same way. Think of rain gauges emptied by different people on different schedules: the readings track the emptying routine, not the weather. Claude will report that artificial trend faithfully, which is exactly the risk.
The fix is to classify every ticket the same way, from the conversation itself. Reporting built on agent-applied category fields is only as consistent as those fields. Revelir's ticket enrichment adds contact reason, recurring issue type, sentiment and custom metrics to every ticket, which Revelir describes as signals the helpdesk does not produce on its own.
How do you set up a Claude MCP integration for ticket data?
Once the labels are consistent, the next step is giving Claude access to them. The Model Context Protocol (MCP) is an open standard from Anthropic that lets Claude query external data sources as tools during a conversation.
- Pick the data layer. Point Claude at enriched ticket data rather than a raw export.
- Connect through MCP. Revelir connects to Claude via MCP, so a Head of CX can ask "Which contact reason is growing fastest?" and get a synthesised answer backed by real ticket data instead of building a dashboard filter.
- Confirm visible fields. At minimum you need date, contact reason, recurring issue type, sentiment, channel and whether a bot or human handled the ticket.
- Test with a known answer. Ask about a spike you already investigated. If Claude misses it, fix the data before trusting new findings.
Which prompts get a reliable answer from Claude?
With access in place, answer quality depends on how precisely you frame the question. Vague prompts invite Claude to pick a definition for you.
| Prompt | What it guards against |
|---|---|
| "Compare contact reason volume for the last 14 days against the previous 14. Show absolute change, percentage change and share of total. Exclude reasons below 30 tickets in either window." | Tiny categories posing as emergencies; volume swings mistaken for shifts |
| "For the top three risers, which recurring issue types drive the growth? Quote three representative tickets each." | Acting on a bucket label without knowing the actual problem |
| "How does sentiment at the start versus the end of these conversations compare with last period?" | Treating resolved tickets as harmless when customers still leave unhappy |
| "Is the growth concentrated in chatbot-handled or human-handled conversations?" | Missing a bot that creates repeat contacts |
| "Break the growth down by channel and language." | A local issue diluted in global totals |
The 30-ticket floor is a starting point. Set it to roughly what one agent handles in a day, so anything below it is noise for staffing.
How do you verify Claude's answer before acting on it?
Stepping back from prompt wording, a separate concern is whether the number holds up. Run these checks:
- Align windows by weekday. Two periods with different weekend counts can create fake growth.
- Check for known events. Promotions, releases and pricing changes explain many spikes.
- Ask Claude to show its counts. Request the raw numbers per period so you can recompute the percentages.
- Read ten tickets. Ten minutes of reading confirms whether the label matches the conversation.
For QA scores, RevelirQA records the model used, the prompt, the documents retrieved and the reasoning behind each score, so you can trace why a grade landed where it did.
What should you do with the fastest-growing contact reasons?
The point of this exercise is action, and contact reason growth is the most direct voice of customer analytics you have: customers chose to reach out. Growth usually signals one of three causes:
- Product defect: send quoted tickets to product with volume attached.
- Policy or process gap: update the SOP. RevelirQA retrieves SOPs by contact reason before scoring, so updates flow into the next relevant evaluation.
- Handling problem: add custom metrics to the QA scorecard for that contact reason. Revelir supports binary, multi-option or scored criteria, configurable per contact reason.
Pairing growth with QA metrics turns raw counts into customer service insights you can assign to an owner.
Frequently Asked Questions
Can Claude find growing contact reasons from raw helpdesk data?
Yes, if tags exist and are applied consistently. Classifying thousands of raw tickets inside a chat session is hard to reproduce identically week to week.
How is conversation intelligence different from AutoQA?
Conversation intelligence and conversation analytics software surface patterns across conversations. AutoQA scores each conversation against your policies. Ticket enrichment lets one data layer support both.
Does customer sentiment analysis matter for growth questions?
Yes. A growing reason with worsening end-of-conversation sentiment deserves attention before a larger one where customers leave satisfied.
Does this work for chatbot conversations?
RevelirQA scores AI agents and human agents with the same rubric, provided the chatbot platform produces a conversation log via API.
Is the over-95% accuracy figure out of the box?
No. Revelir reaches it with each customer through calibration, measured by the customer's own QA analysts validating a sample of evaluations.
About Revelir AI
Revelir AI, headquartered in Singapore, builds RevelirQA, an AI customer service QA software engine that scores conversations against each customer's own SOPs and QA scorecard. It is a fully managed cloud service, set up and run by the Revelir team, and integrates with any helpdesk via API, including Zendesk and Salesforce. Every score carries a full reasoning trace, and every ticket is enriched with contact reason, sentiment and recurring issue type that Claude can query via MCP. Xendit and Tiket.com run it in production on high-volume English and Indonesian conversations.
Want to ask Claude what is really driving your queue, using data you can trust? Learn more at https://www.revelir.ai/.
