Running ai customer service software at enterprise scale is not a configuration challenge - it is an infrastructure and intelligence challenge. Revelir AI is currently processing tens of thousands of tickets per week for Xendit and Tiket.com, two of Southeast Asia's most high-volume digital businesses, with infrastructure designed to scale to millions [1][2]. What we learned from operating at that volume fundamentally changed how we think about quality, sentiment, and what "resolved" actually means for customer retention.
- Revelir AI processes tens of thousands of tickets per week for Xendit and Tiket.com in full production - not a pilot [1].
- At scale, "resolved" is not the same as "satisfied" - sentiment arc data reveals retention risks that standard ticket metrics hide.
- 100% conversation coverage eliminates the blind spots created by manual QA sampling.
- Multilingual ai customer support in Indonesian at high volume proved that language and cultural context are non-negotiable for accurate scoring.
- Contact center ai insights only become actionable when every ticket is enriched, not just a sample.
What Does Processing Thousands of Tickets a Week Actually Require?
Volume is not just a number - it is a stress test for every assumption you made during product design. When Xendit and Tiket.com came on board, we were not running controlled demos. We were ingesting live, high-velocity ticket queues in Indonesian and English, across multiple helpdesk configurations, with real compliance requirements and real customers on the other end [1][2].
Here is what operating at that scale actually demands:
- Infrastructure that does not degrade at volume. Batch scoring is not enough. Every ticket needs to be enriched in near real-time so that insights are current, not historical.
- Multilingual fidelity. Indonesian is not a simplified version of English. Sentiment, tone, and escalation cues behave differently across languages. Generic models trained primarily on English data produce unreliable scores in Bahasa Indonesia.
- Policy-grounded evaluation. At scale, generic QA rubrics produce noise. RevelirQA ingests each client's knowledge base and SOPs via RAG into a vector database, retrieving the actual policy documents before scoring each conversation. This is the difference between measuring compliance and measuring vibes.
- Full auditability. Xendit operates in a regulated fintech environment. Every AI evaluation carries a full reasoning trace - model used, prompt, documents retrieved - creating a compliance-grade audit trail on every score.
Why Does Sentiment Arc Matter More Than Resolution Rate?
Resolution rate tells you a ticket was closed. It does not tell you whether the customer would come back. The single most important insight from operating Revelir Insights at scale is what we call the sentiment arc: the delta between how a customer felt at the start of a conversation and how they felt at the end.
"A technically resolved ticket where the customer started frustrated and ended neutral is a retention risk. At scale, that pattern becomes a churn signal."
Standard customer sentiment analysis tools produce a single sentiment score per ticket. That is a snapshot. A snapshot cannot tell you whether your support operation is making things better or worse during the interaction itself. At Tiket.com, where travel disruptions drive emotionally charged contacts, the difference between a customer who ends a conversation feeling heard versus merely processed is the difference between a rebooking and a lost customer.
At scale, the sentiment arc becomes a strategic metric:
- Which contact reasons consistently produce negative sentiment drift?
- Which agents are turning frustrated customers around - and which are not?
- What percentage of tickets this week started positive and ended negative, and what do they have in common?
These are questions that traditional CSAT surveys cannot answer. They require ai customer feedback analysis applied to 100% of conversations, not post-call surveys completed by fewer than 10% of customers.
What Breaks When You Sample Instead of Covering 100% of Tickets?
Manual QA sampling - reviewing 3-5% of tickets per agent per week - is the industry default. At scale, it is a liability. Here is why sampling fails at volume:
| Dimension | Manual Sampling (3-5%) | Revelir 100% Coverage |
|---|---|---|
| Bias risk | High - reviewers unconsciously favour certain agents or ticket types | None - same rubric applied to every ticket |
| Edge case detection | Low - rare but serious issues are statistically invisible | Full - every outlier is scored and flagged |
| Agent coaching signal | Weak - based on a handful of tickets per period | Strong - patterns across entire conversation history |
| Compliance coverage | Partial - most tickets are never reviewed | Complete - auditable trace on every evaluation |
| AI agent evaluation | Not possible at volume | Evaluated under the same rubric as human agents |
The last row matters enormously as companies deploy customer support ai agent technology alongside human reps. If you are only scoring human agents, you have a blind spot over a growing portion of your customer interactions. Revelir evaluates both under a unified rubric - giving CX leaders a single view of quality across their entire operation.
What Makes Multilingual AI Support Different at Enterprise Scale?
Multilingual ai customer support is not a checkbox feature - it is an accuracy problem. Operating in Indonesia with Xendit and Tiket.com meant confronting a reality that many AI platforms gloss over: sentiment scoring and tone analysis trained primarily on English data perform inconsistently in Bahasa Indonesia [1].
Key challenges we solved at scale:
- Code-switching: Indonesian customer service conversations frequently blend Bahasa Indonesia and English in the same ticket. A model that treats these as separate languages misses context.
- Cultural tone calibration: Directness in Indonesian customer communication reads differently than in English. A politely worded complaint can carry high frustration that a generic sentiment model scores as neutral.
- Policy grounding in local language: RevelirQA retrieves SOP documents in the same language as the conversation before scoring. This keeps the evaluation grounded in the client's actual standards, not a translated approximation.
How Do Contact Center AI Insights Translate into Business Decisions?
Data enrichment is only valuable if it changes decisions. Contact center ai insights that live in a dashboard and require an analyst to interpret them are a step forward from raw ticket data - but they are not the ceiling. Revelir Insights connects to Claude via MCP, meaning a Head of CX can ask in plain English: "What drove negative sentiment last week?" or "Which contact reason is growing fastest?" and receive a synthesised, evidence-backed answer drawn from real ticket data.
This matters for three types of decision-makers:
- CX Operations leaders need real-time quality signals across every agent and every AI interaction - not a weekly sampling report.
- Product teams need to know which product issues are driving contact volume before they show up in NPS. Revelir Insights surfaces product feedback embedded in support tickets automatically.
- Compliance teams in fintech need a full audit trail. RevelirQA's reasoning trace - prompt, retrieved documents, scoring rationale - provides exactly that.
Frequently Asked Questions
Revelir AI is a Singapore-based AI customer service platform built for enterprise teams that need to move beyond CSAT surveys and manual ticket review. The platform operates across three layers: the Revelir Support Agent for autonomous ticket resolution, RevelirQA as an AI scoring engine that evaluates 100% of conversations against your own SOPs, and Revelir Insights as an AI insights engine that enriches every ticket with sentiment arc data, contact reason tags, and custom metrics. With production deployments at Xendit and Tiket.com processing tens of thousands of tickets per week, Revelir brings proven multilingual, high-volume capability to global enterprise CX teams - integrating with any helpdesk via API and connecting to Claude via MCP for natural language querying of your entire support dataset [1][2].
See How Revelir AI Handles Enterprise Scale
If you are managing thousands of tickets a week and still relying on sampling, surveys, or manual review, there is a better approach. Talk to the team that is already running at this scale.
Visit Revelir AI at www.revelir.ai