Most businesses in Malaysia know their customer service operation has gaps. They can feel the escalations. They see the complaints. And they notice when response times slip during peak periods.
What most businesses do not have is a structured way to find those gaps systematically. Before they compound into visible satisfaction problems. That is what a customer service audit checklist provides.
This guide covers eleven audit areas. Each has a specific checklist, benchmark targets, scoring guidance, and the Qiscus product feature that addresses what the audit reveals. Run this audit quarterly. Review the scores. And use the gaps it identifies to prioritise your next 90 days of improvement.
What Is a Customer Service Audit?
A customer service audit is not a performance review of individual agents. It is broader than that. It is a systems review of the processes and workflows that determine whether agents deliver good service consistently at scale.
Based on existing research, an audit that does not result in any action is a waste of time. The goal is to identify specific improvement areas, prioritise them by impact, and empower the team to act on them.
For businesses in Malaysia, a quarterly audit addresses the three operational realities that make ad-hoc improvement fall short. First, WhatsApp volume changes faster than manual monitoring can track. Second, agent turnover in Malaysia is high, and knowledge gaps accumulate quickly. Third, tool configurations that were correct at deployment drift over time as products and policies change. A structured audit catches all three before they become visible in CSAT data.
How to Use This Audit Checklist
A checklist without a process is just a list. This section tells you exactly how to run the audit so that what you find leads to action. Each audit area below contains a checklist, a scoring table, a benchmark target, and a Qiscus product note. Here is how to run the audit effectively.
- Step 1: Pull your data before starting. You need the last 30 days of performance data for each metric area. Do not run it from memory. Pull it from your helpdesk, omnichannel dashboard, and CRM.
- Step 2: Score each checklist item. Each item scores 0, 1, or 2. A score of 0 means the item is not in place. A score of 1 means it is partially in place or inconsistently applied. A score of 2 means it is fully in place and performing within the benchmark.
- Step 3: Calculate area scores. Calculate your percentage score per area.
- Step 4: Identify your two lowest-scoring areas. These are your priority targets for the next 30 days. Do not attempt to fix all eleven areas at once.
- Step 5: Assign an owner to each gap. Every gap needs a named owner and a 30-day target. Without ownership, audit findings become a report that no one acts on.
- Step 6: Re-audit in 90 days. The audit is only valuable if it generates improvement between cycles. Schedule the next one before you finish this one.
Six steps. Roughly two hours to complete if your data is already pulled. The audit is not the hard part. Acting on what it reveals is. The sections below give you exactly what to look for in each area.
11 Audit Area Your Business Must Aware Of
These eleven areas cover every operational dimension that determines whether your customer service delivers consistently or falls short. They are not generic best practices. They are the specific areas where businesses in Malaysia most commonly have gaps. And where those gaps are most likely to be invisible until they show up as complaints, churn, or escalations that were avoidable.
Work through each one in order, score honestly, and let the data tell you where to focus.
1. Response Time
Response time is the most visible customer service metric. And the one most directly tied to first impression. For businesses in Malaysia, response time on WhatsApp is measured in minutes, not hours.
Benchmark targets:
- WhatsApp first response: under 5 minutes during business hours
- Email first response: under 2 hours
- Instagram DM: under 30 minutes
- Live chat: under 1 minute
Checklist:
| Item | Score (0-2) |
| First response time tracked by channel, not just overall | |
| SLA targets defined and configured per channel | |
| Automated alerts fire before SLA breach | |
| After-hours auto-reply active on WhatsApp and primary channels | |
| Response time data reviewed weekly | |
| AI or auto-reply covers tier-one queries during peak periods |
Maximum score: 12
What a low score signals: SLA rules are generic or missing. Peak-period coverage is manual. Or response time data is only reviewed in aggregate. The most common cause is WhatsApp volume spikes during campaigns that push response times above benchmark without automated detection.
Qiscus feature: Qiscus Omnichannel Chat configures channel-specific SLA rules with automated alerts. And Qiscus AgentLabs handles after-hours and peak-period auto-response across every connected channel simultaneously.
Response time sets the first impression. The next area, CSAT, measures whether the full interaction lived up to it.
2. Customer Satisfaction Score
CSAT measures how customers felt about the interaction after it concluded. But collection timing determines whether the data is useful. Most businesses in Malaysia collect CSAT too late and too infrequently. The result is aggregate data, not actionable data.
Benchmark targets:
- Overall CSAT: above 85%
- CSAT collection rate: above 30% of closed tickets
- CSAT response window: within 10 minutes of ticket closure
Checklist:
| Item | Score (0-2) |
| CSAT collected automatically at ticket closure | |
| CSAT delivered within the same channel the customer used | |
| CSAT data segmented by channel, agent, and query category | |
| Negative CSAT automatically triggers alerts to team leads | |
| CSAT reviewed weekly | |
| CSAT trends monitored over time, not treated as one-time scores | |
| Agents receive individual CSAT visibility |
Maximum score: 14
What a low score signals: CSAT is collected via email after the interaction. Response rates are low and data arrives too late. Or CSAT is only reviewed in aggregate, hiding channel-specific and agent-specific gaps. Based on existing research, customer service KPIs tracked at the right granularity are what separate teams that continuously improve from those that plateau.
Qiscus feature: Qiscus Helpdesk Suite triggers in-channel CSAT collection at ticket closure. Results populate the real-time dashboard broken down by channel, agent, and query category. And negative results below a configured threshold automatically alert the responsible team lead.
A strong CSAT score is meaningful only if first contact resolution is also healthy. A high CSAT on three-contact resolutions is a different problem from a high CSAT on first-contact resolutions.
3. First Contact Resolution Rate
First contact resolution (FCR) measures the percentage of customer issues resolved in a single interaction without follow-up or escalation. It is the most direct measure of whether your operation genuinely solves problems or just manages volume.
Benchmark targets:
- FCR rate: above 70% overall
- FCR by channel: track separately, overall FCR masks channel-specific gaps
- FCR trend: improving quarter over quarter
Checklist:
| Item | Score (0-2) |
| CSAT collected automatically at ticket closure | |
| CSAT delivered within the same channel the customer used | |
| CSAT data segmented by channel, agent, and query category | |
| Negative CSAT automatically triggers alerts to team leads | |
| CSAT reviewed weekly | |
| CSAT trends monitored over time, not treated as one-time scores | |
| Agents receive individual CSAT visibility, not only team-level averages |
Maximum score: 14
What a low score signals: Knowledge base gaps are the most common FCR driver. Agents resolve the same queries differently. No standardised resolution path exists. Or AI is not covering the tier-one queries that represent 60 to 70% of volume. Based on existing research, first contact resolution is one of the highest-leverage metrics in customer service because it directly reflects whether agents have the information they need to resolve correctly.
Qiscus feature: Qiscus AgentLabs handles tier-one queries autonomously, improving FCR on automatable query types from the first week of deployment. And the AI copilot surfaces knowledge base content to agents on complex queries. It reduces FCR failure from information gaps.
FCR and escalation rate are the two metrics that most directly reveal your frontline team’s capability floor. FCR shows how often they resolve. Escalation rate shows how often they cannot.
4. Escalation Rate
Escalation rate is the percentage of tickets your frontline team cannot resolve and must transfer to a higher-tier agent or specialist. A high rate is almost always a routing problem, a knowledge base problem, or a training problem. All three are fixable.
Benchmark targets:
- Overall escalation rate: below 10%, target below 5%
- AI-to-human escalation rate: 20 to 35% (healthy range for AI-assisted operations)
- Escalation rate by category: identify top three categories each week
Checklist:
| Item | Score (0-2) |
| Escalation rate tracked by channel and query category | |
| Escalation triggers defined explicitly and documented | |
| Escalation triggers tested and confirmed to transfer full context | |
| Discretionary escalation tracked separately from necessary escalation | |
| Top three high-escalation categories reviewed and addressed weekly | |
| Knowledge base updated based on escalation patterns | |
| Routing rules reviewed whenever escalation rate rises above benchmark |
Maximum score: 14
What a low score signals: Escalation is informal, inconsistent, or undocumented. And escalation data is reviewed monthly in aggregate rather than by category weekly. Based on existing research, AI in customer service reduces escalation rates on human-handled tickets by improving the quality of information agents start each conversation with.
Qiscus feature: Qiscus Helpdesk Suite configures escalation triggers that fire automatically based on ticket age, SLA proximity, sentiment signals, or query category. And escalation reporting by channel and category is available in real time without manual compilation.
Escalation rate reveals where your frontline capability ends. Tool adoption reveals whether the tools that should extend that capability are actually being used.
5. Tool Adoption
A platform agents do not use does not improve service quality. Tool adoption audits the gap between what your customer service stack can do and what your team actually uses it to do.
Benchmark targets:
- Unified inbox adoption: 100% of agents handling all channels from one workspace
- AI suggestion acceptance rate: above 50% within 60 days of deployment
- Knowledge base usage: agents search knowledge base before composing responses
- Helpdesk ticket closure rate: above 95% (tickets not left open without resolution)
Checklist:
| Item | Score (0-2) |
| All agents handling all active channels from one unified inbox | |
| No agents using personal WhatsApp or email for customer interactions | |
| AI response suggestions used by at least 70% of eligible agents | |
| Knowledge base accessed by agents on complex queries | |
| Helpdesk SLA alerts acknowledged and acted on within the alert window | |
| New agents trained on platform use within first week of onboarding | |
| Platform usage tracked per agent |
Maximum score: 14
What a low score signals: Agents are using personal channels alongside the official platform, creating untracked interactions. Or AI features are deployed but agents are not using them. Trust or training gaps are the usual cause. Based on existing research, AI customer support that continuously trains on real customer conversation data consistently outperforms documentation-only systems because real usage drives improvement.
Qiscus feature: Qiscus Omnichannel Chat consolidates every channel into one agent workspace. Platform usage data, including AI suggestion acceptance rate, is available per agent in the Qiscus dashboard.
Tool adoption determines whether your platform investment delivers its potential. Agent skills determine whether your team can use that potential effectively.
6. Agent Skills
Agent skills audits the gap between what your frontline team can handle and what your query mix requires. And escalation data is the most reliable source of evidence for this audit area.
Benchmark targets:
- Agents trained on top 10 highest-volume query categories
- New agents cleared for independent handling within 30 days of onboarding
- Agent performance variance: high and low performers within 15% of team average FCR
Checklist:
| Item | Score (0-2) |
| Escalation data used to identify specific agent skill gaps | |
| Training scenarios built from real escalation patterns | |
| High-escalation query categories covered in onboarding curriculum | |
| Agent performance data reviewed individually | |
| Low-performing agents on specific query categories given targeted coaching | |
| Multilingual capability audited | |
| Agent training updated whenever escalation patterns shift |
Maximum score: 14
What a low score signals: Training is generic product knowledge rather than targeted at the query categories that generate the most escalations. Agent performance is only reviewed in aggregate. No individual-level coaching. And there is no feedback loop between escalation data and training content. Based on existing research, customer service standards that define clear capability requirements for each query category are what protect service quality during high-volume periods and team changes.
Qiscus feature: Qiscus AgentLabs acts as an AI copilot during live conversations, surfacing relevant knowledge base content and draft responses in real time. This narrows the performance gap between new and experienced agents faster than any training programme alone.
Agent skills determine the quality floor for human-handled interactions. Channel coverage determines whether every customer can reach the team at all.
7. Channel Coverage
Channel coverage audits whether every channel your customers use is actively monitored, staffed within SLA, and generating trackable data. For businesses in Malaysia, WhatsApp is always the highest-priority channel. The primary risk is WhatsApp-centric operations that added Instagram, email, or Telegram without the routing configuration to manage them properly.
Benchmark targets:
- Every active customer channel staffed within SLA during business hours
- After-hours AI coverage active on every channel that generates after-hours volume
- No active channel running without SLA configuration
- No active channel unmonitored for more than 60 minutes during business hours
Checklist:
| Item | Score (0-2) |
| Inventory of every channel currently active (WhatsApp, IG DM, email, etc.) | |
| SLA configured for every active channel | |
| After-hours coverage active on every high-volume channel | |
| Every channel visible in a single unified inbox with real-time agent assignment | |
| No customer-facing channel operated outside the main platform | |
| Channel volume data reviewed monthly to identify emerging channels | |
| Response rate above 95% on every active channel |
Maximum score: 14
What a low score signal: Channels have been added without SLA configuration or routing rules. Some channels are staffed manually without automated coverage during gaps. Or an agent is managing a personal WhatsApp number outside the main platform. Untracked and unmonitored interactions.
Qiscus feature: Qiscus Omnichannel Chat connects WhatsApp, Instagram DM, Facebook Messenger, Telegram, TikTok, email, live chat, and 20+ other channels into one agent workspace. Every channel gets SLA configuration, routing rules, and real-time visibility from day one of connection.
Channel coverage determines reach. Reporting quality determines whether the data from all that activity is actually used to improve.
8. Reporting Quality
Reporting quality audits whether your operation produces the data it needs to make improvement decisions. And whether that data is reviewed on a cycle that produces action.
Benchmark targets:
- Weekly performance review covering all eleven audit areas
- Data broken down by channel, agent, and query category (not just overall)
- Named owner for each gap identified in weekly review
- Improvement targets set in each review cycle with 30-day deadlines
Checklist:
| Item | Score (0-2) |
| Weekly performance review scheduled and consistently run | |
| Data reviewed by channel, agent, and query category, not just in aggregate | |
| Named owner assigned to each identified gap | |
| Improvement targets have deadlines | |
| Reporting data comes from one connected system | |
| Historical trend data available to compare current performance vs previous periods | |
| Reporting reviewed with team leads who have authority to act on it |
Maximum score: 14
What a low score signals: Performance data is reviewed monthly in aggregate. Improvement actions are vague and open-ended. And the people who review the data cannot change configurations, update routing rules, or allocate training resources. Based on existing research, scaling customer support requires data infrastructure that surfaces specific gaps at the right granularity, not just summary metrics that confirm things are generally fine.
Qiscus feature: Qiscus Helpdesk Suite and Qiscus Omnichannel Chat deliver a unified real-time reporting dashboard covering all eleven audit areas in one view. Response time, CSAT, FCR, escalation rate, tool usage, channel volume, and SLA compliance are all available by channel, agent, and query category without any manual data compilation.
9. Knowledge Base Quality
The knowledge base is the single resource that determines whether agents resolve independently or escalate. Whether AI answers accurately or hallucinates. And whether new agents reach full capability in two weeks or two months. Most businesses treat the knowledge base as a one-time setup task. It is not. It is a living operational asset that requires the same maintenance cycle as the products and policies it documents.
Benchmark targets:
- Knowledge base covers 100% of top 20 highest-volume query categories
- Knowledge base reviewed and updated within 5 business days of any product or policy change
- Each knowledge base article reviewed for accuracy at least once per quarter
- AI resolution accuracy above 70% on knowledge base-covered query types
Checklist:
| Item | Score (0-2) |
| Knowledge base covers your top 20 highest-volume query categories | |
| Each article reviewed for accuracy at least once per quarter | |
| Knowledge base updated within 5 business days of product or policy change | |
| Owner assigned to each knowledge base category | |
| Informal knowledge (agent memory, personal notes) documented and moved into KB | |
| KB structured for AI use — discrete answerable units | |
| Gaps identified from escalation data and addressed within 2 weeks |
Maximum score: 14
What a low score signals: The knowledge base was built at deployment and has not been updated since. Agents resolve from memory and personal notes rather than a shared source. And AI accuracy degrades over time because the knowledge it draws from is increasingly out of date. The most common symptom is a widening FCR gap between simple queries and policy-adjacent queries. Knowledge currency matters most there.
Qiscus feature: Qiscus AgentLabs trains on your knowledge base and immediately reflects updates. When the knowledge base is updated, AI accuracy on the affected query category improves from the next interaction. And the AI copilot surfaces knowledge base content to agents in real time. Agents rely on the KB, not personal memory.
A complete, current knowledge base is the foundation every other audit area depends on. If AI is below benchmark, escalation rates are high on specific categories, or FCR is inconsistent, the knowledge base is almost always where the root cause sits.
10. Proactive Communication
Most customer service audits measure reactive performance: response speed, resolution quality, and post-interaction satisfaction. Proactive communication audits whether your team is reaching customers before they need to contact you at all. For businesses in Malaysia with high WhatsApp volume, proactive communication reduces inbound volume on the most predictable query types.
Benchmark targets:
- Order and delivery status updates sent before customer inquiry rate exceeds 5% on that order cohort
- Post-resolution follow-up sent within 48 hours of ticket closure
- At-risk customer outreach triggered by defined data signals, not manual monitoring
- Proactive message open rate above 60% on WhatsApp
Checklist:
| Item | Score (0-2) |
| Automated status updates active for orders, appointments, or key customer milestones | |
| Post-resolution follow-up configured and sending within 48 hours of ticket closure | |
| At-risk customer triggers defined based on behavioural or account data | |
| Proactive message performance tracked (open rate, reply rate, inbound reduction) | |
| Broadcast campaigns segmented by customer tier or behaviour, not sent to entire list | |
| Proactive communication reviewed monthly for impact on inbound volume | |
| WhatsApp template messages approved and active for each proactive use case |
Maximum score: 14
What a low score signals: The team only communicates when customers initiate. No automated status updates exist for high-inquiry events like deliveries, appointments, or payment confirmations. And post-resolution follow-up is either manual or absent. It is one of the simplest ways to catch unresolved issues before they become complaints. Based on existing research, proactive customer service reduces inbound contact volume by addressing customer needs before they generate a ticket.
Qiscus feature: Qiscus Omnichannel Chat supports broadcast messaging and automated template delivery across WhatsApp and other channels. Status update automation, follow-up triggers, and segmented campaigns all configure within the same platform as the rest of the operation. No separate tool required.
Proactive communication reduces the volume your team has to manage reactively. AI performance determines how effectively that remaining reactive volume is handled autonomously.
11. AI Performance
If your team has deployed an AI chatbot or AI agent, auditing its performance is not optional. An AI that was accurate at launch degrades over time if it is not trained continuously. An AI that escalates too frequently is not reducing agent load. And an AI that escalates too infrequently may be attempting to resolve queries it should not handle autonomously. This audit area only applies to businesses that have deployed AI in their customer service operation.
Benchmark targets:
- AI tier-one resolution rate: above 70% on knowledge base-covered query types
- AI-to-human escalation rate: 20 to 35% (healthy range)
- AI training cycle: weekly or biweekly update from real conversation data
- Agent AI copilot suggestion acceptance rate: above 50% within 60 days of deployment
Checklist:
| Item | Score (0-2) |
| AI resolution rate tracked by query category | |
| AI-to-human escalation rate within the 20–35% healthy range | |
| AI training updated at least biweekly from real conversation data | |
| Conversations where AI confidence was low reviewed and used for training | |
| AI copilot suggestion acceptance rate tracked per agent | |
| Escalation handover quality confirmed — full context transfers to receiving agent | |
| AI performance reviewed in the same weekly cycle as human agent performance |
Maximum score: 14
What a low score signals: The AI was configured at deployment and has not been retrained since. Resolution rate has drifted below benchmark as products and policies changed. Or escalation rate is above 40%, indicating significant knowledge base gaps or misconfigured intent recognition. Based on existing research, AI customer service tools that continuously train on real conversation data consistently outperform documentation-only systems. Real usage reveals the query variations and intent patterns that static documentation cannot anticipate.
Qiscus feature: Qiscus AgentLabs identifies conversations where AI confidence was low, escalation was triggered, or the initial response was inadequate. These conversations feed the next training cycle automatically. Resolution rate, escalation rate, and suggestion acceptance rate are all visible in the Qiscus dashboard by query category and channel.
Full Audit Scoring Summary
Record your scores across all eleven audit areas after completing each checklist.
| Audit Area | Max Score | Your Score | % Score | Target |
| Response Time | 12 | 80%+ | ||
| Customer Satisfaction Score | 14 | 80%+ | ||
| First Contact Resolution Rate | 14 | 80%+ | ||
| Escalation Rate | 14 | 80%+ | ||
| Tool Adoption | 14 | 80%+ | ||
| Agent Skills | 14 | 80%+ | ||
| Channel Coverage | 14 | 80%+ | ||
| Reporting Quality | 14 | 80%+ | ||
| Knowledge Base Quality | 14 | 80%+ | ||
| Proactive Communication | 14 | 80%+ | ||
| AI Performance | 14 | 80%+ | ||
| Total | 152 | 80%+ | ||
Interpreting your total score:
| Total Score | Interpretation |
| 122 to 152 (80%+) | Strong operation — identify and address the remaining gaps |
| 91 to 121 (60–79%) | Functional but fragile — structural gaps likely present in 3 to 4 areas |
| 61 to 90 (40–59%) | High improvement potential — prioritise the two lowest-scoring areas immediately |
| Below 61 (below 40%) | Foundational gaps — begin with tool consolidation and SLA configuration |
After scoring: Identify your two lowest-scoring audit areas. Those are your immediate priorities. Fix them before expanding to additional areas.
One score that tells you exactly where your operation stands and where it needs work. The gaps you have just identified do not require eleven separate tools to fix. They require one connected system that addresses all of them in the same place. The next section shows how Qiscus does that.
How Qiscus Connects Every Audit Area in One System
Qiscus is an agentic customer engagement platform. The three products referenced across the eleven audit areas are not separate tools requiring separate implementation. They are components of one connected system that addresses all eleven audit areas together.
1. Qiscus Omnichannel Chat
Qiscus Omnichannel Chat addresses audit areas 1, 5, 7, 8, and 10. It consolidates every channel, enforces SLA rules, and delivers cross-channel reporting in real time. Qiscus Omnichannel Chat
2. Qiscus AgentLabs
Qiscus AgentLabs addresses audit areas 3, 4, 5, 6, 9, and 11. It handles tier-one queries autonomously, assists agents with AI copilot, and continuously trains from real conversation data. Qiscus AgentLabs
3. Qiscus Helpdesk Suite
Qiscus Helpdesk Suite addresses audit areas 2, 4, 7, 8, and 10. It triggers in-channel CSAT at ticket closure, enforces SLA by channel, automates escalation triggers, and delivers the reporting granularity this audit requires. Qiscus Helpdesk Suite
The reason most audit findings go unaddressed is not lack of intent. It is that fixing one area requires updating configurations in three different tools, and coordination overhead kills momentum. When all eleven audit areas run on one connected system, fixing a routing rule also fixes the SLA clock, the escalation trigger, and the reporting view.
Run the Audit Easily in One Dashboard with Qiscus
A customer service audit checklist is not a one-time exercise. It is a quarterly operational practice that turns the data your operation already produces into specific, prioritised improvement actions.
The eleven areas in this checklist are the operational dimensions that determine whether your customer service delivers consistently or inconsistently.
Most businesses in Malaysia have strengths in one or two areas and gaps in three or four. The audit makes those gaps visible. Visible gaps get fixed. Invisible ones compound into satisfaction problems that are significantly harder and more expensive to address after customers have experienced them.
Qiscus Omnichannel Chat, Qiscus AgentLabs, and Qiscus Helpdesk Suite address all eleven audit areas in one connected system. A configuration change that addresses an audit finding applies across every connected channel simultaneously.
Book a Qiscus demo for your team and run your next audit with a system built to address what it finds.
Frequently Asked Questions About Customer Service Audits
1. How Often Should a Customer Service Audit Be Run?
Run a full eleven-area audit quarterly. Track the two or three metrics most relevant to your current priorities weekly. The quarterly audit is a comprehensive review. Weekly tracking is your improvement feedback loop. Running both catches acute problems quickly and structural drift over time.
2. Who Should Run the Audit?
The audit should be run by whoever has authority to act on its findings. In most businesses in Malaysia, that is the customer service manager or operations lead who owns the platform configuration. If the person running the audit cannot change routing rules or update the knowledge base, the audit produces a report that requires a second approval cycle before any action happens. That second cycle is where most audit findings go to die.
3. What Is the Difference Between a Customer Service Audit and a QA Review?
A QA review evaluates individual interactions against quality standards. Did the agent follow the script? Was the response accurate? Was the tone appropriate? A customer service audit evaluates the system that produces those interactions. Are the routing rules correct? Is the knowledge base complete? Is the SLA configuration enforced? QA reviews identify agent-level issues. Audits identify system-level issues. Both are necessary. But most teams run QA reviews without running audits. So they coach agents on gaps that are actually configuration problems.
4. Can a Small Customer Service Team Run This Audit?
Yes. The audit scales to team size. A team of five agents managing three channels has the same eleven audit areas as a team of fifty agents managing ten channels. The benchmarks adjust for volume, but the areas remain the same. A smaller team benefits more per agent because each identified gap has proportionally larger operational impact when fewer people share the workload.
5. What Should We Do If We Score Below 40% in an Area?
A score below 40% in any single area indicates a foundational gap. The most common foundational gaps are listed below. No SLA configuration on active channels is the most common. CSAT not collected at interaction level. No knowledge base covering top query categories. And escalation triggers not defined or tested. Fix the foundational gap before anything else.
