# Find out where my customer service chatbot fails and fix it

An analysis of your chatbot's conversations and metrics, with the gaps, root causes, and prioritized fixes to resolve more questions.

- Topic: Customer Support
- Framework: Few-Shot
- Level: Standard
- Tags: Customer Support, Analyze, Fill-in Template
- Source: https://www.thinkwithprompt.com/prompts/find-out-where-my-customer-service-chatbot-fails-and-fix-it

## Prompt

```text
Analyze my customer service chatbot's performance and tell me how to improve it.

**What the bot handles:** [products/services, industry, channels]
**Metrics I have:** [e.g. containment rate, handoff rate, CSAT, conversion, fallback/"didn't understand" rate, drop-off points]
**Conversation samples:** [paste 20-50 transcripts, ideally including failures and handoffs]
**User groups:** [e.g. new vs. returning customers, mobile vs. desktop, regions]
**What the bot can access:** [FAQ, help center, order system, account data, none]
**Goal:** [e.g. resolve more without a human, increase sales, cut wrong answers]

Analyze:
1. **Failure patterns:** group failed conversations by cause: missing knowledge, misunderstood intent, no access to needed data, policy limits, bad flow design, or wrong answer. Count each and quote one example.
2. **Accuracy risk:** answers that were wrong, outdated, or made up, and where they came from.
3. **User groups:** how different groups use the bot and where each struggles.
4. **Handoffs:** whether people reach a human at the right moment, with context passed along.
5. **Metrics:** what the numbers suggest, and which metric I'm missing.

Return:
- A table of issues: issue, frequency, impact, fix (content, flow, data integration, or handoff rule), effort (low/med/high).
- The top 5 fixes in priority order.
- Rewritten responses for the 3 worst examples.
- A test set of 10 questions to rerun after the changes.

Only use numbers from my data; label anything estimated.
```

## What to fill in

- products/services, industry, channels
- e.g. new vs. returning customers, mobile vs. desktop, regions
- FAQ, help center, order system, account data, none

Free to copy and use. From ThinkWithPrompt (https://www.thinkwithprompt.com/), a library of 2,663 curated AI prompts.