live customer service analytics

Analytics here are not vanity dashboards. Every conversation is logged, and two dashboards turn that into something you can act on. This article describes what each one gives you and how to use it.

The Chat Dashboard

Every conversation is logged and inspectable:

• Search, filter, and tag conversations.
• A per-conversation quality rating.
• Per-answer response time.
• Incremental loading with auto-refresh on a live window, so new conversations appear as they happen.

Per-answer feedback (correct, incorrect, or unmarked) flows into this dashboard. You can filter to flagged answers and inspect the sources that were retrieved and the model that produced the answer.

The Analytics Dashboard

A separate Analytics Dashboard summarizes the patterns:

• What customers ask, clustered from the raw questions.
• Where your content has gaps.
• An accuracy view, where per-answer feedback rolls up so you can see how often answers were marked correct.

Closing content gaps

The "where your content has gaps" insight is the one most teams act on first. It surfaces the questions where your published content does not cleanly answer what customers ask. The recommended workflow:

1. Review the gaps weekly (or live, during launches).
2. Group similar questions.
3. For each group, add the missing information and re-index. You can add it to your website, or capture it as manual Q&A through the Instructions field or an uploaded document.
4. Verify the next time those questions arrive that the answer is now correct.

This loop is how the knowledge base gets sharper over time without you guessing what is missing.

A worked example

A retailer launches a new product line on Monday. By Tuesday afternoon, the Analytics Dashboard shows a cluster of questions about "international shipping for the new collection." The product pages mention shipping in general terms, but customers want specifics: which countries, customs handling, delivery time.

The team opens those conversations in the Chat Dashboard to read the exact phrasings, adds a "Shipping the new collection" section to the product launch page covering eight common questions, and re-indexes. By Wednesday that topic stops showing up as a gap, and the related answers come back marked correct.

This is concrete: the analytics surfaced a gap, the fix was visible in the data within a day.

Watching conversations live

The Chat Dashboard auto-refreshes on a live window, so you can watch conversations as they come in. You can:

• Search, filter, and tag conversations.
• Open a conversation and read the full transcript, including the sources retrieved and the model used.
• Tag conversations for follow-up review.

This is useful during launches or when something has gone wrong in your business (an outage, a recall) and you want to see how the front line is handling it in real time.

Exports

You can export from the Chat Dashboard in three formats, respecting your current filters and columns:

• CSV.
• A pretty-printed JSON array.
• A styled Excel workbook (frozen header, auto-sized columns).

This lets you join conversation data with order data, marketing spend, or product analytics in your own tools to ask larger questions of your own.

CSAT and NPS

Customer-satisfaction surveys run through an open-source CSAT survey and analytics platform that you self-host on your own infrastructure. Paste your CSAT site URL and token secret into the Guru admin, choose delivery (SMS via your own Twilio, email, or both), and survey responses live in your self-hosted dashboard.

What to look for after a month

A useful first-month review using the Analytics Dashboard:

• Which topics customers ask about most, and how concentrated that volume is.
• How often those top answers come back marked correct in the accuracy view.
• The content gaps you found initially, and whether they are closed now.

This is operations data, not vanity data. It tells you what your customers want and where your content is failing them.