AI Insights

Discover conversation themes and sales signals with AI-powered analysis

AI Insights analyzes your conversations to identify themes (what visitors ask about) and signals (purchase intent, friction points).

Running an Analysis

  1. Go to your domain's Insights tab
  2. Click Run analysis
  3. Select a date range (quick presets: Last 7/30/90 days, or custom)
  4. Optionally name the analysis (e.g. "April week 2")
  5. Click Start analysis

The analysis takes 10-20 seconds depending on the number of conversations.

What You Get

Themes

Conversations are grouped into topics with:

  • Theme name — e.g. "Returns", "Shipping", "Pricing"
  • Count and percentage — how many conversations fall in this theme
  • Example questions — real visitor questions from this theme

Use themes to understand what your visitors care about most.

Sales Signals Beta

Detected purchase intent signals:

  • Pricing Inquiry — asking about prices, discounts
  • Product Comparison — comparing options
  • Purchase Intent — ready to buy
  • Delivery Question — shipping times, costs
  • Bulk Order — large quantity interest
  • Urgency — time-sensitive need

Friction Signals Beta

Detected pain points:

  • Confusion — visitor seems lost
  • Repeated Question — asking the same thing multiple ways
  • Unresolved — conversation ended without resolution
  • Negative Sentiment — frustration or dissatisfaction
  • Off Topic — questions unrelated to your business

Saved Reports

Every analysis is saved and can be viewed later. Click any past analysis to expand its results. Use this to track trends over time — "returns dropped from 40% to 25% after we improved the return policy page."

How It Works

The analysis uses a two-pass AI pipeline:

  1. Pass 1 — each conversation is individually summarized (theme + signals)
  2. Pass 2 — summaries are aggregated into an overall report with merged themes

Conversations in any language are analyzed correctly — themes are always labeled in English for consistent grouping across markets.

ℹ️

Sales and friction signals are in beta. The detection accuracy will improve as we gather feedback from real usage.