Product Research During a Decision Task

Sample data and analysis code for a LaBrowser example study.

15 participants chose between three laptops (MacBook Pro M3, ThinkPad X1 Carbon, Dell XPS 15) while answering survey questions in LaBrowser's experiment pane. The browsing data reveals what surveys alone never could: what people say matters to them and what they actually research are often very different.

This repository contains the exported data and a Jupyter notebook that reproduces the analysis shown on the showcase page.

What's in the data

File	Records	Description
data/events.json	~2,600 events	Raw event stream including INPUT_SUBMIT from the experiment pane survey
data/google_search_v1.json	~52 sessions	Derived Google search sessions (participants researching laptops)
data/study_config.json	1 config	Study configuration with experiment pane in URL mode

All files use the exact export format from the LaBrowser Study Console.

Key findings

• Participants who said "price was most important" spent 51% of their browsing time reading performance reviews — and only 15% comparing prices
• Participants who chose the MacBook barely looked at ThinkPad reviews despite claiming they "considered all options equally"
• Scroll depth correlates with decision confidence: participants who scrolled deeper through reviews reported higher confidence in their final choice (r = 0.82)
• This analysis is only possible with LaBrowser — no survey tool, browser extension, or screen recording gives you paired explicit + implicit data in a single timeline

Quick start

# Clone and set up
git clone https://github.com/technologylab-ai/labrowser-example-product-decision
cd labrowser-example-product-decision
pip install -r requirements.txt

# Run the analysis notebook
jupyter lab analysis.ipynb

# Or run the script version
python analysis.py

Requirements

• Python 3.10+
• See requirements.txt for packages (pandas, numpy, matplotlib, seaborn, plotly, jupyter)

How the study worked

LaBrowser's experiment pane (URL mode) displayed a survey alongside the browser pane:

1. Participants browsed freely — searching for laptop reviews, specs, and prices
2. The experiment pane showed survey questions: "Which laptop would you buy?", "What was the most important factor?", etc.
3. Survey responses arrived as INPUT_SUBMIT events with service: "experiment_pane" — in the same event stream as browsing events
4. The analysis pairs what participants said (survey) with what they did (browsing behavior)

Data format reference

Events (events.json)

Each event has:

{
    "id": "uuid",
    "session_id": "uuid",
    "tab_id": "uuid",
    "timestamp_utc": "2025-11-20T10:15:30.123Z",
    "event_type": "NAVIGATE|PAGE_LOADED|CLICK|SCROLL|INPUT_SUBMIT|TAB_OPENED|...",
    "page_id": "uuid",
    "url": "https://...",
    "payload": { }
}

Survey responses are INPUT_SUBMIT events with:

{
    "field_role": "survey_response",
    "service": "experiment_pane",
    "text": "MacBook Pro M3",
    "url": "https://survey.example.com/laptop-study",
    "selector": "...",
    "submit_method": "form_submit"
}

Derived sessions (google_search_v1.json)

Each derived session has:

{
    "id": "uuid",
    "study_id": "uuid",
    "session_id": "uuid",
    "parser_id": "google_search_v1",
    "type": "google_search",
    "start_time": "2025-11-20T10:15:30.123Z",
    "end_time": "2025-11-20T10:20:45.678Z",
    "payload": {
        "query": "MacBook Pro M3 review",
        "serp_url": "https://www.google.com/search?q=...",
        "result_clicks": [...],
        "session_duration_ms": 315555
    }
}

About LaBrowser

LaBrowser is a dedicated research browser for behavioral studies. Participants use it instead of their normal browser — everything inside is logged as structured, typed events. Everything outside is untouched.

Learn more at labrowser.app and see the matching showcase page at labrowser.app/examples/product-decision.