VoxTrends
Every opinion. Every channel. One truth.
VoxTrends scrapes, transcribes and analyses customer sentiment across 15+ channels — social media, forums, review sites, call centres, helpdesks and more — and turns raw noise into store-level, topic-level, actionable insights powered by LLM evaluation.
Not a dashboard of charts nobody reads. A system that tells you: "Your store in Pelhřimov has a cleanliness problem. Your store in Ostrava has long queues at checkout. Here is the evidence, here is the trend, here is what to do."
From raw noise to store-level truth
VoxTrends does not show you a wall of charts. It tells you which location has which problem, grounded in cited verbatim from real customers. Every insight is traceable to a source URL, timestamp and author.
Every channel, one pipeline
Social posts, forum threads, Google reviews, call transcripts, helpdesk tickets, app store reviews, mystery shopper reports — all normalised into one schema, one topic taxonomy, one dashboard.
LLM-evaluated, not keyword-counted
Classical sentiment analysis counts positive and negative words. VoxTrends reads the full context, understands sarcasm, detects the specific issue, and summarises it in natural language. The difference is night and day.
15+ channels. One unified view.
VoxTrends ingests customer voice from every touchpoint — digital and physical, public and private, text and voice. Nothing slips through.
Social media
CoreFacebook, Instagram, X (Twitter), TikTok, LinkedIn, YouTube comments, Threads. Continuous monitoring of public posts, comments, mentions and hashtags via official APIs and compliant web scraping.
Review platforms
CoreGoogle Maps / Reviews, Trustpilot, Yelp, Heureka.cz, Zboží.cz, Tripadvisor, G2, Capterra. Structured star ratings + free-text review body, matched to individual locations.
Internet forums & communities
CoreReddit, Quora, Pair, Stack Overflow, niche industry forums, Discord public servers, Telegram public groups. Threaded discussion crawling with context-aware reply chain analysis.
News & online media
National and regional online news sites, press releases, industry portals, blog aggregators. RSS, Atom and custom scraping. Brand-mention extraction and journalist-sentiment scoring.
App store reviews
Google Play, Apple App Store. Version-correlated review tracking — detect spikes after releases, identify UX regressions, compare to competitive apps.
E-commerce marketplaces
Amazon, Allegro, eBay, Kaufland Marketplace. Product review ingestion tied to SKU, seller and fulfilment channel. Detect quality issues before they scale.
Call centre (voice-to-text)
CoreIntegration with cloud telephony (Genesys, Five9, Twilio, Aircall) or on-prem PBX. Calls are transcribed in real time, diarised per speaker, and fed into the VoxTrends pipeline alongside digital channels.
Email & helpdesk tickets
Zendesk, Freshdesk, Intercom, ServiceNow, JIRA Service Management, shared inboxes. Every inbound ticket and reply is ingested, categorised and sentiment-scored.
Live chat & messaging
Intercom, Drift, LiveChat, WhatsApp Business, Facebook Messenger, web widget transcripts. Real-time conversation analysis with immediate escalation alerts.
Surveys & NPS
Qualtrics, SurveyMonkey, Typeform, Google Forms, custom survey tools. Structured scores (NPS, CSAT, CES) combined with free-text verbatim analysis.
In-app feedback
Embedded product feedback widgets (Canny, UserVoice, custom), feature request boards, in-app bug reports. Links sentiment to specific product features and releases.
Employee & mystery shopper
Internal feedback channels, anonymous employee surveys, mystery shopper reports. Cross-referenced with external customer sentiment for a 360° view of each location.
How VoxTrends works — from raw data to actionable insight.
Five stages, fully automated. From scraping a TikTok comment to a board-ready PDF that says "Ostrava has a queue problem" — with every step traceable and auditable.
- 1
Ingest & scrape
Continuous, compliant data collection across all configured channels. Social media via official APIs (Meta Graph, X API v2, YouTube Data) supplemented by headless-browser scraping for platforms without API access. Forums and review sites via targeted crawlers with rate limiting, fingerprint rotation and robots.txt compliance. Call centre audio via streaming integration with cloud PBX. All data lands in a unified raw-event store with provenance metadata (source URL, timestamp, author ID, geo if available).
- 2
Normalise & enrich
Raw events are normalised into a canonical schema: author, timestamp, text, source, channel, language, location (resolved via geo-tagging, store mention detection, or IP/phone prefix). Language detection triggers automatic translation for multilingual markets. PII is stripped and redacted before downstream processing (GDPR by design).
- 3
NLP & LLM analysis
Each normalised event passes through a three-stage pipeline. Stage 1: classical NLP for fast sentiment scoring (positive / negative / neutral / mixed) and named entity extraction (store names, product names, competitor mentions). Stage 2: LLM-powered topic extraction — the model reads the verbatim, identifies the specific issue ("queue length at checkout", "expired produce in aisle 3") and assigns it to the topic taxonomy. Stage 3: LLM summarisation — groups of related events are summarised into natural-language insights with citations.
- 4
Aggregate & score
Events are aggregated into a multi-dimensional cube: by location, by topic, by channel, by time period. Composite scores are computed per dimension (e.g. "Store Pelhřimov, topic: cleanliness, score: 2.1/5, trend: declining"). Anomaly detection runs on every aggregation update, triggering alerts when thresholds are breached.
- 5
Visualise & act
Eight pre-built dashboards (customisable) and a natural-language query interface. Stakeholders ask questions ("Which stores had a cleanliness complaint spike this week?") and VoxTrends answers with data, charts and source citations. Exports to PDF, Slack, Teams, email digest. API for integration into existing BI stacks (PowerBI, Tableau, Looker).
Social media scraping — how we actually do it.
Social media is the highest-volume, most time-sensitive channel in the VoxTrends pipeline. Getting it right requires a layered strategy that combines official API access with compliant supplementary scraping.
API-first, scraping as fallback
For platforms that offer a commercial API — Meta (Facebook, Instagram), X (Twitter), YouTube, LinkedIn — VoxTrends connects via authenticated API endpoints. This guarantees rate-limit compliance, access to structured metadata (author, timestamp, engagement counts, geo tags) and legal clarity.
For platforms without a public API (or with prohibitively expensive tiers), VoxTrends uses a headless-browser scraping layer built on Playwright. Pages are rendered in a real Chromium instance, JavaScript is executed, infinite-scroll is driven programmatically, and the rendered DOM is parsed for post content, timestamps and engagement signals. We rotate browser fingerprints, IP addresses (residential proxy pool) and timing intervals to stay within platform rate limits and terms-of-service boundaries.
Real-time streaming vs. periodic polling
High-priority channels (brand mentions, competitor mentions, crisis keywords) are monitored in near-real-time via streaming API subscriptions (where available) or 60-second polling cycles. Lower- priority channels (historical sentiment, competitive benchmarking) are refreshed on a configurable schedule — typically hourly or daily.
Deduplication & threading
Cross-posted content (the same complaint appearing on Facebook, Google Reviews and Reddit simultaneously) is deduplicated via content fingerprinting (SimHash) and linked into a single "voice event". Reply threads and comment chains are reconstructed so the LLM analysis has full conversational context, not just isolated fragments.
Geo-resolution
For multi-location businesses (retail chains, restaurant groups, franchises), VoxTrends resolves every voice event to a specific store or region. Resolution sources: explicit geo-tags, store mentions in text ("your Pelhřimov branch"), Google Reviews location binding, phone prefix matching for call centre transcripts, and IP-based approximation for web submissions.
Compliance
All scraping respects robots.txt, platform terms-of-service, and GDPR. Personal data is pseudonymised at ingestion; raw author identifiers are stripped before LLM processing. Data retention policies are configurable per client and per jurisdiction.
Eight dashboards — ready to deploy, fully customisable.
VoxTrends ships with eight pre-built dashboards that cover the most common Voice of Customer use cases. Each one is customisable — filters, thresholds, export formats, notification rules — and integrates with your existing BI stack (PowerBI, Tableau, Looker) via API.
Sentiment Over Time
Line chart of daily/weekly sentiment (positive, negative, neutral) across all channels or filtered by location, topic, or channel. Spot downward trends before they become crises.
Topic Clustering
LLM-powered automatic topic extraction from raw feedback. Bubble chart showing volume vs. sentiment per topic — "cleanliness", "staff friendliness", "queue length", "product availability". Click into any topic for verbatim examples.
Store Heatmap
Geographic heatmap of all locations, coloured by composite sentiment score. Drill down to individual stores and see their top 3 issues ranked by frequency and severity. Directly comparable across regions.
Channel Comparison
Side-by-side bar chart showing sentiment distribution per channel (social, reviews, call centre, chat…). Identify where the loudest negative voices are — and whether the same issues repeat across channels.
Alert Dashboard
Real-time feed of critical signals: new 1-star review spike, viral negative post, call centre escalation cluster, sudden NPS drop. Each alert links to the source data and a suggested response. Configurable thresholds and Slack/Teams/email notifications.
Competitive Benchmark
Compare your brand sentiment, topic mix and channel presence against competitors — scraped from the same public sources. Identify positioning gaps and track share of voice over time.
Trend Detection
Time-series anomaly detection: flag emerging issues (new complaint patterns, product defect mentions, regulatory concerns) before they hit critical mass. Powered by LLM pattern recognition across all channels.
Executive Summary
LLM-generated weekly brief: the 5 most important things that happened in customer sentiment this week, with evidence, context and recommended actions. Board-ready, exportable as PDF.
See VoxTrends on your data.
We run a free, no-commitment pilot on your brand: we scrape your public channels for 7 days, build the dashboards, and walk you through what VoxTrends found. Takes 30 minutes of your time. Zero risk.