AI-Powered Google Reviews Intelligence Across a Global Museum Network

We built a centralized review intelligence system using LLM-powered sentiment analysis to transform thousands of Google reviews across 10+ locations into structured, actionable operational and marketing insights.

They assisted with setting up digital marketing controls and reporting, PR monitoring, and paid advertising support, which helped us improve our ability to track campaign performance and better inform our decision-making.

Client

Museum of Illusions

Expertise
  • Google Reviews Monitoring
  • AI-Powered Sentiment Analysis
  • Multi-Location Performance Tracking
  • Review Trend Detection
  • Operational Intelligence
  • Topic-Level Feature Extraction
Year

2025 - 2026

Museum of Illusions operates 60+ locations across 28 countries, generating thousands of Google reviews monthly. As a location-based entertainment brand, reviews are one of the highest-impact touchpoints in the customer decision journey, yet until now, there was no unified system to monitor, analyze, or act on them across the network.

Client words

Over the past year, Granular has worked with us to support the development of a more structured and data-informed digital marketing approach. They assisted with setting up digital marketing controls and reporting, PR monitoring, and paid advertising support, which helped us improve our ability to track campaign performance and better inform our decision-making.

Their work in implementing tracking and reporting tools contributed to a clearer understanding of our marketing data and operational efficiencies, resulting in a more efficient and effective output overall.

The collaboration has been professional and we consider Granular as an extension of our team as we continue to refine and grow our marketing processes.

Ryan Saddik, Global Director of Marketing

Challenges

  • No Centralized Review Visibility

    With over 10 active locations generating reviews independently, there was no single place to monitor review performance across the network. Each location’s Google Business Profile existed in isolation, making it impossible to compare performance, spot emerging issues, or identify which locations were excelling. Leadership had no consolidated view of how the brand was being perceived across markets.

  • Reviews Were Read, Not Analyzed

    Individual reviews were occasionally read and responded to, but there was no systematic analysis of what customers were actually saying. Patterns across hundreds of reviews — recurring complaints about pricing, praise for specific exhibits, frustrations with wait times — remained buried in unstructured text. Without structured extraction, reviews were anecdotal rather than analytical.

  • No Early Warning System for Declining Locations

    A location could experience a gradual decline in review quality over weeks without anyone noticing. By the time a rating drop became visible, the underlying issue had often been compounding for months. There was no mechanism to flag negative trends in real time or compare current performance against recent baselines.

  • Inability to Connect Reviews to Operational Topics

    Even when reviews were read, there was no framework to categorize feedback into actionable operational categories. A review mentioning “not worth the price” and another saying “too expensive for what you get” were treated as isolated comments rather than signals within a “Pricing & Value” theme that could inform strategic decisions.

  • Response Management Lacked Accountability

    Review response rates varied significantly across locations, with no visibility into which locations were responding promptly and which were leaving reviews unanswered. Given the direct impact of response behavior on Google ranking signals and customer perception, this gap carried real business consequences.

Solution

  • Centralized Multi-Location Review Dashboard

    We built a unified monitoring system that aggregates Google review data across all tracked Museum of Illusions locations into a single dashboard. The system provides both an all-time cumulative view and rolling time-windowed analysis (last 14 months, last 30 days), enabling both strategic overview and tactical responsiveness.

  • 30-Day Trend Detection Engine

    The dashboard automatically compares the most recent 30-day window against the prior 30-day period to surface changes in both rating and review volume. Each location receives a dynamically calculated status, flagging whether review trends are very positive, neutral, negative, or very negative so that locations needing attention are immediately visible without manual analysis.

  • LLM-Powered Sentiment and Feature Extraction

    Every review is processed through a large language model pipeline that performs two key functions: it assigns a sentiment score (ranging from -1 to +1 across categories from very negative to very positive), and it extracts the specific feature or topic being discussed. This transforms unstructured review text into structured, queryable data.

  • Topic Taxonomy for Operational Insights Reviews are automatically classified into a predefined set of operational topic categories:

    • Expectations & Experience
    • Safety & Accessibility
    • Pricing & Value for Money
    • Management & Organization
    • Atmosphere & Ambiance
    • Interactivity & Engagement
    • Exhibits & Collections
    • Amenities & Facilities
    • Photography & Visual Experience
    • Family & Child-Friendliness
    • Staff & Service
    • Memorable Highlights
    • Return Visit Likelihood

    This taxonomy was designed to map directly to the operational levers MOI can actually control, ensuring that insights are immediately actionable.

  • Sentiment Distribution Visualization

    A scatter plot visualization maps each topic category by both sentiment score and volume of mentions, with bubble size indicating frequency. This allows leadership to instantly identify which topics generate the most discussion and whether that discussion is positive or negative — revealing, for example, that “Family & Child-Friendliness” drives high volume with neutral sentiment (an improvement opportunity), while “Memorable Highlights” generates high volume with strongly positive sentiment (a brand strength to amplify).

  • Review-Level Drill-Down

    Beyond aggregated metrics, the system preserves access to individual review features — the specific sentence or phrase extracted from each review, its assigned topic, sentiment score, location, and date. This enables teams to move from macro trends to specific customer verbatims when investigating an issue or preparing operational recommendations.

Results

  • Unified Network Visibility All locations now report through a single review intelligence dashboard, enabling instant cross-location performance comparison and trend identification.
  • Proactive Issue Detection The 30-day trend engine surfaces declining locations before rating damage becomes significant, shifting the approach from reactive to preventive.
  • Operational Insight from Unstructured Data LLM-powered analysis converts thousands of reviews into structured topic-level sentiment data, revealing patterns invisible to manual review reading.
  • Response Rate Accountability Tracking of review response rates across locations established clear visibility into which markets are managing their review presence and which require process improvements.
Review Ecosystem Audit

We mapped all active Museum of Illusions Google Business Profiles, assessed existing review volumes and ratings, and identified the full scope of locations requiring monitoring. This established the baseline dataset and surfaced immediate gaps in review management.

Topic Taxonomy Design

Working with stakeholders, we defined the operational topic categories that reviews would be classified into. The taxonomy was designed around MOI’s actual operational structure — ensuring each category maps to a team or decision-maker who can act on the insights.

Sentiment Analysis Pipeline Development

We built the LLM-powered processing pipeline that ingests raw review text, extracts the discussed feature or topic, assigns it to the correct taxonomy category, and scores sentiment on a continuous scale. The pipeline was calibrated against manually labeled review samples to ensure accuracy.

Dashboard Architecture and Build

The interactive dashboard was constructed with multiple analysis layers: an executive summary with 30-day trend detection, a location breakdown with time-series charts and KPI scorecards, and a sentiment analysis section with both longitudinal tracking and topic-level distribution views.

Historical Data Processing

All available historical review data was processed through the pipeline to establish baselines, enabling meaningful trend analysis from day one rather than requiring months of data accumulation before delivering value.

Alert and Trend Logic Implementation

We implemented the automated comparison logic that evaluates 30-day rolling windows, calculates rating and volume changes, and assigns status labels (very positive, neutral, negative, very negative) to each location — creating an at-a-glance early warning system.

Ongoing Monitoring and Refinement

The system runs continuously, processing new reviews as they appear. The topic taxonomy and sentiment calibration are periodically reviewed and refined as new patterns emerge or operational priorities shift.