Correlation Research & Analysis
A correlation research was conducted during my data analytics internship with The Ascott Limited, an owned lodging business unit of CapitaLand Investment that operates a massive global portfolio of serviced apartments, hotels, and co-living spaces under numerous brands with over 970 properties across 14 brands in over 40 countries.
I analysed three years of monthly data across 300+ hotels in 20 countries to answer one business question asked by leaders of my department: does customer satisfaction actually translate into stronger commercial performance?
To reach a conclusion, I gather and compiled data, followed by building various analytical charts comparing four satisfaction metrics (CSAT, NPS, GRI, CQI) against 4 business performance metrics (RGI, Occupancy%, Revenue vs Budget %, GOP%).
Stack: Python (NumPy, Matplotlib, Seaborn, Pandas)
Getting Data
Extract and ingest raw data: guest satisfaction scores (GRI, CQI, CSAT, NPS), business performance metrics (RGI, Occupancy %, Revenue vs Budget %, GOP %), and booking metrics (reviews count and avg booking window) were extracted from the database into an Excel and ingested.
Clean data: Standardising through a cleaning pipeline that ensured a consistent structure, resolved property identifiers, handled missing months, and deduplicated records.
Merging data: All sources were then merged onto a full property × year × month skeleton to ensure every property had a complete time series across 2023-2025.
Analysis
Establish assumptions: Acknowledged limitations of this research such as correlation ≠ causation, sample size per quartile, and external factors (e.g., seasonality, macro events) which are uncontrollable.
Analysis: The cleaned master dataset was analysed cross-sectionally. The use of various chart types were involved in this step namely, line chart, scatter plot, bar chart, and more.
To uncover how customer satisfaction translates into business outcomes, I ran pairwise correlation analysis across every CSAT and business performance metric. This is followed by extending it across time lags of 1 to 12 months which revealed not just which metrics move together, but when. This answers key business questions like: how long does it take for customer satisfaction to show up in business performance?
Bringing in number of reviews and avg booking window allows further analysis to answer key business questions like: is the time lag longer because the booking window is long? does the number of reviews impact the time lag?
Properties were segmented into quartiles by satisfaction score within each year with avg scores of the business performance for quarterly comparison.
Challenge
Over the years, the hospitality industry research consistently shows that guest satisfaction drives revenue.
But when I ran the standard correlation analysis on our portfolio, the relationship had weak correlation compared to the published industry benchmarks (r ≈ 0.42).
The question became: is satisfaction really a weak driver in our portfolio, or is the correlation hiding the real story? Perhaps taking a step back and evaluate whether the current method of analysing the relationship is the most suitable.
Approach to overcome challenge
Through the use of AI as a learning stepping stone while avoiding data biasness, it discovered interesting patterns of evaluating it from a wider perspective. This enabled me to build a four-view analytical view for each satisfaction metric:
Quartile gap: split properties 4 quartiles including the top and bottom 25% by satisfaction and compared their RGI.
Threshold effect: tested cutoff values to find where the the RGI gap between the top and bottom quartile is the most prominent and compare the customer satisfaction values.
Consistency theory: identified properties in the top quartile for 3+ consecutive years vs other properties, and compare their RGI.
Scatter view: visualised the underlying property-level distribution against the threshold set above.
Applied across all four satisfaction metrics so the story could be told consistently and defensibly.
Conclusion
Specific findings are confidential, but the analysis delivered actionable insights that shaped the recommendation presented to senior management, specifically a shift in the FY benchmark from tracking multiple customer satisfaction metrics to consolidating around a single primary metric discovered through this research.