Cluster (Segmentation) analysis
A good segmentation has the ability to be the bedrock upon which all proposition development and customer activity sits. We know from experience that segmentations fail most often due to a lack of clarity on the objectives for the segmentation, or due to poor implementation. Of course, there’s a need to get the analytical part which lies in between right, and that is where Cluster Analysis comes in. Cluster analysis collects things (people, normally) into groups based on information about them. The analysis will aim to include similar people in the same groups, and to spread people who are different to each other further away. A successful cluster analysis will look for several tight-knit, evenly distributed, distinct and easily explainable clusters. There are a number of differing approaches to arrive at the most commercially intuitive cluster solution, and we will bring our wealth of experience in these approaches (k-means, hierarchical cluster, latent class, OCEAN) together to develop clusters that are easily identifiable and targetable.
One of the often-mentioned issues with segmentation outcomes is mapping specific behaviours and features in the real world. Discriminant analysis is often used to find those few “golden questions” that can determine the outcome of the prediction without needing all of the data. This can be particularly useful when you can predict in which segment a consumer belongs without needing them to fill in the entire online survey!
Data reduction approaches
Sometimes, you can end up with hundreds of different data variables or questions, and that can be too many. These can often be simplified, into themes, i.e. service, cleanliness, or timeliness. Simplifying lots of different data points can really help with further analysis, and can make understanding the data and patterns easier. Our approaches include Exploratory and Confirmatory Factor Analysis, Principle Components Analysis and Random Forests/Ensemble Trees.
Advanced key drivers analysis
Our suite of advanced key drivers analysis tools are used to distinguish the key drivers of a certain measure. Depending on the type of regression, this type of analysis takes into account that several of the drivers may be interlinked (i.e. satisfaction with customer service and length of time waiting on the phone). Key drivers analysis is often used for key KPIs, such as overall customer satisfaction, NPS, etc. Our battery of tools include multivariate regression methods, logistic approaches (for when the dependent variable is a binary measure), and Structural Equation Modelling (SEM) methods including Partial Least Squares (PLS) and path modelling, which can be used to understand more complex and latent relationships between variables. We also use CHAID approaches to help understand predictors of certain behaviours (e.g most satisfied customers), yielding highly visual decision trees that can be easily interpreted.
Developing winning propositions in competitive markets means understanding the specific proposition elements that drive customer preferences. The area of Choice modelling enables us to do this with a greater degree of certainty. Choice models fall into several categories:
Simple (Pairwise) Trade Off is a relatively straightforward approach which forces participants to choose preferences across a number of alternatives. Participants are asked to allocate seven points between two alternatives in an iterative experiment. From these paired scores, the results are indexed to determine which alternatives are the most and least preferred.
Maximum Difference Scaling (Max Diff) analysis is a type of trade-off technique which makes it possible to understand people’s priorities from a long list of features that may be closely linked to another (i.e. importance of clean carriages, working air conditioning, and clean platforms at train stations). Participants will be shown groups of features and asked to choose the best and worst of the group – this process results in a calculated level of importance or preference for each feature.
Conjoint analysis spans a range of similar trade-off techniques, and often looks at how consumers make decisions between “bundles” of products. Where services or products are offered as bundles (i.e. a number of texts, minutes, and data available in a mobile phone contract), we can use conjoint to understand how much more, if any, consumers are willing to pay for an extra 2GB of data, or whether they even value that free music subscription you’re offering. We have experience in a wide range of specific approaches, including ACA, CBC, ACBC and Menu-based conjoint.
Cost is always a difficult subject to discuss with research participants – if asked, most will claim they want as much as they can for as little as they can! There are several different pricing techniques that can determine optimal price ranges for products or services based on other information that is easier to ask, and easier to answer. These details can also be used to understand how prices might affect consumer demand, revenue, and profitability. Our range of pricing approaches include the Gabor-Granger method, Price Sensitivity Meter (also known as Van Westendorp) and Brand Price Trade Off (BPTO).
TURF analysis (Total Unduplicated Reach and Frequency) is particularly useful to understand which product, marketing messages, etc. will reach the widest audience whilst taking into account that some consumers might be easily influenced by multiple products or messages.
It quite often doesn’t make any sense to have the same benchmarks and the same targets for everyone, because you’re not comparing apples with apples. A 100m relay attempt our team might be proud of would make the likes of Bolt want to resign. Intelligent benchmarking takes into account different factors (i.e. Bolt is an Olympian, we are… not…) and re-aligns benchmarks so that both us and Bolt can aim for something realistic.
Approaches for non-structured data – Sentiment and Textual Analysis
Sometimes, you might have access to open text data without a score attached to quantify the emotion, or sentiment behind the text. Automatic sentiment analysis uses algorithms and machine learning to explore the words used in the text – the types of words, the order of the words, etc. to quantify the thoughts, themes, and emotions that may not have been explicitly mentioned.
Open ended text data is often thought of as unruly, and difficult to analyse. We can take this data and use machine learning to understand it and automatically categorise open ended text into themes, saving hours of manual work. Text can also be scored on sentiment and emotions, and can be analysed for patterns.
Approaches for web and digital data – Web Scraping and Passive Metering
There is so much information available on the internet. As you’re browsing your competitors’ sites, you’re skimming past reams and reams of valuable information about them; their products, their prices, their discounting strategies…. Web scraping means we can take all of this data from under their noses, analyse it, and give it back to you, to help you compare your business with theirs and ensure you stay ahead of the market.
Advancements in technology and levels of smartphone adoption now also means we can track real-time behaviour data from participants’ devices. We ask participants to download an app which runs in the background of their phone and collects data on their activities and behaviours. We can then use this data to inform statistical modelling and profiling of consumers, prospects, brand advocates, etc. Behavioural data can also be linked to survey data to provide a more rounded picture, combining the “what” with the “why”.
What-if? Approaches – Forecasting and Simulation
Armed with your internal data, we can analyse past behaviours, stock levels, resource levels (staffing or otherwise), etc. in order to predict future forecasting. We often extend the data available for this type of analysis by using additional external data, such as weather patterns, or local footfall to improve the accuracy of context analysis. Forecasting can help with stock ordering and staff planning amongst other business critical decisions that are often made on estimates, or assumptions.