Clustering and Segmentation are important data analysis techniques used to identify groups of similar records, customers, products, regions, or business entities. In Tableau, clustering helps analysts discover hidden patterns in data and divide similar marks into groups based on selected variables.
Segmentation is the broader process of dividing data into meaningful categories for analysis and decision-making. Clustering is one technique used to create segments automatically based on similarities in the data.
For example, an e-commerce company can analyze customer spending, order frequency, and purchase value to discover groups such as high-value customers, occasional buyers, and low-spending customers.
Clustering is an unsupervised machine learning technique that divides observations into groups called clusters. Observations within the same cluster are generally more similar to one another than observations belonging to different clusters.
In Tableau, clustering can group marks in a visualization according to selected variables. Tableau assigns different colors to the resulting clusters so users can identify patterns visually.
Suppose a retail company has customer data containing annual spending, number of orders, and average order value. Clustering may reveal three groups:
These are illustrative interpretations. Actual clusters depend on the available data, selected variables, and model configuration.
Segmentation is the process of dividing a larger population or dataset into smaller groups based on shared characteristics. Segments can be defined using business rules, demographic attributes, categories, or statistical techniques such as clustering.
| Feature | Clustering | Segmentation |
|---|---|---|
| Meaning | Groups observations according to similarities discovered by an algorithm. | Divides a population into meaningful groups. |
| Approach | Usually data-driven and unsupervised. | Can use business rules, existing categories, or statistical methods. |
| Group Creation | Determined by the clustering model and input variables. | Defined by analysts, business rules, or models. |
| Example | Automatically discovering groups of customers with similar purchase patterns. | Classifying customers as new, returning, or loyal based on predefined rules. |
| Purpose | Discover hidden structure in data. | Make groups easier to understand and act upon. |
| Tableau Use | Use the Cluster feature in the Analytics pane. | Use groups, sets, calculated fields, filters, or clustering results. |
Clustering is generally considered an unsupervised learning technique because it does not require a predefined target label for every observation. Instead, the algorithm looks for patterns in the input variables.
| Feature | Supervised Learning | Unsupervised Learning |
|---|---|---|
| Training Data | Uses labeled examples with known outcomes. | Uses data without a predefined target label. |
| Goal | Predict an outcome or class. | Discover structure, patterns, or groups. |
| Example | Predict whether a customer will cancel a subscription. | Group customers according to purchasing behavior. |
| Common Methods | Regression and classification. | Clustering and dimensionality reduction. |
Tableau uses the K-means clustering algorithm for its built-in clustering feature. K-means attempts to divide marks into a specified number of clusters so that each mark belongs to a cluster with a relatively nearby center.
K-means is a clustering algorithm that partitions data into K clusters. Each cluster has a centroid representing the mean of the observations assigned to it for the numerical variables used by the model.
A company wants to group customers using two variables: annual spending and number of orders. If K equals 3, the algorithm attempts to create three clusters with similar customers grouped together.
Input Data
Customer ID | Annual Spending | Orders
C001 | 50000 | 20
C002 | 45000 | 18
C003 | 10000 | 4
C004 | 12000 | 5
C005 | 30000 | 12
Clustering Process
1. Select the input variables.
2. Initialize cluster centers.
3. Assign customers to the nearest center.
4. Recalculate cluster centers.
5. Repeat assignments until convergence.
6. Review the resulting customer groups.The customer groups shown in this example are not predetermined; the algorithm determines membership from the data and model settings.
Data preparation is essential because clustering results depend on the fields selected, the quality of the data, and the level of detail in the visualization.
Open Tableau Desktop and connect to a suitable source, such as Excel, CSV, or a supported database.
Decide what each mark should represent. For customer segmentation, each mark may represent one customer. For product segmentation, each mark may represent one product.
If the view aggregates several customers into a single mark, the resulting clusters may represent aggregated groups rather than individual customers.
Choose variables that relate to the business question. For customer analysis, useful measures might include total sales, order count, average order value, and days since the last purchase.
Investigate outliers, duplicate records, missing values, and fields that may not contribute useful information. Tableau can assign records with missing clustering measures to a Not Clustered category.
Tableau provides a built-in Cluster feature in the Analytics pane. Follow these steps to create a cluster analysis.
In Tableau Desktop, the number of clusters can be specified between 2 and 50. If the number is not specified, Tableau selects a cluster count automatically, up to its default search limit of 25.
Variables are the fields used by the algorithm to compare marks and determine cluster membership. Selecting appropriate variables is one of the most important decisions in clustering.
| Business Objective | Possible Variables | Potential Insight |
|---|---|---|
| Customer Segmentation | Total Sales, Order Count, Average Order Value | Groups with different purchase patterns. |
| Product Segmentation | Sales, Profit, Quantity Sold | Groups of products with similar commercial performance. |
| Regional Segmentation | Sales, Profit Margin, Customer Count | Regions with similar performance characteristics. |
| Employee Analysis | Experience, Training Hours, Performance Score | Groups with similar workforce metrics. |
| Marketing Analysis | Campaign Spend, Conversions, Clicks | Campaigns with similar performance profiles. |
Variables can have very different numerical ranges. For example, annual revenue may range into millions, while order frequency may range from 1 to 100. If variables are not scaled appropriately, a large numerical range can dominate distance calculations.
Tableau's built-in clustering feature automatically uses min-max normalization for its clustering inputs. This maps each numeric variable to a scale between 0 and 1 when the variable has a nonzero range.
Min-Max Normalization
Normalized Value = (x - Minimum) / (Maximum - Minimum)
Example:
Minimum Revenue = 10
Maximum Revenue = 110
Observed Revenue = 60
Normalized Value = (60 - 10) / (110 - 10)
Normalized Value = 50 / 100
Normalized Value = 0.5Normalization helps reduce the influence of differences in numerical scale, but it does not remove the effect of unusual outliers. Review the data and interpretation before relying on the clusters.
The number of clusters affects the level of detail in a segmentation. Too few clusters may combine distinct patterns, while too many clusters may create groups that are difficult to interpret or act upon.
These labels are possible interpretations, not guaranteed outcomes. Tableau does not automatically know which group should be called high-value or low-value; analysts must examine cluster characteristics.
When Tableau selects the number of clusters automatically, it evaluates cluster quality using the Calinski-Harabasz criterion. This criterion compares separation between clusters with variation within clusters. A larger value generally indicates more distinct clusters relative to their internal spread.
Use automatic selection as a starting point, then compare alternative cluster counts for stability, interpretability, and business usefulness.
After Tableau creates clusters, examine the variables and characteristics of each group. Do not assume that Cluster 1 is better or more valuable than Cluster 2 just because of its number or color.
Tableau provides a Describe Clusters dialog box to help inspect cluster characteristics and model information.
| Statistic | Meaning | Interpretation |
|---|---|---|
| Number of Clusters | The number of clusters produced by the model. | Shows how many groups were created. |
| Number of Points | The number of marks included in the clustering view. | Shows the number of observations represented by the marks. |
| Cluster Size | The number of marks assigned to a particular cluster. | Helps identify large and small groups. |
| Cluster Centers | Average numerical values for variables within each cluster. | Helps characterize each group's typical profile. |
| Between-Group Sum of Squares | Measures variation between cluster centers and the overall center. | Larger separation can indicate more distinct groups. |
| Within-Group Sum of Squares | Measures variation of marks around their cluster centers. | Smaller values indicate tighter groups, though the value also depends on the number of clusters. |
| F-Statistic | Measures the relative variation between and within groups for a variable. | Can help identify variables that distinguish clusters. |
After creating clusters, you may want to reuse them in other worksheets, filters, or dashboards. Tableau allows cluster results to be saved as a group dimension.
The saved group and the original clustering result become separate objects. Changes to the original clustering do not automatically update the saved group, and editing the saved group does not change the original model results.
To use meaningful business names, first save the clustering result as a group. Then edit the saved group and rename its members.
For example, after reviewing customer behavior, an analyst might label groups as Frequent High-Value Customers, Occasional Customers, and Low-Activity Customers. These names should be based on the actual cluster characteristics rather than assumptions.
Use Tableau clustering to identify groups of customers according to their purchase behavior and develop a dashboard for business analysis.
| Customer ID | Annual Sales | Orders | Average Order Value |
|---|---|---|---|
| C001 | ₹1,20,000 | 24 | ₹5,000 |
| C002 | ₹1,00,000 | 20 | ₹5,000 |
| C003 | ₹25,000 | 5 | ₹5,000 |
| C004 | ₹18,000 | 3 | ₹6,000 |
| C005 | ₹60,000 | 12 | ₹5,000 |
| C006 | ₹35,000 | 7 | ₹5,000 |
The sample values are illustrative. A real clustering model needs enough observations to identify meaningful patterns, and its results may differ from any expected business categories.
Group products according to sales, profit, quantity sold, and return rates. This may help identify high-performing products, low-volume products, or products that require further investigation.
Compare regions using sales, profit margin, customer count, and order frequency. Clusters may reveal areas with similar business performance.
Analyze campaign cost, clicks, conversions, and revenue to discover groups of campaigns with similar outcomes.
Group stores according to revenue, foot traffic, transaction count, and product mix to support operational planning.
Use appropriate workforce metrics to explore patterns in training, performance, and experience. Interpret the results carefully and avoid treating cluster membership as a standalone judgment of an employee.
| Problem | Possible Cause | Solution |
|---|---|---|
| Cluster option is unavailable | The view or data source does not meet clustering requirements. | Check the data source, view structure, and eligible input fields. |
| Many marks are Not Clustered | Some clustering inputs contain null values or unsupported mixed categorical values. | Inspect missing values and check the selected variables. |
| Clusters are difficult to explain | Variables are irrelevant, redundant, or poorly chosen. | Review the business objective and test a smaller set of useful variables. |
| One variable dominates the interpretation | The variable has extreme values or an unusual distribution. | Inspect outliers and understand Tableau's scaling behavior. |
| Too many clusters are created | The chosen cluster count may be too detailed for the business question. | Compare smaller cluster counts and review their usefulness. |
| Clusters change after editing | Variables, data, level of detail, or cluster count have changed. | Document the model settings and compare the updated results. |
| Saved group does not update | The saved group is separate from the original clustering model. | Review the cluster results and update the saved group when needed. |
Clustering in Tableau groups marks according to similarities in selected variables. It is an unsupervised learning technique used to discover patterns in data.
Segmentation is the process of dividing a population or dataset into meaningful groups. The groups can be defined using business rules or statistical techniques such as clustering.
Tableau uses the K-means clustering algorithm for its built-in clustering feature.
A centroid is the center of a cluster, calculated as the mean of the observations assigned to it for the clustering variables.
Create a suitable visualization, open the Analytics pane, drag Cluster into the view, select variables, and configure the number of clusters.
Normalization places variables on comparable scales so that differences in numerical magnitude are less likely to dominate distance calculations.
You can specify a number or allow Tableau to select one using its cluster-quality criterion. Compare the results for interpretability and business usefulness.
Clustering uses an algorithm to discover groups based on similarities. Grouping usually combines selected dimension members manually or according to defined rules.
Yes. You can save cluster results as a group dimension and use that group in other worksheets, subject to Tableau's grouping constraints.
It generally indicates that Tableau could not assign a mark to a normal cluster, for example because a clustering input contains missing values or incompatible categorical information.
No. Cluster numbers are identifiers, not rankings. You must inspect the variables and characteristics of each cluster to interpret its meaning.
K-means can be sensitive to outliers and variable selection, and it works best when the chosen variables and resulting cluster structure are appropriate for the distance-based model.
Clustering and Segmentation help analysts discover meaningful groups within large datasets. Tableau's built-in K-means clustering feature makes it possible to identify patterns visually, examine cluster characteristics, and save results for reuse. By choosing relevant variables, checking data quality, comparing different cluster counts, and validating the results, analysts can build more informative dashboards and support better business decisions.
