A few editions ago we looked at how AI can help coworking operators sell more, with three use cases from different stages of the customer lifecycle:

All three point at new revenue, either by acquiring new members or by extracting more from the inventory already there. This edition looks at a fourth, further along the lifecycle: cross-selling and personalisation, aimed at the members already in the building.

Coworking spaces have a lot of data on how their members use the space: how often a member enters the building each week, how long they stay, how many rooms and resources they book, which extra services they buy, what plan they are on, what they pay. Years of data, in most cases, used for little beyond an occupancy report.

So how can operators turn that data into custom experiences and personalised services?

AI and member data let operators move from a single experience built for every member to many tailored experiences built for one. Segmentation grounded in behaviour and preferences shapes offers around how each segment actually uses the space, driving engagement, retention and revenue.

But the value of segmentation only appears at the point of action. Segmentation that stays in a dashboard is a report: useful, but not actionable. Pushing those segments into campaigns is what changes the member experience.

A few questions this kind of segmentation opens up for an operator:

All of this is available in Nexudus through a feature called Marketing Insights. Here is how it works.

Marketing Insights is an AI feature in the Nexudus platform that looks at how members use a space, detects patterns in that usage, and groups members with similar behaviour into segments across a wide range of variables. What it returns is not a list of members sorted by one or two fields. It is a small set of behavioural profiles, each with a name, a description, and a set of recommended strategies (commercial, engagement, retention) to apply to that specific group.

The feature sits under Nexudus CRM > Marketing Insights. Click Get Recommendations and, within seconds, the system analyses member space usage, engagement levels and booking trends, and returns a set of behavioural segments.

Those segments show operators the main patterns of use in their space, and give each group a starting point for a custom experience or a cross-sell, grounded in how members actually behave.

The features (data variables) used to build the segments fall into four groups:

That combination is what separates this from plan-based segmentation, or from a group built on one or two variables. Two members on the same dedicated desk contract can land in completely different segments: one is in four days a week and never buys anything beyond the desk, the other comes in twice a week but books meeting rooms constantly and spends on catering every time.

The model works on usage patterns, not identities: no names, emails or contact details are fed into it.

This section is a bit technical, but it's also important to know how the segments are built. There are three steps (scaling, dimensionality reduction and clustering) and an operator does not need to run any of them, but should understand roughly what happens before acting on the result.

The first step involves standard scaling of all features. Scaling the data is crucial as it ensures consistency and prepares it for the dimensionality reduction step in the next stage.

The next step is to perform dimensionality reduction using UMAP (Uniform Manifold Approximation and Projection). UMAP is a powerful, non-linear dimensionality reduction technique designed to preserve the underlying structure of high-dimensional data. It works by constructing a weighted graph of the data's nearest neighbours and optimising a low-dimensional representation that maintains these relationships as closely as possible.

Unlike traditional methods like PCA (Principal Component Analysis), which rely on linear transformations, UMAP excels in capturing complex patterns, making it particularly effective for clustering, visualisation, and preprocessing for machine learning tasks. By reducing the dimensionality, UMAP simplifies the dataset while retaining its key features and enabling better performance.

After dimensionality reduction, the data is fed into the HDBSCAN (Hierarchical Density-Based Spatial Clustering of Applications with Noise) model. HDBSCAN is a clustering algorithm that extends the DBSCAN method by incorporating a hierarchical approach to detect clusters of varying densities. Unlike traditional clustering algorithms like k-means, HDBSCAN does not require the number of clusters to be specified in advance, and can automatically identify and separate noise (outliers) from meaningful clusters.

Each segment, along with all the variables that shape it, is then fed into an OpenAI LLM with a single goal: to build a readable profile with recommended strategies grounded in that data.

What the operator gets back for each segment is:

An example of the segment output, using fake data.

We said earlier that the value of segmentation only appears at the point of action. Knowing the main usage profiles in a space is useful in itself. It tells an operator how members actually behave, rather than how their contracts say they should. But the most relevant part is turning those segments into real campaigns that offer custom services or cross-sell opportunities to each group.

This is one example of how data and AI can work on both sides: a more personalised experience for members, and new revenue from cross-sell opportunities that were already there, just unread.

Start with a campaign for the segment where the cross-sell opportunity is clearest and the value to the member is real. Launch it and measure the conversion. Once the results build confidence, experiment with new campaigns. Both member segments and marketing campaigns are available in the Nexudus platform.

That's it for today. See you in two weeks.