A retailer’s product assortment frequently contains sets of products that if purchased in combination with each other, make up a certain latent ‘project’. For example, the purchase of mold proof paint, bathroom tiles and a sledgehammer indicates that a customer will renovate his bathroom. Often retailers don’t consider the existence of projects in their assortment, or they rely on existing product hierarchies to combine products into projects. Such a strong and unique match between products and projects, however, will not always hold. We propose an innovative model-based approach that uses purchase history data to find and identify latent projects. By adopting this approach, a project can be characterised by the products that are likely to be purchased given that a customer is working on that project. Our proposed model-based approach allows all kinds of information to affect the likelihood that a customer is working on a specific project. For example a customer’s previous purchases and demographics. In addition, we allow for persistency which is a natural feature in this context, e.g. at some point in time the customer will finish renovating his bathroom. In the end, our approach enables us to infer the probability with which a customer is working on a specific project, which in turn facilitates project-specific targeting for marketing actions at the customer-level.