Mapping emerging demand to localise a retailer's vegan range

QuantSpark's location-intelligence work for a leading UK grocery retailer shows what a data-led approach to an emerging category can achieve: a probabilistic demand model that started as a postcode-level heatmap ended up as the retailer's entire vegan ranging strategy, rolled out nationally across more than 750 convenience and supermarket stores from an estate of more than 1,000.

The problem was not that vegan food is a difficult category, it is that it is a young one. The retailer runs over a thousand convenience and supermarket stores and wanted to localise its vegan range at postcode level rather than apply the same assortment everywhere. That distinction matters because emerging categories rarely leave a clean sales history to plan from: shelf space allocated on gut feel or a national average either overstocks stores where genuine demand is low, tying up space and generating waste, or understocks stores in areas where demand is real but under-served, losing sales and disappointing shoppers who cannot find what they came in for. The retailer needed a way to anticipate demand store by store, ahead of the sales data catching up.

QuantSpark's answer combined demographic modelling with the retailer's own transaction history. The team began with a literature review of vegan population characteristics, then built a probabilistic demographic demand model using open-source location intelligence datasets alongside UK census data and postcode boundaries. Machine-learning-inspired methods estimated demographic density across the country and converted it into a heatmap of likely vegan population by postcode. In parallel, natural language processing was applied to the retailer's existing range assortment, apportioning the probability that each product belonged to a vegan basket. Those two outputs, the demand heatmap and the product-level vegan-basket probabilities, were then fed into a supply-demand gap model at store level, which produced concrete, store-specific ranging and space recommendations. The whole model was grounded in scale: the underlying analysis drew on more than 500 million product-level transactions spanning five years, so the recommendations were not a demographic guess sitting on top of the business, but a model tested against the retailer's own purchasing history.

The workflow runs in five stages. First, a data foundation combining the literature review, open-source location and census data, and five years of transaction history. Second, that foundation is converted into a demand heatmap through ML-inspired density estimation. Third, NLP scores the existing assortment for vegan-basket probability. Fourth, a store-level supply-demand gap model combines the heatmap and the assortment scores into specific ranging and space recommendations. Fifth, those recommendations become policy: adopted as the retailer's vegan ranging strategy and extended nationally.

None of the systems involved are named proprietary tools in the public record, but the categories are clear: open-source location intelligence and demographic datasets, UK census and postcode boundary data, probabilistic and machine-learning-inspired demand modelling, and natural language processing applied to product assortment. Together they replace a single national assumption about vegan demand with a store-specific one, built from public data rather than a costly bespoke survey.

The value delivered is best read as adoption at scale rather than a modelled financial return. The retailer did not pilot the model and shelve it: the location-intelligence module became the basis for its entire vegan ranging strategy and was rolled out across more than 750 convenience and supermarket stores, a strong signal that the store-specific recommendations were trusted enough to replace the blanket approach network-wide. The approach was also built to be reusable, extendable to other customer-profile demographics using the same open-source data sources, so the value is not confined to one category.

No financial uplift, waste-reduction or sales-lift figures are disclosed in the public source, so none are claimed here. In a comparable engagement, the commercial case for this kind of work is normally made by comparing category sales, availability or waste in stores that received localised ranging against a matched control group still running the blanket assortment, alongside the planning cost saved by automating what would otherwise be manual, category-by-category store reviews. Those comparative figures were not published for this project, so the rollout scale, more than 750 stores, stands as the headline measure of impact.