Four Business Model Archetypes Emerge Among Flemish Dairy Farms

A new analysis of Belgium’s 2020 agricultural census has mapped nearly all Flemish dairy farms onto four distinct business model archetypes, and found that most of them do not fit neatly into any single one. Using a statistical technique called archetypal analysis, researchers classified 3,810 Flemish dairy farms into four profiles: Grassland Stewards, Integrated Feed Landholders, Entrepreneurial Diversifiers and Scale Maximisers. The striking result is that only a minority of farms align strongly with one archetype. Applying a threshold at which a farm’s dominant membership coefficient must exceed 0.67 to count as “pure,” just 19 percent of Grassland Steward members and 23 percent of Integrated Feed Landholder members were pure, compared with 31 percent for Entrepreneurial Diversifiers and 40 percent for Scale Maximisers. Overall, 74 percent of farms were classified as hybrid, combining features of multiple configurations.

The study, published in the journal Agricultural and Food Economics, was motivated by a long-standing problem in agricultural policy: farm typologies used for monitoring, benchmarking and designing targeted interventions typically force every farm into one mutually exclusive category. The authors argue that such hard classifications obscure the mixed configurations and boundary cases that dominate real farm populations. Their alternative treats heterogeneity as configurations of strategic choices rather than positions along a single structural gradient, drawing on the business model literature, which frames firms through the linked dimensions of value creation, value delivery and value capture.

To build the typology, the team used official census data from Statbel, Belgium’s national statistical office, which covers the full population of agricultural holdings rather than a survey sample. Starting from 4,991 Flemish holdings reporting dairy cows in 2020, they excluded farms with fewer than 15 cows, a threshold corresponding to the first local minimum in the smoothed herd-size distribution, leaving 3,832 commercial holdings. A further 22 holdings were dropped because missing or inconsistent denominator values made key land and tenure variables impossible to construct, producing a final analytical sample of 3,810 farms, or 99.4 percent of the commercial sample. The researchers note that a census base matters because sample surveys such as the Farm Accountancy Data Network may under-represent less common configurations, which are precisely the cases a mixed-membership approach is designed to capture.

The indicators feeding the model were constrained by what the census records. Value creation was operationalised through fodder-based composition, including shares of grassland, maize and other fodder crops, a Shannon diversity index of the fodder mix, the number of income-generating activities, and labour per hectare. Value delivery was captured by two ordinal indicators: the importance of direct sales and the share of income from non-agricultural outputs. Value capture, which in the broader business model literature includes bargaining position and revenue logic, could only be proxied, because detailed prices, margins and contract conditions are absent from census data. The authors instead used standard output as a measure of economic size, alongside the split between family and hired labour and the share of leased land. Herd size and stocking rate were excluded due to multicollinearity with structural indicators, and organic status was set aside because organic prevalence was too low in the sample to yield a meaningful contrast.

Archetypal analysis, originally developed by Cutler and Breiman, identifies a small set of extreme reference points within the convex hull of the data and expresses every observation as a weighted mixture of them. The team estimated models for two through eight archetypes with multiple random starts. Although fit improved with each additional archetype, solutions beyond four mainly subdivided existing gradients rather than adding distinct business model logics, so the four-archetype solution was retained on the combined grounds of fit, stability and interpretability. The resulting profiles differ jointly in fodder base, labour organisation, land tenure and market orientation, not scale alone.

The four archetypes span a spectrum from grass-based family stewardship to capital-intensive scale maximisation. Grassland Stewards are small, family-operated farms on predominantly leased grassland, about 57 percent of utilised agricultural area, with low labour intensity of 0.04 AWU per hectare, modest standard output and no recorded diversification. Integrated Feed Landholders are medium-scale family farms combining grass, roughly 48 percent of area, with maize at about 46 percent and other fodder crops around 5 percent, showing a high fodder-mix diversity index of 0.84, relying almost entirely on family labour and operating on a more ownership-oriented land base. Entrepreneurial Diversifiers stand out through multiple income-generating activities, deriving 11 to 50 percent of output from non-milk products and with direct sales in the 50 to under 100 percent category, supported by higher labour intensity of 0.060 AWU per hectare and a mixed family and hired workforce. Scale Maximisers mark the intensive pole, with the highest standard output per farm at around EUR 1.44 million, the lowest labour per hectare, a maize-oriented forage base of roughly 50 percent maize and 29 percent grass, and labour organised entirely as hired work at the archetype level.

The distribution of farms and resources across these profiles reveals a dual sector structure. Grassland Stewards and Integrated Feed Landholders together account for the majority of holdings and the workforce, but a smaller share of cows and milk than their numbers suggest. Scale Maximisers comprise fewer than one-fifth of farms yet hold about one-third of cows and milk, indicating larger herds and markedly higher output per farm and per unit of labour. Entrepreneurial Diversifiers form a small segment using marginally more labour relative to their land and output, consistent with a labour-intensive diversification model. Geographically, West and East Flanders together host about two-thirds of holdings and are dominated by the family-farm continuum, with Grassland Stewards particularly prominent in West Flanders. Antwerp and Limburg show higher relative shares of Integrated Feed Landholders and especially Scale Maximisers, while Flemish Brabant, despite its small farm base, has the highest relative presence of Entrepreneurial Diversifiers.

The hybridity itself turned out to be structured rather than random. Normalised Shannon entropy of farm memberships averaged 63 percent with a median of 66 percent, implying that unambiguous single-archetype memberships are relatively rare. Secondary memberships show a close Grassland Steward to Integrated Feed Landholder continuum: 61 percent of Grassland Steward farms list Integrated Feed Landholder as their second archetype, and 62 percent of Integrated Feed Landholder farms list Grassland Steward. Entrepreneurial Diversifiers most often blend with the family-farm profiles, leaning 36 percent towards Grassland Stewards and 39 percent towards Integrated Feed Landholders, with only 26 percent blending towards Scale Maximisers. Scale Maximiser farms mainly hybridise with Grassland Stewards at 51 percent, attributed to leased-land strategies, and with Integrated Feed Landholders at 29 percent, primarily because of the feed base. Hybridity therefore concentrates along the family-farm axis, with diversification and scale partly distinct but still connected to that core rather than to each other.

The authors subjected the solution to extensive reliability checks. Alternative preprocessing pipelines, including different scaling and transformation choices, produced nearly identical residual sums of squares, with changes of at most 0.4 percent, and most reassignments of farm labels concentrated near archetype boundaries. Across 10 to 100 random initialisations, the best-fit residual for the four-archetype solution varied only slightly, and a non-parametric bootstrap of 100 resamples yielded archetype coordinates close to the full-sample estimates. In an 80/20 hold-out test, projecting held-out farms into the full-sample simplex produced high agreement, with an Adjusted Rand Index of 0.94 and 97.6 percent identical dominant labels. Changing the indicator set itself had a larger effect, with Adjusted Rand Index values ranging from 0.246 to 0.677, though the core profile logic remained interpretable. Four semi-structured interviews with experts familiar with the Flemish dairy sector supported the realism of the profiles, describing Integrated Feed Landholders as the standard Flemish model and Scale Maximisers as high-performing yet vulnerable to permitting regimes and societal acceptance of intensification.

The researchers are careful about the limits of what the typology can show. It rests on a single cross-section from 2020, a year in which the COVID-19 pandemic may have temporarily inflated direct-sales activity, and it cannot speak to trajectories, profitability, bargaining power or environmental and social outcomes. The archetypes should not be read as a sustainability ranking, and the authors caution that farm configurations are shaped by land quality, regulation, family labour availability and historical paths, so the language of business model choice should not imply that farms can freely move between configurations. Still, they argue, the configuration map offers a practical starting point: continuous membership coefficients can inform benchmarking and differentiated instrument design by showing where diversity is concentrated, how boundary configurations are structured, and which operating logics face distinct organisational and market conditions, from tenure-linked advisory packages for the family-farm core to environmental sequencing questions for the scale-intensive pole centred on Antwerp.

Subject of Research: A census-based archetypal analysis identifying four business model configurations and widespread hybrid membership among Flemish dairy farms.

Article Title: Mapping farm business model configurations with archetypal analysis: evidence from Flemish dairy farming

Article References: Kasaw, M. T., Contor, H., De Keyser, E., Bourgeois, L., Verplancke, F., Van Meensel, J., & Mathijs, E. (2026). Mapping farm business model configurations with archetypal analysis: evidence from Flemish dairy farming. Agricultural and Food Economics, 14(1), Article 69. https://doi.org/10.1186/s40100-026-00514-6

Image Credits: AI Generated

DOI: 10.1186/s40100-026-00514-6

Keywords: Flemish dairy farming, archetypal analysis, business models, farm typology, mixed membership, agricultural census, value creation, value capture, farm diversification, scale intensification, Belgium, agricultural economics