Why More Isn't Always Better: What Business Networks Really Mean for Innovation

Using data on thousands of Dutch firms across various industries (e.g., manufacturing, scientific activities, construction etc.) and sizes (from 10 to more than 250 employees), we examined how the density of connections among a firm's extended network, not just direct partners, but partners' partners and beyond, affects the likelihood of introducing a new product. What emerged was a pattern that defies simple intuition.

At low levels of density (see Figure 2), firms lack the interconnections needed for knowledge to flow effectively. Ideas get stuck. There aren't enough pathways for useful information to travel from one part of the network to another. At moderate levels, something productive happens: knowledge circulates, trust builds between firms that can vouch for each other through mutual connections, and firms tap into the complementary expertise of others. This is the sweet spot: the point where the village is connected enough to share knowledge, but not so connected that nothing stays private.

But push density too high, and the network starts working against you. When everyone is connected to everyone, proprietary knowledge leaks. The unique insight you gained from a partner — a new manufacturing technique, a novel approach to a market problem — quickly becomes common currency across the network. Your competitive edge dissolves. What was once a valuable piece of knowledge becomes freely available to firms you may be competing with, even if you never shared it with them directly.

There's a twist. This pattern depends heavily on how different the firms in a network are from one another — what we call cognitive distance. Cognitive distance captures the extent to which firms differ in their knowledge bases, their technological expertise, and their organisational focus. It's the gap between, say, a biotech start-up and a logistics firm, or between a software company and a traditional manufacturer.

If your network partners operate in very different fields and think about problems in fundamentally different ways, the potential for breakthrough innovation is high. You're exposed to ideas you'd never encounter within your own sector — the kind of unexpected combinations that lead to genuinely new products. But that same diversity raises the stakes. In a tightly connected network, those valuable, unfamiliar ideas spread to competitors just as fast as they reach you. The more novel the knowledge, the more damaging its leakage.

When firms in a network are more similar — sharing the same industry background, similar expertise — the risks of a dense network are lower. There's less unique knowledge to leak, and the trust-building benefits of tight connections tend to prevail. In these more homogeneous networks, density can actually become an accelerator rather than a liability.

In short: The same level of density that helps one firm innovate can hinder another, depending on how different its neighbours are.

How do you even map a network like this? Traditionally, researchers have relied on patent records, alliance databases, or survey data. These sources are useful but limited: patents only capture a slice of innovative activity, alliance databases miss informal partnerships, and surveys are expensive, slow, and confined to the firms that choose to respond.

We took a different approach. We scraped the websites of thousands of Dutch firms and traced the hyperlinks between them. A link from one company's website to another is a small but meaningful signal: it suggests a relationship, whether a formal partnership, a supplier connection, or a knowledge-sharing tie. Individually, each link tells you little. But taken together, across thousands of firms, these digital traces reveal the hidden structure of the Dutch business landscape — clusters of interconnected firms that emerge organically, cutting across traditional sector and geographic boundaries.

We also used natural language processing on the text of corporate websites to measure cognitive distance. By analysing what firms write about themselves — their products, their expertise, their markets — we could quantify how similar or different any two firms are in their knowledge and organisational focus. This gave us a much broader picture of interfirm differences than patents alone can provide, capturing not just technological distance but also differences in market orientation and organisational identity.

This web-based data was then combined with the Community Innovation Survey (CIS) conducted by Statistics Netherlands (CBS), which provided reliable, validated measures of whether firms had actually introduced new products to the market. The result was a unique hybrid dataset: one that pairs the structured, representative innovation data of an official survey with the rich, boundary-spanning network information hidden in the web.

This approach speaks to a broader ambition. As Jonas Klingwort, member of this research team and methodologist and data scientist at Centraal Bureau voor de Statistiek (CBS), puts it: 

"Advancing official statistics requires not only methodological innovation but also the systematic exploration of new and alternative data sources to complement and enhance traditional data collection." 

Web-based methods offer exactly that — a scalable, timely, and boundary-spanning alternative to traditional data sources. Members of this research team are actively pushing this frontier further, using data from platforms such as Reddit and Glassdoor to explore factors affecting entrepreneurship and start-up success, demonstrating that the web holds a wealth of untapped insight for understanding how businesses form, connect, and innovate.