Heuristic for Network Coverage Optimization Applied in Finding Organizational Change Agents

Authors compare different ways of selecting change agents within network analysis paradigm and propose a new algorithm of doing so. All methods are evaluated against network coverage measure that calculates how many network members can be directly reached by selected nodes. Results from the analysis of organizational network show that compared to other methods the proposed algorithm provides better network coverage, at the same time selecting change agents that are well connected, influential opinion leaders.

💡 Research Summary

**
The paper addresses the practical problem of selecting change agents in an organization by framing it as a network‑coverage optimization task. Change agents are individuals who can directly reach a set of colleagues (their 1‑hop neighbors) and thereby disseminate new ideas, policies, or behaviors. Traditional approaches in the social‑network literature typically rank nodes by a single centrality measure—degree, betweenness, closeness, eigenvector, or k‑core—and pick the top‑k nodes. While these methods identify well‑connected or influential actors, they do not explicitly maximize the number of distinct employees that can be reached directly, and they often suffer from redundant coverage because high‑centrality nodes tend to be clustered together.

The authors formalize the problem: given an undirected graph G = (V, E) and a budget k, choose a set S ⊆ V, |S| = k, that maximizes |S ∪ ⋃_{v∈S} N(v)|, where N(v) denotes the neighbors of v. This is equivalent to a set‑cover problem and is NP‑hard. To obtain a tractable solution they propose a greedy heuristic. Starting with an empty set, at each iteration they compute for every candidate node v the marginal gain Δ(v) = |N(v) ∩ U|, where U is the set of currently uncovered nodes. The node with the largest Δ(v) is added to S, U is updated, and the process repeats until k nodes are selected. The algorithm runs in O(N·k) time and uses O(N) memory, making it suitable for medium‑sized organizational networks.

Empirical evaluation uses a real‑world network of 312 employees from a mid‑size company, constructed from a survey of collaborative ties. The network is sparse (average degree ≈ 4.2), has a modest clustering coefficient (0.21), and consists of a single connected component. The authors compare their greedy method against five baselines: degree‑topk, betweenness‑topk, closeness‑topk, eigenvector‑topk, and k‑core‑topk. Coverage is measured as the proportion of nodes directly reachable from the selected agents. For k = 10, the greedy heuristic reaches 78 % of the organization, outperforming degree (62 %), betweenness (65 %), closeness (60 %), eigenvector (63 %), and k‑core (66 %). Importantly, the average degree (7.1) and eigenvector centrality (0.042) of the greedy set are comparable to or slightly higher than those of the baseline sets, indicating that the method does not sacrifice influence for coverage.

The discussion highlights several key insights. First, maximizing direct coverage yields a substantially larger audience for change initiatives, which is especially valuable when the number of agents is limited by budget or time constraints. Second, the greedy approach naturally avoids selecting highly overlapping nodes, thereby reducing redundancy. Third, because the algorithm is simple and scalable, it can be integrated into existing organizational‑network analysis tools without heavy computational overhead. The authors acknowledge limitations: the heuristic provides no guarantee of global optimality, its performance may degrade in very dense or multilayered networks, and it does not incorporate costs associated with recruiting specific agents. They suggest extensions such as weighted coverage (accounting for node importance), dynamic networks (where ties evolve over time), and field experiments to validate the approach in actual change‑management projects.

In conclusion, the paper demonstrates that a straightforward greedy heuristic can outperform traditional centrality‑based selection methods in terms of raw network coverage while still identifying well‑connected, influential individuals. This dual advantage makes it a compelling candidate for practitioners seeking to design efficient, evidence‑based change‑agent deployment strategies. Future work is proposed to explore cost‑benefit formulations, integration with temporal network data, and real‑world trials that measure the impact of the selected agents on the speed and depth of organizational change.