Organizational Network Analysis (ONA): finding the people your org chart misses

60-Second Summary

ONA maps the *real* informal network in an organisation — who collaborates with whom, who's central, who bridges otherwise-disconnected groups — using either surveys or passive collaboration data (email/Slack/Teams metadata).
Three patterns matter most: high-centrality nodes (your hidden critical people), brokers (people connecting otherwise-separate groups — losing them shatters the network), and isolates (under-connected employees, often high attrition risk).
Use ONA for: M&A integration diagnostics, identifying succession risks the org chart misses, post-reorg health checks, RTO-vs-remote collaboration audits, and finding the people who actually move work.
Passive ONA (metadata mining) is powerful but ethically loaded. Aggregate-only, opt-in where possible, never used for individual performance decisions, GDPR-compliant by design.
Tools: Microsoft Viva Insights / Workplace Analytics, Worklytics, OrgVitals, Polinode, Cognitive Talent Solutions, or a custom build on NetworkX/igraph.

The org chart says who reports to whom. ONA says who actually works with whom. The gap between the two is where the real organisation lives — and where most HR decisions go wrong. This guide covers what ONA is, when it's worth doing, and how to do it without creating a surveillance program.

What ONA actually is

Organizational Network Analysis applies graph theory to people data. Each employee is a node; each interaction (email, meeting, Slack DM, survey-reported collaboration) is an edge. The resulting graph reveals patterns the org chart cannot.

Origins: social network analysis (Moreno 1934, sociograms), formalised in the 1970s (Granovetter on weak ties), commercialised in HR in the 2010s. The 2020–2026 boom came from RTO debates and AI-augmented collaboration data — organisations suddenly had the data and the question 'is remote work breaking our network?'.

Passive vs survey-based ONA

Two approaches

Passive ONA

Data source: email/Slack/Teams metadata (who emails whom, how often) — never content
Always-on, no employee effort
Captures actual behaviour, not self-report
Higher ethical/legal load (GDPR Art. 22, works-council scrutiny)
Tools: Microsoft Viva Insights, Worklytics, Humu, custom

Survey-based ONA

Data source: short survey ('who do you collaborate with weekly?')
Quarterly or annual, requires participation
Captures perceived collaboration (often more useful than actual)
Lower ethical load — explicit consent each cycle
Tools: Polinode, OrgVitals, custom with SurveyMonkey + NetworkX

Start with surveys

For first-time ONA, do a survey-based pilot. Lower ethics overhead, easier exec buy-in, and the data is often more interesting (perceived networks reveal influence, not just message volume). Graduate to passive once you've earned trust and have a clear use case.

The five key metrics

Graph metrics that matter in HR

1
Degree centrality
How many connections a person has. High = popular, well-connected. Often correlates with influence but also with overload.
2
Betweenness centrality
How often a person sits on the shortest path between two others. High = a broker — they connect otherwise-separate groups. Losing them fragments the network.
3
Closeness centrality
How quickly a person can reach others in the network. High = they get information fast and can move it fast — often the informal go-to person.
4
Clustering coefficient
How tightly someone's neighbours are connected to each other. Low = the person's connections are diverse (cross-functional); high = they live in a clique.
5
Eigenvector centrality / PageRank
How influential someone is, weighted by how influential their connections are. Surfaces 'kingmakers' — people not flashy themselves but connected to the powerful.

The broker risk

In most companies, 1–3% of employees are critical brokers — losing them cascades. ONA is the only way to find them; the org chart and 9-box completely miss this risk. The single most actionable use of ONA is identifying brokers and putting succession + retention plans around them.

Use cases that justify the investment

Use case	What ONA reveals	Typical action
M&A integration	Whether the two companies are actually connecting 6/12/18mo post-close	Targeted integration events, identifying integration-broker roles
Succession risk	Hidden brokers and central nodes the formal chart misses	Retention + succession plans for the top 1–3% by betweenness centrality
Post-reorg health	Did the reorg actually change collaboration, or did old patterns persist?	Recalibrate org design, identify reorg-survivors who never moved
RTO vs remote audit	Whether remote teams are forming silos vs office teams	Targeted in-person events, structural fixes (not blanket RTO mandates)
Diversity & belonging	Whether under-represented groups are well-connected or isolated	Targeted sponsorship, mentor matching, structural fixes
Innovation networks	Which teams generate ideas that cross silos vs stay local	Resource the brokers, restructure the silos
Attrition risk (early signal)	Sudden drops in centrality often precede exits by 60–120 days	Targeted manager check-in (handled carefully — see ethics)

Ethics: the non-negotiables

Hard rules for ONA programs

Aggregate-only reporting (no individual scores shared with managers). Opt-in with clear notice (especially for passive collection). GDPR Art. 13/14 notices for EU employees. No content analysis (metadata only). No use in performance or hiring decisions. Works-council consultation in countries that require it (Germany, France, Netherlands). Periodic ethics review.

Document the lawful basis (GDPR Art. 6 — legitimate interest typically, with a documented balancing test).
Pseudonymise individual identifiers in the analyst workflow; only re-identify when an authorised action is approved.
Run a Data Protection Impact Assessment (DPIA) before deploying passive ONA in any EU country.
Publish what you measure and what you don't. Employee trust collapses when surveillance is discovered, not when it's disclosed.
Set a hard retention limit (typically 12–24 months) and delete on schedule.

Getting started: the 90-day pilot

Weeks 1–2: Define one specific use case (e.g. 'post-reorg health check for the engineering org'). Without a question, ONA generates pretty graphs and no decisions.
Weeks 3–4: Choose survey vs passive. For a first pilot, choose survey. Draft a 5-question instrument ('who do you go to weekly for advice / decisions / information?').
Weeks 5–6: Run the survey on a single BU (50–500 people). Aim for 70%+ response rate. Lower than that and the graph is biased.
Weeks 7–10: Analyse — visualise the graph (Gephi or Polinode), compute the five metrics, identify brokers and isolates. Triangulate with HR data (tenure, performance, attrition).
Weeks 11–12: Share findings *aggregate-only* with the BU leader. Propose specific actions (succession plans for brokers, integration interventions for isolated subgroups). Decide whether to expand or stop.

FAQ

Frequently asked questions

Is ONA legal in the EU?

Yes, with proper basis. Survey-based with explicit consent is straightforward. Passive (metadata) requires DPIA, legitimate-interest balancing test, works-council consultation where applicable, and aggregate-only reporting. Don't deploy passive in EU without legal review.

Will employees revolt if they find out?

Yes, if you're hiding it. No, if you're transparent. The companies that have done this well (Microsoft, Vodafone) publish the program, the safeguards, and the use cases. Trust is earned by disclosure, not erased by it.

How does ONA compare to engagement surveys?

Different questions. Engagement says how people feel; ONA says how they connect. Use both. The pairing is powerful — e.g. 'isolated employees with low engagement are 4x more likely to leave in next 6 months'.