Organizational Network Analysis: Mapping How Work Really Flows
Your org chart shows who reports to whom. It has almost nothing to say about who actually gets work done together. Organizational network analysis maps the second structure, and the gap between the two is usually where your delivery risk lives.
What Organizational Network Analysis Is
Organizational network analysis, usually shortened to ONA, treats an organisation as a network rather than a hierarchy. People are nodes; interactions are edges. Once work is represented that way, structural questions become answerable: who connects otherwise separate parts of the business, which teams have drifted into isolation, and which individuals would leave a hole if they resigned tomorrow.
The technique is decades old in academic sociology. What changed recently is the data. ONA once required an all-staff survey asking people to nominate collaborators, which was slow, expensive and heavily biased by memory and politics. Meeting, messaging and document co-editing metadata now describes the same network continuously and without asking anyone to recall anything.
That shift is what moved ONA from a consulting engagement into something an internal team can run each quarter.
Three Networks, Not One
A common early mistake is treating collaboration as a single undifferentiated network. In practice at least three overlap, and they answer different questions.
The workflow network describes who hands work to whom. It is the closest thing to a map of your actual operating model and it is where bottlenecks appear. The communication network describes who talks to whom, which is far denser and reflects social ties as much as work ties. The expertise network describes who is consulted for answers, and it is usually the most revealing, because being asked is a much stronger signal of influence than being copied in.
The three rarely coincide. The person at the centre of your expertise network frequently sits several levels below the person at the centre of your org chart, and finding that mismatch is often the single most valuable output of a first analysis.
The Roles a Network Reveals
Certain structural positions recur across almost every organisation, and they carry different risks.
Connectors have unusually many ties within their group. They are the people who make a team cohere, and they are also single points of failure for that team's morale. Brokers sit between groups that otherwise barely interact. They are disproportionately valuable and disproportionately overloaded, because every cross-functional request routes through them. Peripherals have few ties in any direction, which sometimes means deep focused work and sometimes means an onboarding failure nobody noticed.
None of these is a performance judgement. A broker is not better than a peripheral. They are structural facts about position, and they call for different management responses: succession cover for connectors, load relief for brokers, and a conversation for peripherals.
What ONA Is Genuinely Good At
Four use cases justify the effort more reliably than the rest.
Post-reorganisation verification is the strongest. A reorganisation is a hypothesis about how work should flow. Comparing the collaboration graph before and against three months after tells you whether the hypothesis held, and organisations are usually surprised by how often the old structure persists underneath the new boxes. Onboarding effectiveness is the most actionable: new joiners who have not formed ties beyond their immediate manager by week six are a well-established attrition risk. Silo detection finds teams whose cross-functional ties have thinned over time. Key-person risk identifies the brokers whose departure would sever connections rather than merely lose a headcount.
For the attrition-signal side of this, our guide to employee engagement metrics covers the complementary measures.
Reading the Graph Without Overreading It
Network diagrams are seductive, and it is easy to draw conclusions the underlying data does not support.
Interaction volume is not influence. A person copied into every thread will look central and may be entirely peripheral to decisions. Absence of a tie is not absence of a relationship, because two people who sit together and talk in person generate almost no metadata. And network position is heavily role-determined: an executive assistant will always look like a hub, which is a description of the job rather than a finding.
The defensible readings are comparative rather than absolute. How did this team's external tie count change after the reorganisation? How does this cohort's week-six connection count compare with the previous cohort's? Change over time is far more robust than any single snapshot.
Cross-Team Ties, Last 30 Days
Cross-team interactions per week
Tie distribution by role type
▲ Cross-team ties rose 19 points after the shared planning ritual was introduced in W2.
Illustrative eMonitor dashboard.
Privacy: The Non-Negotiables
ONA is among the more sensitive things you can do with workforce data, because a network graph is inherently relational. Even without message content, the edges themselves are revealing, and in several jurisdictions a collaboration graph is unambiguously personal data.
Three rules keep this defensible. Analyse metadata only, never content: who and when, not what. Report at group level with a minimum cohort size, commonly five, so no individual is identifiable from a cell. And separate the analytical view from the managerial one, so that individual-level graphs are available to a small analytics function under documented conditions and not to line managers browsing their teams.
Announce the analysis before running it. An ONA discovered rather than disclosed will damage trust more than any insight it produces is worth.
Running Your First Analysis
Scope it narrowly. A first ONA covering the whole company produces a hairball nobody can act on. Two or three interdependent functions is the right size.
Choose one question. Did the platform reorganisation actually reduce hand-offs between product and engineering? Are new joiners in the support organisation connected by week six? Pull ninety days of metadata, which is long enough to smooth holidays and short enough to reflect the current structure. Build the graph, identify brokers and peripherals, then, and this is the step most teams skip, go and talk to the people concerned before concluding anything.
The interviews are not a formality. A peripheral position that looks like an onboarding failure regularly turns out to be a specialist working exactly as intended, and no amount of graph analysis would have told you that.
Map the Organisation You Actually Have
eMonitor turns collaboration metadata into group-level network views, with minimum cohort sizes enforced by default.
Connecting ONA to the Rest of Your Data
ONA is at its weakest as a standalone exercise and at its strongest joined to workload and output data.
A broker with rising external tie counts and rising working hours is a burnout risk with a name and a date. A team with thinning cross-functional ties and lengthening cycle times has a coordination problem rather than a capacity problem, and hiring into it will not help. These joins are where the analysis stops being interesting and starts being useful.
Our guide to predictive analytics in workforce data covers the modelling side of these combinations.
Common Mistakes
Running ONA as a one-off is the most common. A single snapshot has no baseline, and almost every defensible ONA finding is a comparison. Quarterly cadence turns it from a curiosity into an instrument.
Presenting individual network positions to managers is the most damaging. It converts a structural analysis into a performance ranking, and it will be the last ONA you are permitted to run.
Confusing the communication network with the workflow network is the most technically consequential, because it leads to reorganising around social ties rather than work dependencies.