This recipe executes an organizational network analysis (ONA) from communication metadata, producing a directed weighted graph with centrality scores, bottleneck identification, structural hole analysis, and a formal-vs-informal gap report. The output reveals who actually drives information flow versus who the org chart says should. [src1, src2]
Which tool path?
├── Enterprise client with Microsoft 365
│ └── PATH A: Viva Insights — automated ONA
├── Technical analyst + org size < 1000
│ └── PATH B: Python NetworkX + Gephi — full control
├── Technical analyst + org size > 1000
│ └── PATH C: Python igraph + D3.js — scalable
└── Non-technical, no Viva Insights
└── PATH D: Gephi desktop — visual-first| Path | Tools | Cost | Speed | Output Quality |
|---|---|---|---|---|
| A: Viva Insights | Microsoft Viva Insights | $5K-$10K | 1-2 days | Excellent |
| B: NetworkX + Gephi | Python, NetworkX, Gephi | $0 | 3-4 days | Excellent |
| C: igraph + D3.js | Python, igraph, D3.js | $0 | 3-5 days | Excellent |
| D: Gephi desktop | Gephi, spreadsheets | $0 | 4-5 days | Good |
Duration: 1-2 hours · Tool: Decision based on client environment
Choose tool path based on Microsoft 365 license, analyst skill, org size, and visualization requirements.
Verify: All dependencies installed. · If failed: If pip fails, use conda.
Duration: 4-8 hours · Tool: Python NetworkX or igraph
Build directed weighted graph from communication metadata. Nodes = anonymized employee IDs, edges = communication frequency weighted by interaction type. [src2]
# Weight by interaction type: direct_message=3.0, thread_reply=2.0,
# channel_message=1.0, reaction=0.5, email_direct=3.0, email_cc=0.5
G = nx.DiGraph()
# Aggregate weighted edges from communication metadataVerify: Node count matches roster (+/- 5%). · If failed: Check data collection coverage.
Duration: 4-8 hours · Tool: Python NetworkX
Calculate betweenness centrality (bottlenecks), degree centrality (hubs), eigenvector centrality (influence). Identify structural holes using Burt's constraint measure. Flag statistical outliers (> 2 SD). [src1, src3]
Verify: Top 10 bottlenecks identified, centrality distributions follow power-law shape. · If failed: Re-check weight normalization.
Duration: 4-8 hours · Tool: Python + org chart data
Overlay formal org chart onto actual communication graph. Identify shadow influencers (high eigenvector, no formal authority) and bypassed managers (formal authority, low communication centrality). [src1]
Verify: Gap report generated with shadow influencer and bypassed manager lists. · If failed: Document gaps in org chart data.
Duration: 4-8 hours · Tool: Gephi, D3.js, or matplotlib
Generate annotated network visualization. Produce written analysis with findings and recommendations. [src1, src3]
Verify: Client validates top 5 bottlenecks and shadow influencers. · If failed: Re-run with different edge weight calibration.
{
"output_type": "ona_analysis_report",
"format": "PDF + CSV + SVG",
"key_metrics": [
{"name": "bottleneck_count", "description": "Nodes > 2 SD above mean betweenness"},
{"name": "shadow_influencer_count", "description": "High eigenvector, no formal authority"},
{"name": "bypassed_manager_count", "description": "Formal authority, low centrality"},
{"name": "structural_hole_count", "description": "Nodes spanning disconnected clusters"}
]
}| Quality Metric | Minimum Acceptable | Good | Excellent |
|---|---|---|---|
| Node coverage (% of roster) | > 80% | > 90% | > 98% |
| Bottleneck identification (client-validated) | > 50% confirmed | > 70% | > 85% |
| Visualization readability | Clusters visible | Annotated clusters | Interactive exploration |
| Gap analysis coverage | > 60% | > 80% | > 95% |
If below minimum: Extend data collection, add missing platforms, or use interview-based network mapping.
| Error | Likely Cause | Recovery Action |
|---|---|---|
| NetworkX convergence failure | Graph too sparse or disconnected | Use eigenvector_centrality_numpy or fall back to PageRank |
| Uniform centrality scores | Edge weights not normalized | Re-check weight calibration, verify interaction type distribution |
| Too many bottlenecks (> 20%) | Threshold too low | Increase to 2.5 or 3 standard deviations |
| Disconnected graph | Missing department data | Analyze largest connected component, document gaps |
| Visualization unreadable | Too many overlapping nodes | Use community detection to cluster |
| Component | Free (NetworkX) | Mid (Gephi Pro) | Enterprise (Viva) |
|---|---|---|---|
| Graph analysis tool | $0 | $0 | $5K-$10K |
| Visualization | $0 (matplotlib) | $0 (Gephi) | Included |
| Developer time | 3-5 days | 4-5 days | 1-2 days |
| Total | $0 | $0 | $5K-$10K |
Same edge weight for broadcasts and 1:1 messages. Result: channel spammers appear as influential hubs. [src2]
1:1 messages 3x weight, thread replies 2x, reactions 0.5x. This reflects actual relationship strength.
Focusing only on bottleneck centrality. Result: missing the people who bridge disconnected groups. [src3]
Calculate Burt's constraint measure. Low-constraint nodes span structural holes — the organization's innovation bridges.
Showing raw network graph without first validating findings. Result: false positives undermine credibility. [src1]
Share top 10 findings with 2-3 internal stakeholders before formal presentation.
Use when an agent needs to execute organizational network analysis from communication metadata. This is Step 3 of the OIA engagement lifecycle. Requires the combined communication metadata dataset from the data collection recipe. Output feeds into autoimmune scan and stress test recipes.