Your organization as a graph

We think of companies as hierarchies: people form teams, teams sit in departments, and departments belong to divisions. But does this mental model help us understand how work actually gets done?

Not really.
We create cross-functional groups and work streams around projects to foster collaboration. In extreme cases, we even go through reorgs and layoffs to help improve the structure. But it's still difficult to quantify how much all of these initiatives help because we lack a way to visualize how work flows through the organization.

What if we could map who is involved in a project, which teams frequently interact, and where the bottlenecks are? What if we could identify pockets where knowledge accumulates, as well as the areas that are left in the dark? What if we could leverage the people who act as bridges and support those who unintentionally create isolation?

With such insights, we can enhance our organizational design by bringing teams that need to collaborate closer together while removing the barriers that hamper their efficiency. We could even take it a step further and "test" organizational changes before implementing them by simulating how information would flow in the new structure.

In this talk we'll explore how visualizing your organization as a graph can uncover hidden structures, reveal communication pathways, and help you build better products.

Transcript

1. Dunya Kirkali Your organization as a Graph The hidden network of work behind every org chart
2. Org Chart
3. Team Chart
4. Real Chart
5. 🌍 • Team Leader at • Co-author of the • Engineering Manager’s Compass • Blogger on • blog.incrementalforgetting.tech • Graphs ❤ Dünya Kırkalı
6. 🤔 Why
7. Why graphs Isn’t everything a graph? • Relational Data • Relationships are as important as entities • Combine datasets • It let’s you ask interesting questions
8. Why now Just because Dünya likes graphs? • Layo ff s • Reorgs • Hiring freezes • E ffi ciency programs • Platform consolidation • AI-era productivity pressure
9. Gut feeling 🫀
10. Can we 🧠
11. ⭐ Northstar Systems
12. ⭐ Northstar Systems • CEO • 1 Chief Product O ffi cer • 1 Chief Technology O ffi cer • 4 teams • Core Work fl ows • Growth & Mobile • Platform Experience • Executive Leadership
13. ⭐ Northstar Systems • 4 services • rider-service • regions-service • car-service • payments-service
15. Boring 💤
16. 🪜 Augmentations
17. Communication
18. Communication • Chat • Slack • Teams • E-mail • Survey • Meetings Input
19. Communication w/o
20. Communication w/
21. Communication w/
22. Communication • Some people act as bridges • Coupling between areas • Too little communication Takeaways
23. “The org chart shows accountability. Communication shows execution” Communication Quote
24. Ownership
25. Ownership Input • Services • Backstage by Spotify • SERVICEOWNERS by GitHub • In-House • CODEOWNERS • GitHub or GitLab
26. Ownership w/o
27. Ownership w
28. Ownership • Some boundaries are unclear • Not every team has balanced amount of ownables • Some teams depend heavily on others Takeaways
29. “Things that change together should be closer together” Quote Ownership
30. Pull Requests
31. Pull Requests Input • GitHub • GitLab • Bitbucket • Etc.
32. Pull Requests w/o
33. Pull Requests w/
34. Pull Requests Takeaways • Incorrect ownership • Technical bridges • Domain kings
35. “The codebase remembers how the organization actually works” Quote Pull Requests
36. Other
37. Other • Trust • “Who do you go to when you need help” • “Who do you think I should talk to” • “Who would hurt the most if they leave” • Documentation • Notion • Con fl uence • Work Items • Jira • Linear
38. 🧮 Techniques
39. Leiden Community Detection • Finds densely connected clusters • Allows you to take a step back • Breaking down monoliths • Designing teams
40. Leiden Community Detection
41. Betweenness • Finds nodes that sit on many shortest paths between other nodes • Identi fi es bridges • Find bottlenecks • Coordination chokepoints
42. Betweenness • Lena Martinez • Marcus Chen • Fatima Ibrahim • Nina Okafor • Matteo Rossi • Jules Bennett • Zara Ali 35.48 35.01 28.40 27.79 25.30 23.20 20.81
43. Betweenness
44. PageRank • Rank in fl uence based on incoming connections • Identify the most in fl uential people, services, or teams in the graph • Highlight points that require more care
45. PageRank • regions-service • Lena Martinez • car-service • payments-service • Elena Petrova • Growth & Mobile • Marcus Chen • rider-service 1.205 0.660 0.658 0.646 0.615 0.598 0.590 0.584
46. PageRank
47. LLMs What would happen if we loose Developer X?
48. LLMs Which team is becoming the organization bottleneck?
49. LLMs Who should work together on a new strategic project?
50. 🏭 Pipeline
51. Pipeline 1. Make your hypothesis
52. Pipeline 1. Make your hypothesis 2. Collect data
53. Pipeline 1. Make your hypothesis 2. Collect data • Script • Run daily • Save as CSV
54. Pipeline 1. Make your hypothesis 2. Collect data • Ingestion pipelines • Data Lake
55. Pipeline 1. Make your hypothesis 2. Collect data 3. Convert them into Cypher queries and import into Neo4j
56. Pipeline 1. Make your hypothesis 2. Collect data 3. Convert them into Cypher queries and import into Neo4j • Each dataset in it’s own DB • Then create a combined DB where you associate all data • Install the Neo4J CLI
57. Pipeline 1. Make your hypothesis 2. Collect data 3. Convert them into Cypher queries and import into Neo4j 4. Use your favourite agent harness
58. Pipeline 1. Make your hypothesis 2. Collect data 3. Convert them into Cypher queries and import into Neo4j 4. Use your favourite agent harness 5. Correlate with other data sources
59. Pipeline 1. Make your hypothesis 2. Collect data 3. Convert them into Cypher queries and import into Neo4j 4. Use your favourite agent harness 5. Correlate with other data sources 6. Experiment
60. Pipeline 1. Make your hypothesis 2. Collect data 3. Convert them into Cypher queries and import into Neo4j 4. Use your favourite agent harness 5. Correlate with other data sources 6. Experiment 7. Repeat
61. 📓 Closing
62. Considerations • Ethics • Con fi dentiality • Anonymity • Expectations • It’s not a 💊, it’s a 🔍 • You always need to interpret the data • Leverage LLMs to pose your questions
63. Acknowledgements • 📖 Hidden power of social networks
 by Andrew Parker and Rob Cross • 📖 Linked
 by Albert-László Barabási • 🎤 Engineering Leadership through a Social Network Lens
 by Gabriel Ramirez • 🫶 Support
 by Maxim Schepelin
64. Resources • SERVICEOWNERS by GitHub • CodeScene • Backstage by Spotify • Understanding the Leiden Algorithm • Pi Coding Agent • Neo4j
65. @dunyakirkali Questions
66. https://schepelin.com/go/book