Collaboration diagnosis: diagnose your team's collaboration with data｜Nova

Nova is a project collaboration diagnostics tool by Qinghuijia for teams of 5–50. You enter metadata such as how often each pair of members talks, and Nova draws the collaboration network, marking bottlenecks, isolated members and weak links. Message content is never read.

Core capabilities: a collaboration health score, detection of bottlenecks, isolated members and weak links, and recommendations tied to each finding. Start with a free diagnosis.

NovaQinghuijia
Back to resources
Product

After the report lands: from metrics to action

Health score, centralisation index, weak ties — these are not a scorecard, they are a way to order your to-do list. How to translate each of the three structural problems into one executable action.

Qinghuijia ResearchAug 21, 20265 min read
report readingaction items

Let's correct a common mistake first: treating the health score as a performance KPI.

It is not. The health score is a high-level summary figure designed to help you decide "do we need to look into things closely this time?" and "did our recent changes make any difference?". Its absolute value has limited meaning — the reasonable range for a health score naturally varies across teams of different sizes and business tempos. There is only one meaningful way to use it: comparing it against the same team's own historical values.

Below is the recommended sequence for translating the report into action items.

Step 1: Verify data credibility first

Before looking at the conclusions, look back at your data. Check three self-check items:

  • Is the time window reasonable? (Ideally 3 to 6 weeks, avoiding obvious public holidays and release peaks)
  • Are the channels properly categorised? (Group chats, emails, and meetings should not be compressed into a single channel)
  • Are there any anomalous edges? (e.g., the communication frequency on a single edge exceeds the entire team's total volume)

If the data is flawed, you can skip the conclusions — they are merely calculating incorrect raw data faithfully.

Step 2: Focus on structures, not individuals

The report will include individual-level rankings (such as who has the most connections or who is isolated). The easiest mistake to make is jumping straight to this level and reading it as a personnel evaluation.

The correct sequence is to read team-level structural conclusions first:

  • Density: Is the overall connectivity too low or too high?
  • Centralisation: Is connectivity excessively concentrated around a few individuals?
  • Number of isolated nodes: How many members have almost zero active ties?
  • Number of cross-group edges: Are the channels between different sub-teams sufficient?

These four metrics define the nature of your structural issues and dictate the direction of your next steps.

Step 3: Map three categories of problems to three actions

Structural issues generally fall into three categories, each pointing to a distinct action direction:

1. Bottlenecks (Excessively Concentrated Connectivity)

  • Symptom: The in-degree of a few individuals far exceeds the average; everything must pass through them.
  • Action: De-centralise and delegate, do not just reduce their workload. Simply reducing their workload leaves the structural bottleneck intact. De-centralisation means delegating decision-making authority so that certain classes of tasks do not need their involvement at all. A common approach is to clearly define authorization boundaries for specific types of decisions.

2. Information Silos (Isolated Nodes and Structural Holes)

  • Symptom: Certain members have almost no active connections, or there are no bridging ties between distinct clusters.
  • Action: Add ties. Establish direct edges, tie connections to shared artifacts, or implement cross-role pairings. The core lesson is to never expect an edge to persist purely out of goodwill; it must be backed by a workflow.

3. Missing Weak Ties

  • Symptom: The number of cross-group channels is small or decreasing.
  • Action: Establish routine channels. Remember that weak ties are characterised by low frequency, so do not try to build them using high-frequency mechanisms. A weekly meeting often becomes hard to sustain, whereas a persistent shared artifact is much more reliable.

Step 4: Prioritise action items

When all three categories of problems coexist, the recommended priority is:

  1. Fix data credibility (if any issues were flagged in Step 1)
  2. Resolve bottlenecks first — bottlenecks distort and contaminate all other structural metrics around them
  3. Resolve information silos next
  4. Cultivate weak ties last

The reason to tackle bottlenecks first is their overwhelming systemic impact: when a bottleneck exists, the connection profiles of everyone around it are warped. Trying to fix other areas first will likely lead to wasted effort.

Step 5: Run a follow-up diagnosis in one month

To see if your changes were effective, look at two metrics in the follow-up report:

  • Have the structural metrics moved in the desired direction? (Density, centralisation, cross-group edges)
  • Do the new edges persist? (Do they show up in the next round, or did they only flash briefly after the intervention?)

Make sure to use a time window of the same duration as the initial run to ensure the data sets are comparable.

If the metrics haven't budged, it usually boils down to two reasons: the change was not truly structural (e.g., you just held a single meeting or sent out a memo), or the time window was too short to reflect the shift. The former is far more likely.

In summary

Nowhere in the report is anyone evaluated. It describes positions and channels — who happens to stand at a critical intersection, and where a line is missing between two sub-groups.

This also means that structural problems cannot be solved by simply asking "someone to work a bit harder." Conversely, once you get the structure right, no extra heroism is required.

Curious what your own team looks like structurally?

Run it on your own relationship data. Provide members' work relationships, we will analyze only the collaboration' content.

Try a free diagnosis