Last month a study of tens of thousands of information workers reported something that eighteen months of dashboards had failed to notice. When an entire firm moved to remote work, its collaboration network became more static and more siloed: people spent more time with the contacts they already had, formed fewer new ties, and the bridges that connected one group to another thinned out. Communication shifted toward asynchronous channels, and the network stopped rewiring itself. The finding is interesting. The measurement is the important part. That research did not count messages, meeting hours or active days. It looked at who was connected to whom, and at how those connections changed. Almost no organisation can produce that view of itself, because almost every collaboration analytics deployment in existence is counting activity — and activity is the wrong quantity.
Activity metrics measure exhaust
Open any collaboration dashboard and you will find messages sent, meetings attended, hours in meetings, after-hours activity, channel counts and active users. Every one of those is a by-product of work rather than a description of it, and several of them move in the wrong direction. Meeting hours rise when coordination is failing. Message volume rises when decisions are unclear. After-hours activity rises when the day is too fragmented to finish anything, which is a warning rather than an achievement. A metric that increases when the organisation is working badly is not a measure of collaboration. It is a measure of friction, and reading it as engagement is how companies end up celebrating their own dysfunction in a quarterly slide. The deeper problem is that activity is individually attributable. Once a number exists per person, somebody will put it in a performance conversation, and the moment that happens the measurement is over: people manage the number and the data stops describing reality.
Four questions a network view can answer
Network measurement asks about structure instead. Four questions carry most of the value. Are ties crossing team boundaries, and is that growing or decaying? The proportion of interaction that happens outside a person's immediate group is the closest thing to a health indicator that collaboration data offers. Decay is slow, invisible in activity counts, and expensive: it shows up two quarters later as duplicated work and decisions that surprise the people who should have been consulted. How long does a new joiner take to build ties outside their own team? This is the single most useful measure in a distributed organisation. In an office, that integration happened accidentally. Distributed, it happens only if somebody designs it. If your median joiner has no meaningful contact outside their team at ninety days, you have found the cause of an attrition problem you are currently attributing to compensation. Who is the only path between two parts of the business? Every organisation has two or three people through whom information between functions actually travels. Those people are usually praised. They should instead be treated as a resilience risk, because the organisation has an undocumented dependency on an individual's relationships, and it will discover this the week they resign. Is information flowing in both directions? A function that sends far more than it receives is broadcasting, not collaborating. That pattern is common in finance, compliance and IT, and it predicts exactly the complaint those functions hear about not being involved early enough.
What you can compute without buying anything
Most of this can be derived from metadata you already hold: the directory and reporting lines, calendar attendee lists, message and channel metadata, document co-authorship and comment threads, and assignment records in whatever tracks work. Content is not required and should not be used. Ties come from who appeared together, not from what they said. Be honest about the limits. Channel membership is not a relationship. A recurring meeting of thirty people generates ties that do not exist. The most consequential conversations still happen in calls, corridors and messaging apps outside the tenant. A network map is a hypothesis about the organisation, not a photograph of it, and every finding worth acting on should be checked with a handful of interviews before anybody reorganises anything.
Practical Guidance for Collaboration Analytics
- Write the question before you pull the data, and refuse any analysis that begins with what the tool can show us.
- Use metadata, never content, and record that decision in the documentation.
- Set a minimum group size — five is a reasonable floor — and never publish a cell below it.
- Never produce an individual-level output for a manager. If a finding can be used against a named person, the analysis is designed wrong.
- Measure integration of new joiners at thirty, sixty and ninety days as a standing indicator rather than a one-off study.
- Treat single-path individuals as succession and resilience risks, and act structurally rather than by praising them.
- Compare only against your own trend, because external benchmarks for this are marketing.
- Time-box each study and delete the derived dataset on a published schedule.
Findings only matter if they change a decision
The common failure is a beautiful network diagram presented to a leadership team, admired, and filed. Each of the four questions has an intervention attached, and the intervention should be specified before the analysis runs. If cross-team ties are decaying, the answer is shared work rather than shared socials: put people from two functions on the same deliverable with the same deadline. Cross-team coffee schemes generate ties that do not survive the first busy week. If new joiners are isolated at ninety days, assign a structured set of introductions with a named person and a purpose for each, and review it as part of probation. If one individual bridges two functions, document what flows through them and create a second path deliberately. If a function is broadcasting, change who is in the room at the start of decisions, not the volume of updates at the end. And resist the temptation to set a target. The moment a tie count becomes an objective, people will produce ties.
| Question | Response proposed |
|---|---|
| Are cross-team ties decaying? | Assign a shared deliverable across functions |
| Are new joiners isolated? | Plan purposeful introductions with named contacts |
| Does one person bridge two functions? | Document what flows through them and create a second path |
| Is a function broadcasting rather than listening? | Include it earlier in decisions rather than adding updates |
Qualitative summary of this article's source text, not a measured outcome or performance estimate.
The Regional Angle
Three things make this analysis different in Gulf organisations, and the first will invalidate your results if you ignore it. The most influential ties in a regional business frequently run outside every system you can measure. A decision travels through a majlis, a personal messaging group, a call in a car, or a five-minute conversation with a group chairman. None of that appears in calendar metadata. If you build a network map from corporate tools in a family-owned group, you will reliably identify a mid-level coordinator as the most connected person in the company, because that person lives in email while the actual bridges do not. Treat the map as a partial view of the formal organisation, state that limitation in the first slide, and validate the structural findings through interviews with people who know how things really get decided. Used that way, the analysis is still valuable, and the gap between the measured network and the real one is itself a finding worth discussing. The second is more delicate. In a workforce drawn from sixty or a hundred nationalities, network analysis will surface clustering by language and nationality, because it is really there: the Arabic-speaking commercial team, the finance function recruited largely from one country, the operations floor that works in a third language. The operational consequence is real, since information genuinely fails to cross those boundaries, and identifying it is legitimate. Publishing it carelessly is not. A report that names communities by nationality is inflammatory and creates exposure that no analytics project should be generating. The disciplined approach is to analyse clustering by function, site and language of work, keep any demographic dimension aggregated and inside the human resources function, and direct every intervention at structure — rotating who chairs, mixing project teams, requiring written bilingual summaries of decisions — rather than at individuals or groups. The third is legal footing, which is changing under this analysis as it is being designed. There is no works council here to consult, which removes an obstacle and also removes the legitimacy that consultation confers, so the organisation has to construct its own. Do it explicitly: an approved purpose, a documented lawful basis, an aggregation policy, a retention limit and a note to staff describing what is analysed and what is not. That housekeeping is about to matter more than it did. Saudi Arabia published its personal data protection law in the past fortnight and a federal law for the Emirates is expected shortly, and communication metadata is personal data under every modern regime including the one already in force in the Dubai financial centre. A programme documented properly this quarter will survive the arrival of implementing regulations next year. One assembled informally by an enthusiastic analyst will not.
The objection worth taking seriously
The strongest objection is that this is surveillance with a better vocabulary. Aggregation thresholds are a policy rather than a technical guarantee, and anybody holding the underlying dataset can re-identify individuals trivially. Managers who are told they cannot see person-level data will ask for it, and eventually someone will agree. And the interventions the analysis recommends — connect the disconnected, plan introductions for new starters, reduce dependence on one person — are things a competent manager already knows without a graph. The re-identification point is correct, and the honest response is not a reassurance but a design: the derived dataset sits with a small, named team, is never exposed through a self-service tool, and is deleted on schedule. If your vendor's product puts a person-level view one click away from a manager, that product is a monitoring tool whatever the brochure says. Where the objection fails is in the comparison it implies. The alternative to network measurement is not an unmeasured workplace. Activity is already being collected, already being read, and already being used: presence indicators, attendance records, response times, the green dot. Those measures are individually attributable, easy to weaponise and describe nothing useful. The choice is between a bad measure of individuals and a good measure of structure. And the past eighteen months made the case for structure empirically. An organisation's network changed materially, in ways that will affect innovation and integration for years, and the change was invisible to everyone whose dashboard counted messages.
Common Questions
Do we need to buy a dedicated tool?
Not for a first study. Directory data, calendar metadata and message metadata will answer the four questions well enough to decide whether the topic deserves investment.
How do we tell staff about this?
Before, in plain language: what is analysed, what is not, the minimum group size, who sees output, and how long the data is kept. The announcement is cheaper than the rumour.
Is any of this predictive of attrition?
Weak external ties in the first ninety days tracks with early leaving in most organisations that have looked. Treat it as a prompt to intervene rather than as a score attached to a person.
What should we expect over the next twelve months?
Expect these capabilities to arrive inside the collaboration platforms you already pay for, bundled into the employee experience suites launched this year, which turns this from a procurement decision into a configuration decision that IT will make by default unless someone governs it. Expect the new Gulf data protection laws to publish implementing regulations during next year that treat communication metadata as personal data. Expect more published research as the natural experiment continues, and expect the findings to become more nuanced than the early headlines. And expect the argument about returning to the office to start citing network decay as evidence — at which point your leadership will ask what your own numbers say, and the organisations that began measuring structure this year will be the only ones able to answer.
Collaboration Analytics Review — we define the questions worth asking, build the network view from metadata you already hold, and put the governance around it that keeps the answer useful and the exercise legitimate.