Company Culture: What Your Communication Data Can and Cannot Tell You

By Nicolas Saliba · Published April 2025, Updated August 2026 · 8 min read

A CHRO asks the internal communication team to help strengthen the culture ahead of a three-year plan. The team builds the programme: a values refresh, a leadership video series, a run of intranet stories about people who exemplify the behaviours. Twelve months later the same CHRO asks whether it worked. The engagement survey has moved two points, which is inside the noise. The campaign pages have view counts. Nobody in the room can say whether anything changed in how the organisation actually behaves.

That is the honest state of culture measurement in most large organisations, and it is not a failure of effort. Culture is asserted in communication and revealed in behaviour, and those two things are measured with different instruments. This article is about what your communication data genuinely tells you about the second one, and, just as importantly, about the claims it will not support.

Culture has two layers, and internal communication sees both

Long-running work on organisational culture in Harvard Business Review describes it as the shared assumptions, norms and behaviours that persist without anyone enforcing them^[1]. For a measurement conversation, it helps to split that into two layers.

The stated layer is the culture the organisation publishes: values, the leadership narrative, the recognition programme, the story you tell candidates. Internal communication produces most of it. It is easy to count and almost worthless as evidence. Campaign impressions tell you the message was manufactured, not that anything is enacted.

The enacted layer is what people actually do: who talks to whom, which questions get asked in public, what gets answered, whose posts get ignored, where a cascade stops. HBR’s culture research makes the point that the enacted layer is where strategy is either supported or quietly overruled^[2]. Internal communication has partial visibility of this layer, which is more visibility than most functions have, and the partiality matters.

The engagement survey is the right instrument, and an insufficient one

It is worth conceding this properly before arguing anything else. Surveys measure belief, perception, intent and felt experience. Those are real and they matter, and no amount of communication data substitutes for asking people what they think. Gallup’s global workplace research has spent two decades demonstrating that self-reported engagement tracks to outcomes an executive committee cares about^[3]. If you want to know whether people feel respected, ask them.

The insufficiency is structural rather than methodological. Surveys run once or twice a year, so they cannot tell you whether a specific change landed in March. They rely on self-report, so they carry the usual attribution problems. Response rates skew towards the populations that already feel heard, which is exactly the wrong bias when you are looking for the parts of the organisation that are disconnected. And a survey gives you a score, not a mechanism: it says trust in senior leadership is at 54, not that the reason is a cascade that dies at the second management layer in the manufacturing division.

Gallagher’s State of the Sector research consistently reports measurement as the capability internal communicators name first when asked what they lack^[4]. The gap is not that teams do not survey. It is that they have one instrument, used twice a year, and nothing continuous to put next to it. Our note on engagement surveys compared with communication data works through where each one is the better answer.

The behavioural signals you already hold

Communication activity is behavioural residue. It records what people opened, where they stopped reading, what they replied to, which questions they were willing to ask with their name attached, and which populations never appeared at all. None of that was collected as culture research, which is precisely why it is useful: nobody was performing for it.

Four signals carry genuine information about the enacted layer. Each one is paired below with the claim it will not support, because over-claiming on this material is how an IC team loses credibility with a CHRO.

Signal	What it may indicate	What it does not prove
Reply rate on leadership posts, against read rate on the same post	Whether speaking up in front of the organisation feels normal or exposed	That people trust leadership. Silence has many causes, including indifference
Share of reading that crosses a function, country or site boundary	Whether the organisation is connected laterally or only vertically	The quality of that collaboration, or whether anything came of it
Reach at headquarters compared with reach three management layers down	Where a manager cascade stops working	Why it stopped. Manager capacity and message relevance look identical in the data
A population that is persistently absent across every channel	That the channel model structurally excludes them	Disengagement. Shift patterns, shared devices and language often explain it

The over-claiming trap

The single fastest way to discredit this work is to narrate causation from a trend line. Viva Engage replies falling by a third during a redundancy consultation is not evidence that trust collapsed. People post less in public when their job is under review, for entirely rational reasons, and the same pattern would appear in an organisation with a healthy culture handling a difficult period well.

The same caution applies to sentiment analysis. Automated sentiment is useful for triage across thousands of comments in several languages, and it is genuinely helpful for spotting where to read closely. It is not a culture verdict, and it should augment the judgement of an experienced practitioner rather than replace it. We set out the limits in the piece on sentiment analysis in internal communications.

The workable rule: communication data is strong at locating a question and weak at answering why. Use it to decide which three sites to visit and which two manager populations to talk to, then go and ask.

Signals mean nothing without a denominator

Every measure above is a ratio, and the denominator is a population rather than a page. “1,400 readers” is not a finding. “38 per cent of the commercial function and 4 per cent of manufacturing” is a finding, and it points at a specific room.

Producing that requires joining communication activity to organisational structure: function, grade, site, country, language, contract type, manager population. In a Microsoft 365 organisation those attributes usually live in Entra ID or in an HR extract^[5], not in the analytics panel of any individual channel. Getting the attribute model right before you start measuring saves a year of rework, which is the subject of our guide to audience segmentation for internal communications. The population most likely to be invisible is the one furthest from a desk, and the practicalities of that are covered in measuring frontline worker communications.

The honest position on Microsoft native analytics

SharePoint site analytics and Viva Engage community insights are free, already switched on, and sufficient for a single team measuring a single surface^[6]. If you run one community and want to know whether your posts land inside it, the native panel does that job and nothing here argues otherwise.

The limits appear the moment the question becomes organisational. Native reporting does not join to HR attributes, so the denominator above is not available. Retention windows are finite, which makes multi-year culture narrative difficult precisely when a three-year plan needs it. And each surface reports on itself, so a person who reads the intranet, ignores Viva Engage and opens every newsletter appears as three unrelated fragments.

Teams often try to close this with a Power BI build on the Microsoft Graph reporting API. That route can work, but it is more demanding than it looks. The reporting API is not consistently reliable and produces days with missing data^[7]. Customers have observed KPI values in Power BI that do not reconcile with the native SharePoint reports for the same period. A dependable build needs a robust ingestion process and scheduled reconciliation against the native figures, so budget for the pipeline rather than the dashboard. Our note on SharePoint analytics in Power BI covers the specific failure modes.

This is the layer Tryane occupies. It reads SharePoint, Viva Engage, Teams and your newsletter platform together, joins that activity to organisational structure through Entra ID or an HR file, and keeps history across product transitions so a culture trend can span more than one reporting window. Tryane is SOC 2 Type 2 certified and GDPR compliant by design, hosts in the EU by default with US data residency available on request, deploys in a couple of hours using single sign-on through Azure AD or Entra ID, and installs no agent on any device.

How to put this in front of a CHRO

Structure the conversation in three parts and resist the temptation to merge them. First, what we produced: the campaign, the cadence, the channels. Second, what people did: the behavioural signals, segmented, with the caveats stated out loud. Third, what we still need the survey to tell us, and what we will go and ask in person.

An IC function that says “here is where the message stopped, here is the population it never reached, and here is what we cannot conclude from this” is more credible than one presenting a culture score. The measures that hold up under that kind of questioning are set out in our guide to internal communication KPIs, and the framing that works with an executive audience is covered in proving internal communications works to leadership.

Frequently asked questions

Can communication data measure company culture?

Not on its own, and any vendor claiming otherwise is overselling. Communication data evidences behaviour: what was read, replied to, ignored, and by which parts of the organisation. Culture also includes belief and felt experience, which requires asking people. The useful position is that communication data is a continuous behavioural signal that sits alongside a periodic survey, and that each one answers questions the other cannot.

Is a drop in engagement on our intranet a culture problem?

It might be, and it might be a publishing problem, a search problem or a reorganisation. Aggregate movement rarely carries a single explanation. Segment the drop: if it is concentrated in one function or one country, you have a specific question to investigate rather than a general anxiety.

How often should we look at these signals?

Monthly is enough for the trend, with a closer look around known events such as a restructuring or a leadership change. Weekly reporting on culture signals tends to generate noise-chasing. The exception is a live crisis, where daily reach by population is genuinely useful.

Does this require HR data, and is that a privacy problem?

It requires organisational attributes, not personal records. Function, site, country and manager population are enough, and reporting should be set to a minimum group size so no individual is identifiable in a segment. Agreeing that threshold with your works council or data protection officer before the first report is the step teams most often skip.

Do we still need the engagement survey if we have this?

Yes. They are different instruments. The survey tells you what people believe and how they feel, at a point in time. Communication data tells you what people did, continuously, with segmentation the survey cannot reach. Organisations that run both tend to use the behavioural signal to decide where to look and the survey to understand what they found.