Monitoring exists to answer one question: can you show, with evidence, that a decision affecting trans and nonbinary people was reasoned rather than assumed? That is what the Public Sector Equality Duty asks for — due regard, evidenced — and it is what any organisation needs if a policy or decision is ever challenged. It is not a mandate to track everything you are technically able to capture. Between those two things sits a line that well-intentioned monitoring programmes cross without noticing: the line between measurement that supports a decision and collection that just accumulates sensitive personal data because the fields were available in the HR system. This playbook is about staying on the right side of that line.
What monitoring is for
Good monitoring exists to evidence two things: that your organisation is meeting its Public Sector Equality Duty obligations (or, for non-public bodies, that it can show a comparably reasoned approach), and that specific decisions — a policy change, a facilities decision, a case-by-case judgement — were made on evidence rather than instinct. Both uses point outward, to a decision or a duty you can name. If a metric cannot be tied to either, it is not evidencing anything; it is just data sitting in a system, waiting to become a liability.
That distinction is also where over-collection creeps in. It is easy to justify capturing “one more field” — a finer breakdown, a more frequent survey, a system that records gender history alongside current presentation — on the grounds it might be useful later. Under data protection law, “might be useful later” is not a lawful basis. Monitoring that cannot point to a specific, current purpose is surveillance wearing a governance label: it carries risk without buying any defensibility in return.
Lawful data handling: minimise, don’t just collect
Gender identity and gender reassignment information sits in a category that data protection law treats with particular care. Gender history and transition status should be treated as sensitive personal information requiring a clear lawful basis and a specific processing condition before they are processed at all, the same tier of protection applied to health data. For this characteristic the usual equality-monitoring gateway does not fit: the equality-of-opportunity condition in Data Protection Act 2018 Schedule 1 (paragraph 8), which underpins much routine diversity monitoring, does not extend to gender reassignment, so it cannot be the processing condition for gender-reassignment data. Voluntary monitoring of this data therefore usually relies on the individual’s explicit consent as the lawful gateway — though the right condition still turns on the specifics, so take your own data protection advice. Three principles follow directly from that, and all three should be checked before a single field is added to a form or a system:
- Minimise. Collect the least data that will answer the question you actually have. A headcount by broad category is often enough to evidence a decision; a granular breakdown by site, team and grade rarely is, and it multiplies the risk described below.
- Purpose-limit. Data collected for one stated reason should not quietly get reused for another. If a self-ID field was introduced to support facilities planning, using the same data for a different purpose later needs its own justification, not an assumption that “we already have it.”
- Ground it in a lawful basis, then stop there. If you cannot articulate the lawful basis for holding a piece of data — and articulate it in terms someone outside HR would accept — you cannot justify holding it. Don’t collect what you can’t justify, and don’t keep it after the justification has expired.
Self-identification should be voluntary in substance, not just in name: a genuine “prefer not to say” option with no downside, a clear explanation of what the data will and won’t be used for, and no treatment of non-disclosure as a gap to be closed through pressure. The Beyond Compliance survey found organisations themselves doubt how safe disclosure feels: 30.6% estimate that no more than one in ten of their trans and nonbinary staff would feel comfortable disclosing their identity at all. Self-ID that people don’t trust does not just under-collect — it signals that disclosure is being watched rather than supported, which suppresses the openness the monitoring was meant to encourage.
Small numbers change everything
This is the point most monitoring frameworks miss, because it cuts against the instinct to report data at the most granular level available. Trans and nonbinary staff are, in almost every organisation, a small fraction of the workforce — and visible, disclosed numbers are smaller again than the true population, for the trust reasons above. That combination means granular breakdowns are exactly where anonymity fails. A report that slices trans-inclusion data by department, site, grade or any other dimension can, with very few individuals in the relevant group, effectively identify a specific person — even without a name attached.
The safeguard is small-number suppression. Set a minimum reporting threshold. A widely used convention in statistical disclosure control — the kind of approach ONS and NHS bodies apply to their own published statistics — is not to publish or circulate any count below five. No single threshold is fixed in law, though: the right number depends on your organisation’s size, and your data protection officer should sign off the approach you adopt. Part of what that threshold protects is preventing the inference of any individual’s gender history — including that of a gender recognition certificate holder, where section 22 of the Gender Recognition Act 2004 is engaged — from a figure that is otherwise anonymous. Suppress or aggregate any figure below it, rather than rounding it or footnoting it as “small sample.” Apply the threshold consistently to combinations of variables too: a small overall headcount can still become identifying once crossed with one other characteristic, such as site or seniority. Build this into the reporting template itself, not into the judgement of whoever pulls the figures that quarter. Where the underlying number is too small to report safely, the correct output is “insufficient data to report,” not a workaround that publishes it in a different shape.
What good metrics look like
A metric earns its place by being tied to a decision, not by being available. That means favouring a small set of process and outcome measures over a long dashboard of everything the HR system can produce.
Useful process measures include whether the inclusion policy has a review date and whether that date was kept, whether managers have received training and can say with confidence that they understood it, and whether there is a working, documented process for name and gender-marker changes. These sound administrative, but each is a direct proxy for whether the organisation can act consistently when a real situation arises — and each is something the organisation controls, unlike headcount figures that depend on disclosure.
Useful outcome measures are tied to a specific commitment: time to resolve a raised concern, consistency of facilities provision across sites, or whether a policy’s stated coverage matches what managers report applying in practice. What doesn’t belong is anything collected because it’s easy to produce rather than because a decision depends on it — a running count of “diversity events attended” tells you activity happened, not that anything got safer, fairer, or more defensible.
The absence of a metric is itself informative. The Beyond Compliance survey found a striking share of organisations simply do not know the answer to basic questions here: 30.1% don’t know whether their systems can even record pronouns, and 33.1% don’t know how confident their own managers feel handling this area. Not knowing is not neutral — it means the organisation cannot currently evidence anything either way, which is its own governance gap.
Monitoring as part of defensible governance
Monitoring only earns its keep as part of a wider governance structure — it needs an owner, a reporting line, and a reason to exist beyond its own collection. The Beyond Compliance survey found that just 6.1% of organisations tie inclusion outcomes to executive KPIs or accountability measures, and 70.9% have no named lead for the area at all. Data collected into that vacuum tends to sit unused, unreviewed and undefended — worse than not collecting it, because it still carries the handling risk without any of the accountability benefit.
The EHRC’s updated statutory Code of Practice is guidance, not law in itself — a court or tribunal must have regard to it, but it does not by itself create new legal obligations. What it usefully models is the expectation running through this area: that organisations should be able to show their working. Lawful, proportionate monitoring — minimised, purpose-limited, suppressed once numbers stop being safe, and tied to named ownership — is what makes that possible. Monitoring built to look thorough rather than to answer a specific question does the opposite: it creates a paper trail of sensitive data with no decision to defend, which is a liability the organisation has built for itself.