Can I analyse pay data for the gender pay-gap report without payroll leaving the company?

Yes: the processing goes to where the data already is. The sources enter an on-premise environment or the dedicated cloud with its data centre in Italy, and from that moment the files do not move. What comes out is the report and the analyses, not the export with names, salaries and leave records.

We have to prepare the first pay-gap report by 7 June 2027: where do we start with the data?

With a census of the sources — payroll, the HR system, external consultants’ spreadsheets, time and attendance — and with who sees what. Art. 11 of Italian Legislative Decree 96/2026 requires access to be limited to pay-equality purposes only: the restriction is set before processing starts, not after.

Our consultant asks us to email the payroll export: is there an alternative that stays inside the perimeter?

Yes. Categories, quartiles and gaps are calculated in the environment the company controls, and the documents the law requires are produced there. The consultant works on the results and on the criteria, not on the file with each person’s name, salary and leave record.

Can we use AI to analyse employees’ internal communications, and what are the limits?

The limit is written into the method: we compute aggregates and prepare filings, we do not infer emotions, moods or judgements about people. Italian data protection authority decision no. 342 of 14 May 2026 on employee sentiment analysis marked where the line runs, and art. 4 of the Workers’ Statute remains the constraint on any tool from which remote monitoring could follow.

An AI system touching recruitment or career progression is high risk: which obligations land on us?

In the work we do here the system does not propose actions about people: which corrections to make, and for whom, is decided by the employer through its own procedures. If instead you are introducing a system that touches recruitment or progression, the first step is to qualify the role and the risk class — the path described under the “AI governance and compliance” solution.

Solutions · Analysis of sensitive data

The data stays in. The analysis comes out.

Pay, job grades, confidential HR files: the processing goes to where the data already sits and returns reports and gap analyses. The file never leaves the perimeter.

A brass two-pan balance scale, perfectly level, resting on a wooden shelf

The shortcut costs more than the deadline.

From 7 June 2027 Italian Legislative Decree 96/2026 requires the first gender pay-gap report from companies with at least 250 employees and, on the same date, from those between 150 and 249 (art. 9). Producing it means cross-referencing, person by person, payslip, job grade, gender, bonuses and pay progression after parental leave: the most sensitive body of data a company holds after its trade secrets. The same decree, in art. 11, requires it to be protected and its access restricted to the purposes of pay equality alone.

The pattern that repeats is this — a typical model, not the account of a specific case: the payroll export uploaded to a cloud people-analytics tool, or emailed to a consultant, to save time. The report is delivered on schedule; the file with names, salaries and leave records has already left the perimeter, without anyone deciding so at policy level. It holds for pay data as it does for behavioural data: the Italian Garante’s decision no. 342 of 14 May 2026 on sentiment analysis of employees marked where the boundary runs, and art. 4 of the Workers’ Statute remains the constraint on any tool from which remote monitoring can follow.

Solutions

The processing travels to the data, not the reverse.

A census of the sources

Payroll, HR system, external consultants’ spreadsheets, time and attendance: first you map where pay data actually lives, then you set who sees what, and for which purpose.

Dedicated AI inside the perimeter

Processing runs on-premise in your environment, or on the CSIDIA dedicated cloud, with its data centre in Italy. Closed models, disconnected from the open web: the payroll file never leaves the company.

Aggregates and filings, not people

We compute quartiles, categories and gaps, and prepare the documents the law requires. We do not infer emotions, moods or judgements about people: the AI measures and prepares, it does not appraise. Every decision on staff stays with the employer and its own procedures.

Traces that withstand an inspection

A stated legal basis and purpose for every processing operation (GDPR arts. 5, 6, 9 and 88), restricted access, logs of every run, and documented training for those who handle this data.

An example, step by step.

A typical case, not a real client. A group of 900 employees across three companies, facing the first gender pay-gap report due on 7 June 2027. Payroll sits with an external firm, job grades live in the HR system, bonuses in a spreadsheet kept by the finance team. Nobody has ever joined the three sources, and nobody wants to send them around in order to do it.

01

Import into the perimeter.

The three sources go into the processing environment: on-premise at the company, or on the dedicated cloud with a dedicated VPN and a data centre in Italy. From that point the files stop moving; access is narrowed to pay-equality purposes only, before any processing starts, as art. 11 requires.
02

Computing the gaps.

The categories of workers doing equal work or work of equal value are reconstructed — eighteen in this example — and for each one mean and median pay, quartile composition and the variable component are computed. The calculation runs in minutes; the categories, though, are defined by the company, against criteria put in writing.
03

The gaps to justify.

The system flags the categories above the 5% threshold which, without a justification on objective and gender-neutral criteria, trigger the joint assessment with worker representatives. In this example there are four out of eighteen.
04

Simulating the justifications.

For each category the gap is tested against the objective criteria already in use — seniority, responsibility, certified skills, recorded performance — and what remains unexplained is measured. The system proposes no action on individuals: which corrections to make, and for whom, is decided by the employer through its own procedures.
05

Dossier for the report.

Out come the report in the format the law requires, the note on the criteria adopted, and the register of processing runs: which data, for which purpose, on which legal basis, run by whom and when.

On this example the preparation goes from weeks of work spread across HR, the external firm and the finance team to around ten person-days, almost all of them spent writing the criteria rather than reconciling spreadsheets. That is an estimate on a typical case, not a promise. The point that matters is a different one: no export containing names, salaries and leave records leaves the company’s perimeter, and no decision about staff is taken by the system.

An illustrative example on a typical case: the assumptions are recalibrated on your own data.

What we deliver.

The gender pay-gap report and the gap analysis behind it, produced without the payroll export ever leaving the company.
A map of pay-data sources and of access rights, restricted by purpose as art. 11 requires.
Objective, gender-neutral criteria for grading and progression, written down before a gap has to be justified.
A register of processing runs and a record of training for those who handle this data.
Analyses of confidential HR documentation — minutes, contracts, personnel files — under the same rule: inside the perimeter.

Who this is for.

HR and people directors above 150 employees, with the first report due on 7 June 2027.
Data protection officers who must restrict access to pay data without stopping the analysis.
Executive teams that want to know the gap, and the criteria behind it, before an inspector or a court asks.
Companies and public bodies handling confidential HR documentation that cannot be uploaded to external tools.

Two delivery modes.

On-premise, in your own environment

The AI runs on infrastructure you already control. Documents never cross the boundary of your network, and administration stays with your IT department.

Dedicated cloud, in Italy

An environment reserved for a single client, a dedicated VPN, a data centre resident in Italy, in premises we staff ourselves. Nothing is shared with other clients.

In both cases the models are dedicated and closed, disconnected from the open web: nothing they read feeds third-party services. They are open-weight models, with the weights archived inside the perimeter where they run: the version you use changes when you decide it does.

Who owns the machine and the model

Frequently asked questions.

Can I analyse pay data for the gender pay-gap report without payroll leaving the company?

Yes: the processing goes to where the data already is. The sources enter an on-premise environment or the dedicated cloud with its data centre in Italy, and from that moment the files do not move. What comes out is the report and the analyses, not the export with names, salaries and leave records.
We have to prepare the first pay-gap report by 7 June 2027: where do we start with the data?

With a census of the sources — payroll, the HR system, external consultants’ spreadsheets, time and attendance — and with who sees what. Art. 11 of Italian Legislative Decree 96/2026 requires access to be limited to pay-equality purposes only: the restriction is set before processing starts, not after.
Our consultant asks us to email the payroll export: is there an alternative that stays inside the perimeter?

Yes. Categories, quartiles and gaps are calculated in the environment the company controls, and the documents the law requires are produced there. The consultant works on the results and on the criteria, not on the file with each person’s name, salary and leave record.
Can we use AI to analyse employees’ internal communications, and what are the limits?

The limit is written into the method: we compute aggregates and prepare filings, we do not infer emotions, moods or judgements about people. Italian data protection authority decision no. 342 of 14 May 2026 on employee sentiment analysis marked where the line runs, and art. 4 of the Workers’ Statute remains the constraint on any tool from which remote monitoring could follow.
An AI system touching recruitment or career progression is high risk: which obligations land on us?

In the work we do here the system does not propose actions about people: which corrections to make, and for whom, is decided by the employer through its own procedures. If instead you are introducing a system that touches recruitment or progression, the first step is to qualify the role and the risk class — the path described under the “AI governance and compliance” solution.

Worth reading before you decide.

Operational from week one.

A real use case, on your data, in production. Then it grows, week after week.

30 minutes video call ~~€150~~ free

Start an operational trial

It starts with a session with our engagement expert. Your data stays yours, always.