People analytics with AI

People analytics helps HR find mistakes and trends in HR data. Clear rules do the finding. AI explains what was found, groups survey comments into themes, and answers questions about your HR numbers in plain words. Nothing here changes a record, pays anyone, sends a warning or makes a decision. Every result is information for a person to review.

Contents: Who can use it · Turning it on · Checks: attendance, payroll and data quality · Predictions · Burnout indicators · Survey comment themes · Ask a question · Privacy and safety · FAQ

Who can use it

It's on every plan. Roles get these permissions on the Roles screen, under AI in HR:

• See checks, burnout by department and survey themes: AI analytics (List). You see flags only for the employees you can already see. Payroll flags also need access to payroll and pay. Attendance flags need the attendance summary. Data quality flags need the employee list.
• Run checks, ask AI to explain, close flags, change settings: AI analytics (Run).
• See predictions and burnout per person: AI predictions (List). Running predictions also needs AI analytics (Run). Give this only to the few HR people who need it.
• Ask questions in plain words: AI questions (Use). You can only get answers from the HR report data you can already open in the report builder.

Survey themes also need access to the survey results.

Roles that could already see or change the AI settings for HR get the analytics and prediction permissions. Roles that could already use the report builder get AI questions (Use).

Where: HR → AI in HR → People analytics, with the Checks, Predictions, Burnout and Ask tabs. Survey themes are on a survey's results page, under Comment themes.

Turning it on

Each part starts off. An HR admin turns it on in AI in HR, on the Features tab:

• Payroll, attendance and data quality checks for the checks.
• Attrition and absence predictions for predictions.
• Survey comment themes for survey themes.
• Questions about HR numbers in plain words for the Ask tab.

You can turn each one on for every country or for one country. People in a country where a feature is off are left out. If all AI is paused, nothing runs.

Checks: attendance, payroll and data quality

1. Open People analytics → Checks.
2. Click Run checks. You can pick only Attendance, Payroll or Data quality.
3. Flags appear when the checks finish. The most serious ones are at the top.

What the checks look for:

Area	What is flagged
Attendance (last 90 days)	Late arrivals, absences or overtime well above the rest of the organization, and several days with a check-in but no check-out
Payroll (latest payroll month)	Net pay that changed sharply from the month before, net pay below zero, zero net pay when gross is not zero, deductions that take most of the gross, and a person paid twice in one month
Data quality (active employees)	Missing core fields (joining date, department, designation, date of birth, worker type, branch), impossible dates (born in the future, joined before working age, left before joining, joining more than a year ahead), and possible duplicates (same national ID, or same name and date of birth)

"Well above" means far above your own organization's average. A minimum also applies, so a quiet month doesn't flag small differences.

Each flag shows the figures behind it. Then:

• Explain with AI gives a short explanation and a list of things to check. AI gets only the rule and the figures, never the person's name.
• Dismiss means it's not a problem. Mark resolved means you fixed it at the source. Both need a short note. Reopen puts it back on the list.

When you run the checks again, flags you dismissed or resolved stay as they are. An open flag that is no longer found is marked Cleared. The checks never correct anything themselves. Fix the record in its own screen.

Predictions

Predictions estimate attrition risk (someone may leave) and absence risk (someone may be absent more often). They are high-risk information, so:

• Only people with AI predictions (List) can see them, and only for employees they can already see.
• Every prediction lists the factors behind it. The score is the sum of those factors' weights, from 0 to 100: Low (under 35), Medium (35 to 59), High (60 and over).
• Nothing happens because of a prediction. No message, task, status change or decision follows.
• Every prediction is saved in the AI decision log.
• People who opted out of AI processing are left out.

The factors are about work patterns only:

Attrition	Absence
Joined less than a year ago	Absence rate well above the organization
No leave taken in the last 180 days	Absences rising in the last 45 days
Overtime well above the organization	Frequent late arrivals
Absences rising in the last 45 days	No leave taken in the last 180 days
Late arrivals rising in the last 45 days	Overtime well above the organization
Many very long working days	Many very long working days
Several people left the same department recently

Age, gender, religion, marital status, nationality, health and family are never used.

To run them, open People analytics → Predictions and click Run predictions. Filter by Attrition or Absence, by band or by department. Open one to see its factors. Explain with AI writes a plain-words explanation and suggests supportive topics for a conversation, such as workload or time off. It never suggests discipline, dismissal or pay changes.

Fairness: the prediction model can be tested in AI in HR → Fairness. The test uses made-up people who differ only in a protected attribute, and checks that every group gets the same results.

Burnout indicators

Burnout indicators are off until HR turns them on. When they're on, HR sees them by department by default.

1. In People analytics → Burnout, turn on Burnout indicators.
2. The page shows each department's people, how many are at risk, and how many have each indicator.

The indicators are high overtime, many very long working days (over 10 hours), work on holidays, and no leave for more than 180 days. A person with two or more of these counts as at risk.

A department with fewer people than the minimum group size of HR reports (5 unless 9bx support changed it) is hidden. If only one department is hidden, the next smallest one is hidden too. That way nobody can work out the hidden figures from the total.

To see indicators per person, also turn on Show per person. Only people with AI predictions (List) can then see the list. People who opted out of AI processing are never listed. Nothing is sent to the employee or their manager.

Survey comment themes

A survey's written comments can be grouped into themes, with a sentiment for each theme and for the survey overall.

• The survey must have Allow AI analysis of comments turned on. People see this setting when they answer.
• Themes appear only after at least the survey's minimum group size has answered.
• AI gets the comments of the whole survey, sorted alphabetically and with names masked. It returns themes and counts only, never quotes. A theme that only one comment mentions is left out.

Open the survey results, go to Comment themes, and click Find themes. The latest result is kept, so you can look at it again later.

Ask a question

On the Ask tab, type a question about your HR numbers, for example "How many people joined each department this year?" or "Average late minutes by branch last month".

The answer shows:

• What was looked up: the report data, measures, breakdown, filters and dates that AI picked.
• Source: the report builder data it came from.
• The figures, as a small table.

AI can only pick from the HR report data you can already open in the report builder. That data has a fixed list of measures, breakdowns and filters. AI cannot read anything else. Answers are always totals, never lists of people. A group smaller than the minimum group size is hidden, just like in a report. Questions about one named person, or about data outside the list, get "This question cannot be answered" with the reason. Your recent questions are listed under the box.

Privacy and safety

• Rules find the flags, indicators and prediction factors. AI only explains, groups or translates a question.
• Every AI result is saved in the AI decision log, with the model, the version and who asked.
• Names, email addresses and phone numbers are masked before anything goes to AI, unless your organization turned masking off.
• An employee can opt out of AI processing in My HR. They're then left out of predictions and burnout per person, and AI won't explain flags about them.
• Employees see the notice "Your organization uses AI to help spot errors and trends in HR data…" where AI is used for them.

Troubleshooting and FAQ

"This AI feature is not available: off". The feature is off for your country. Turn it on in AI in HR → Features.

"This AI feature is not available: opted_out". The employee opted out of AI processing. You can still fix the flag by hand.

"This AI feature is not available: held". A fairness report failed and must be reviewed in AI in HR → Fairness before predictions run again.

I clicked Run checks but see no flags. The checks run in the background. Refresh after a minute. If nothing is unusual, there's nothing to flag.

A department shows "Hidden". It has fewer people than the minimum group size, or it had to be hidden so a smaller department can't be worked out.

Can a prediction or a flag start a disciplinary case or change pay? No. Nothing in People analytics acts on its own. A person decides, in the normal screens.