AI engagement assessment and continuous eNPS

The manager learns that an employee is burned out or planning to leave during the exit interview - when the decision has already been made.

An annual engagement survey provides a snapshot once a year and misses the moment when sentiment changes.

A line manager oversees dozens of people and simply cannot track who is growing tired, who is being overlooked, or who stopped suggesting ideas a month ago.

Hub AI implementations for business already promises an HR assistant that suggests how to retain a specific employee.

This page explains the mechanism: how the assistant continuously evaluates engagement and calculates eNPS - and where the line is between "helps the manager"

eNPS (Employee Net Promoter Score) is an employee loyalty metric adapted from customer NPS.

It is based on one question: 'On a scale from 0 to 10, how likely are you to recommend the company as a place to work?'

Responses are divided into three groups: promoters (9-10), neutrals (7-8) and detractors (0-6). The result is calculated as the share of promoters minus the share of detractors; neutrals are excluded from the formula. The range is from -100 to +100.

The NPS approach was created by Fred Reichheld (Bain & Company) and presented in Harvard Business Review in December 2003 in the article "The One Number You Need to Grow."

eNPS applies the same logic to employees.

Not all processes and not all data fall under GDPR, but personal - those that identify a person: full name, phone number, email, and any information tied to a specific employee.

Employee data is personal data, and processing it requires a lawful basis.

Employment relations are a recognized basis, but AI analysis of internal communications without an explicit basis may be treated as a violation, and penalties have been tightened.

There are two clean ways to keep the process within the legal framework.

First - consent: it must be specific, informed, voluntary, and unambiguous, and from September 1, 2025 it must be issued as a separate document from other signed paperwork.

Second - anonymization: anonymized data can be processed without consent.

That is why eNPS is calculated from aggregated and anonymized signals - this is an architectural response to the risk, not a paperwork one.

The mechanism is already implemented in LLM & Security Gateway: two-way personal data obfuscation before the model and deobfuscation in the response - the model and the provider logs see only anonymized text.

This is the proof of the boundary: the employee's personal data does not leave the perimeter.