RESPONSIBLE USE
Several Mondonomo models return a gender posterior alongside the name they're analyzing. This page explains how that number is derived, what it is intended for, and the uses it must not be put to.
The gender field is the conditional probability P(gender | name, locale) computed directly from MondoGraph — the proportion of recorded bearers of that exact name (in that locale, if supplied) who are documented as male or female.
It is a statistic about a population of past name bearers. It is not an inference about the person whose name was submitted.
The posterior is built for aggregate and tie-breaker tasks where calibrated population statistics are the right tool:
· Disambiguating duplicate records in CRM/CDP merges (e.g., picking the more likely
salutation when two records collide).
· Population-level analytics — how is our customer base distributed across countries
and demographic markers.
· Localization and culturally-aware fallback (e.g., choosing a default honorific in
languages that require gendered grammar).
· QA and detection of clearly-mislabeled records.
The posterior is not a personal-data inference and must not be used as one:
· Individual-level classification of a real person's gender.
· Inputs to decisions that affect a person's access to a product, service, price, credit,
insurance, employment, housing, or healthcare.
· Any context where the EU AI Act, GDPR Article 22, US ECOA, UK Equality Act, or other
regulations restrict automated decisions on protected characteristics.
· Targeted advertising of regulated categories (alcohol, lending, political content)
on an individual basis.
For these use cases, ask the person directly.
Pilot deployments can disable the gender field at the API gateway with a single header, return only the type/language/country fields, or replace the gender posterior with a binary “ambiguous” flag above a configurable confidence threshold. Talk to research@mondonomo.ai if any of these apply.
Calibrated, auditable population statistics on names are useful and lawful for the tasks above — and pretending the signal doesn't exist would just push customers toward worse substitutes (raw LLM guesses, scraped social profiles, name-only Bayes classifiers without evidence counts). Shipping the statistic with this explainer is the better answer.
Questions: research@mondonomo.ai · Related: data handling · Last updated 2026-05-18.