Skip to main content
We use generative AI technology to provide useful information while striving to minimise inaccuracies. We use reasonable efforts to ensure that personal information is not used in this process and, to the extent possible, automatically redact personal information before using AI. We strongly advise end-user consumers not to enter personal information into any AI-powered feature. For questions, contact legal@mention-me.com.

​
AI Capabilities

Generative AI — We use generative AI to summarise, categorise, and make actionable the large volumes of verbatim text from your customers’ share messages and/or NPS feedback. This helps you optimise advocacy. We are also experimenting with generative AI to support programme optimisation as a platform assistant. Machine Learning — We use ML techniques for customer segmentation and personalisation predictions.

​
Technologies and Frameworks

  • Predictive models: Individual models are built for each client using only their own data. XGBoost powers Propensity To Refer, predicting a customer’s likelihood to refer a friend. Keras powers the predictive elements of Extended Customer Revenue, predicting customer churn and lifetime value.
  • Summarisation and sentiment analysis: LLM models (ChatGPT 3.5, 4, and later, upgraded as new models are released).

​
Controls and Safeguards

  • LLM output is never displayed to end customers. Only client employees see generative output.
  • Input filters reduce the likelihood of inappropriate content.
  • We have completed a DPIA covering our use of generative LLM and are satisfied the risk is low given our use cases, particularly because the risk is significantly lower than if generative AI were used in a live chat format with end users.
  • We use a combination of caching, filtering, and monitoring to balance accuracy, performance, and reliability.

​
Model Maintenance

  • ML models are retrained at least monthly (some weekly) using the latest data.
  • Performance is monitored over time against a holdout group with incremental upside reported.
  • A team of data engineers monitors and manages pipelines with weekly training schedules.
  • We continuously monitor the output of our generative AI models and adopt the latest model features as they become available.

​
Transparency and Explainability

  • Feature importances are recorded for all ML models.
  • The SHAP framework is used to understand individual decisions, though this is not currently surfaced to users or regularly calculated.

​
Ethics and Fairness

  • Protected characteristics are not used as features in our models.
  • We have completed a DPIA covering our use of OpenAI to process share messages and NPS feedback, and are satisfied we take the necessary steps to reduce the risk of personal information being processed and to ensure appropriate accuracy and lack of bias.
  • We have covered risks in our DPIA and are satisfied that, because of the use cases we adopt, our use cases carry low risk.

​
Performance Metrics

We focus on metrics for production performance and the accuracy of decision-making.
  • Propensity To Refer: Model ROC AUC, predicted propensities in high and low groups, feature importances, and incremental revenue generated.
Last modified on July 31, 2026