Medallia AI
Artificial-intelligence-driven features for Medallia Experience Cloud
Medallia AI is the Medallia Experience Cloud layer that powers artificial-intelligence-driven features in Experience Cloud, including:
- Action Intelligence
- Designed to enable actionable, adaptive intelligence in experience programs with natural language processing, deep learning, continuous learning, and predictive analytics
- Generative AI reporting
- Restriction: This feature is part of a generative AI feature set available to customers with the requisite subscription agreement. Once the agreement is signed, feature entitlement is automatic after 5-7 business days. Activation may require additional configuration work. For more information, contact Medallia Support.Reporting components for root-cause analysis and data summarization, including Root Cause Assist, Themes with generative AI, Intelligent Summaries for conversational data, and Intelligent Summaries for Text Analytics
- Medallia Speech
- Experience Cloud an add-on to Medallia Experience Cloud that quickly and accurately transforms voice signals, such as voicemail messages and customer interactions with call center agents, into text
- Smart response
- Restriction: This feature is part of a generative AI feature set available to customers with the requisite subscription agreement. Once the agreement is signed, feature entitlement is automatic after 5-7 business days. Activation may require additional configuration work. For more information, contact Medallia Support.Type of Feedback response that generates personalized and contextually appropriate response using generative AI
- Text Analytics
- Omnichannel sentiment analysis, which identifies the sentiment associated with individual phrases in customer feedback
CoreText model processing and development
Text Analytics and Action Intelligence use a proprietary multi-lingual, text-understanding system called CoreText, which turns every phrase (regardless of language) into a vector with one key property. The same meaning produces the same vector. Medallia uses these vectors to train models that predict properties of new text data, such as its sentiment, or whether the text is a suggestion. Medallia AI does not use CoreText models to generate text.
A consequence of this approach is that examples in one language can make the model better in all languages. This is called transfer learning, because the benefit of learning in one language transfers to others. This means that every model Medallia develops is inherently multilingual.
CoreText uses an emerging deep learning technique in language understanding called sentence encoding. CoreText is similar to systems such as BERT (from Google) and LASER (from Meta), and is trained using millions of pieces of text across all supported languages, from many different sources, including social reviews, feedback, news, and so on.
To construct a model, Medallia curates thousands of examples from human experts, defines a set of measures of performance, and identifies rough thresholds on these metrics for the model to be useful. These seed examples and intermediate models help retrieve a much larger set of similar examples (typically hundreds of thousands to millions) from our collection of feedback data spanning many industries and lines of business. Medallia then uses this much larger set of data to develop production models, following an iterative process of training and human feedback until achieving an acceptable level of performance. When a feature that uses a model becomes available in Medallia products, Medallia uses anonymized feedback from the field to drive continued improvement in the models.
Generative AI model training, data transparency, and user feedback
Medallia generative AI products use large language models trained on Medallia experience data. Medallia models are not trained with client data.
When AI-generated content or data is displayed in Experience Cloud, the generative AI icon displays with the module title or data, and a text disclaimer displays in the module's footer or after the relevant content.
When all fields in a module are populated with AI-generated data, the generative AI icon displays with the module title.
When a text field contains AI-generated data, the generative AI icon displays next to that text field's label, and the disclaimer displays after the relevant content. When only some text fields in a module are populated with AI-generated data, the icon is displayed on each text field but is not displayed with the module title.
Root Cause Assist, Themes with generative AI, Intelligent Summaries for conversational data, and Intelligent Summaries for Text Analytics include functionality that prompts the user for feedback on generated text.
For more information, see each feature's relevant section:
-
Results feedback for Root Cause Assist
-
AI-generated data transparency and feedback for Themes with generative AI
-
Model data, feedback, and training for Intelligent Summaries for conversational data
-
Summary traceability and feedback for Intelligent Summaries for Text Analytics
Medallia Speech transcription models
Speech uses acoustic and language models to convert speech to text. For more information, see Language models.
