Building AI to drive B2B marketing through the Integration of Content Intelligence, Recommendation and Conversational Agents

Building AI to drive B2B marketing through the Integration of Content Intelligence, Recommendation and Conversational Agents

The Challenge

A leading SaaS platform for B2B marketing in the US, aims to nurture prospects, improve content performance by delivering intelligent content experiences and virtual events across the buyer’s journey and improve win rates. Tatras Data was asked to develop an AI-based solution using unstructured content data, user content consumption data and CRM data to provide personalised content recommendation to accelerate the user journey. Tatras was also tasked with providing deep insights, to marketers, into content consumption and to correlate it with consumer firmographics and gauge buyer intent. Using this buyer intent, a chat agent was to be developed that could engage in a conversation with the buyer, personalized to their intent and stage of the buyer journey.

Hypothesis

  • NLP techniques can help organize large B2B marketing corpora and make it easy to navigate for marketers in large corporations.
  • Visitor Intent and Stage of Funnel can be identified through content consumption analysis leading to personalized pathways for visitors.
  • A personalized conversation through a chat agents, based on the current stage of the user and their intent, is achievable with current LLM technology.

Execution

  • Tokenization, remove stop-words and special characters, Lemmatize data, Keyphrases Extraction & utilize other NLP-based cleaning techniques.
  • Topic Modelling along with a novel Topic seeding to generate topics in line with marketer needs Autoencoder based vectorization of CRM data (Salesforce)
  • Hybrid Recommender System (Collaborative and Content based) implemented with Explainability feature.
  • Automated tagging of content using Natural Language Understanding.
  • Multi-agent GenAI based chatbot with question recommendation based on User profile and intent.
  • Hybrid Graph RAG and Vector RAG with cross-encoder based reranking for context generation.
  • Action identification using LLMs to bring users in the loop for closing sales.

Outcomes

  • Content Intelligence based on Topic and Taxonomies made content consumption analysis easier to comprehend and to identify gaps in marketing material.
  • Improved stickiness on client websites through recommendation of relevant content.
  • Chat based interface integrated with recommendation technology to provide a personalized conversation based on stage of the customer in the funnel.