DialpadGPT: Generative AI built for business conversations

What is DialpadGPT?

DialpadGPT is Dialpad's proprietary large language model (LLM), fine-tuned on business conversations, that powers many of the AI features in Dialpad AI. Because it's trained for the language of sales calls, support conversations, and meetings, it can turn everyday conversations into summaries, action items, coaching cues, and customer satisfaction insights in real time.

Why a model built for business conversations

General-purpose LLMs are trained mostly on written text from across the internet. Business conversations are different: they're spoken, fast, full of industry jargon, and often captured through imperfect transcripts. A model fine-tuned on that kind of data can be better suited to the tasks teams rely on every day.

Built on original AI research

DialpadGPT and the features it powers are shaped by original research from Dialpad's in-house AI Research Lab. The team studies real-time communications intelligence, agentic AI systems, multilingual understanding, and responsible AI, and publishes findings at venues like ACL, EMNLP, and NAACL. A few papers connected to DialpadGPT's work:

Adapting LLMs to business conversations

• DACP: Domain-Adaptive Continual Pre-Training of Large Language Models for Phone Conversation Summarization (EMNLP 2025, NewSumm Workshop) explores how continued pre-training on phone conversation data can adapt LLMs for summarizing business calls.

• Domain Adaptive Continual Instruction Pre-Training via Reading Comprehension on Business Conversations (EMNLP 2025, Industry Track) studies a reading comprehension approach to teaching models the structure and language of business conversations.

Efficiency without sacrificing quality

• How Accurate Are LLMs at Multi-Question Answering on Conversational Transcripts? (EMNLP 2025, Industry Track) introduced a batch-prompting framework that lets AI answer multiple questions from a single conversation in one pass, cutting cost and latency, while fine-tuned smaller models matched GPT-4o's accuracy.

• Tiny Titans: Can Smaller Large Language Models Punch Above Their Weight in the Real World for Meeting Summarization? (NAACL 2024, Industry Track) examines whether smaller, more efficient LLMs can compete with larger models on real-world meeting summarization.

Real-time understanding of live conversations

• Topic Matching in the Wild: Benchmark and Lessons from Real-World ASR Transcripts (W-NUT at EMNLP 2026) benchmarked regex, zero-shot embeddings, and LLMs for recognizing customer topics in noisy, real-world contact center transcripts. Lightweight LLMs using natural-language topic descriptions delivered the strongest results.

• AI Coach Assist: An Automated Approach for Call Recommendation in Contact Centers for Agent Coaching (ACL 2023, Industry Track) explores how AI can identify the calls most useful for coaching contact center agents.

• Predicting Customer Satisfaction with Soft Labels for Ordinal Classification (ACL 2023, Industry Track) studies how to predict customer satisfaction from conversations, the kind of work behind AI CSAT.

Rigor and responsible AI

• LLM Evaluate: An Industry-Focused Evaluation Tool for Large Language Models (COLING 2025, Industry Track) describes an on-premise framework for testing LLMs with rigor and privacy, enabling reproducible, cost-efficient benchmarking across open and closed models.

• Data Anonymization for Privacy-Preserving Large Language Model Fine-Tuning on Call Transcripts (EACL 2024, Workshop on Computational Approaches to Language Data Pseudonymization) examines ways to anonymize call transcripts before they're used to fine-tune LLMs.

These papers are a sample of the published research from Dialpad's AI Research Lab, which also spans speech recognition, multilingual understanding, and agentic AI.