Cloud communications provider Dialpad Inc. is continuing its aggressive rollout of artificial intelligence capabilities with today’s launch of DialpadGPT, which it describes as a domain-specific large language model that’s tailored to the needs of customer service, sales, and recruiting operations.

The company said the generative pretrained transformer offering is meant to address businesses’ concerns about scalability, accuracy, relevance, privacy and security when adopting generative AI.

Dialpad has positioned itself as the most AI-focused vendor in the call center market. Last spring it rolled out a slew of new features incorporating natural language processing automated transcription, performance monitoring and personalized guidance.

5 billion minutes

DialpadGPT was trained on more than 5 billion minutes of proprietary conversational data gathered over a five-year period. It’s described as a business-oriented model that uses conversational and messaging data across customer service, sales, recruiting and employee collaboration.

Those capabilities will be used as new features and to embellish existing services. A new GPT-powered automated recap feature will summarize the contents of a conversation immediately after the call has ended and alert callers to action items that were mentioned like plans to schedule a meeting. Generative AI will also be used to automatically label calls according to their content, determine how successful they were and generate analytics about popular topics and sentiment across a large volume of calls.

GPT technology will also be applied to existing capabilities including playbooks, which automatically assess whether a sales representative has covered all the talking points in a script and scorecards that rank the success of individual calls and the relative performance of agents.

“We already have scorecards but GPT will make them more accurate,” said Co-founder and Chief Technology Officer Brian Peterson. “We’re 80% accurate in detecting customer satisfaction using transcription analysis. GPT will make that better.”

Deep AI roots

Dialpad has been working with AI technologies like machine and deep learning for more than five years, “but it’s gone super-advanced in the last several months,” Peterson said. The company made the strategic decision to build and train its own models rather than rely on partnerships or commercially available services like OpenAI LLC’s ChatGPT.

“ChatGPT doesn’t work for us,” Peterson said. “The data isn’t relevant to support and sales calls, it hallucinates, there are privacy concerns and it isn’t real-time.” That last capability is important in applying generative technology to real-time conversation monitoring, which enables supervisors to intervene in a conversation that’s going badly while the call is ongoing.

Third-party services “aren’t fast enough,” Peterson said. “We were able to strip out overhead to monitor conversations in real time.”

Although the training data set for DialpadGPT spans a large number of customers, the company said privacy isn’t an issue. “One of the biggest reasons we wanted to build our own GPT was so we could control the data,” Peterson said. “All the data you’ve trusted us with is in our system and we have full retention and compliance with the AI model. There’s no chance of data leakages like there is in a public model.”

GPT capabilities will be provided to every customer starting in October as part of their subscriptions. Some features that benefit from GPT such as customer satisfaction and sales playbooks are carry an additional fee of $10 per user per month.

Photo: photopin