Pathlight CEO explains how its AI agents will perform customer research 24/7
Customer insights platform Pathlight plans to rely on AI agents to extract strategic insights from large amounts of customer conversations in a way no human could manage alone, it announced this week. However, bringing this ambitious agent-powered vision to life posed technical challenges that required the Palo Alto based company to build custom infrastructure from the ground up.
In an exclusive discussion with VentureBeat, Pathlight CEO and co-founder Alexander Kvamme delved into the advantages and hurdles of designing an AI system capable of tackling such immense analytical jobs at scale. According to Kvamme, while executives are eager to understand pressing customer issues, dedicating the resources to deeply investigate every interaction simply isn’t feasible as businesses grow larger.
“One of the reasons why startups are so successful is they’re so close to their customers. They can move so quickly,” said Kvamme. “But as the company scales and becomes an enterprise, there’s just no possible way for you to review all that information.” To fill this gap, Pathlight set out to develop “24/7 research teams” that could monitor conversations without the constraints of constant human data collection.
While the deployment of AI agents provides Pathlight with a competitive differentiator, its customers will interact with familiar software interfaces to unlock these new powers.
Within Pathlight’s admin panel, executives can spin up “insight streams” — focused agents trained on specific analytical directives like understanding product issues or offering opportunities to try new strategies.
Pathlight is not alone in this approach. Using multiple generative models, working in harmony to produce results is an emerging element of the still-young AI sector.
Earlier this month, Microsoft announced AutoGen, which is “a framework for simplifying the orchestration, optimization, and automation of LLM workflows.” As well, new AI labs like Imbue focus on the research and development of cooperative foundation models which will eventually be able to learn, adapt and reason.
AI Agents will do jobs humans shouldn’t
Kvamme outlines the obvious: asking a human to sit at a desk and listen to every single customer interaction to aggregate individual insights is not a realistic proposition. Instead, the company’s agents take directives to actively analyze conversations.
“The way to think about agents and the way to think about insight streams, is to think about jobs that we would never be able to hire someone for,” Kvamme explained.
The agents don’t work alone, though. Kvamme described a hierarchy where agents actively flag insights. Then parent agents consolidate feedback into coherent summaries. This then equips company executives to make informed decisions and answer those burning questions which might have been impossible before.
“What we have found is every single executive has a series of questions in their head that they don’t have answers to that keeps them up at night,” said Kvamme.
During the interview, Kvamme provided a demo of Pathlight’s AI agent dashboard. He walked through how the system actively analyzes customer conversations in real-time.
Kvamme showed how calls and messages come into the platform, are handled by a human customer support specialist, and are processed by AI. Summarization, sentiment analysis, and other insights are automatically added. Perhaps most importantly, the system flags key themes and issues for agents — and ultimately, human managers and executives — to review.
In the demo account, themes like “order placement inquiries” were displayed. When selected, executives could see the reflections and insights flagged by agents. For example, one reflection noted an issue with “incorrect package delivery by FedEx.”
Kvamme emphasized this level of granular insight would be nearly impossible for a human to glean without AI assistance. AI agents will allow business leaders to have access to the full context and memory across all conversations, he explained.
Early AI agents need custom integration strategies
Bringing such a solution online demanded building custom infrastructure from scratch, however.
You can’t just plug massive datasets containing an enormous amount of customer interactions into existing AI tools like ChatGPT, Kvamme explained. The scale and technical needs required Pathlight to develop its own backend systems to handle the new workload demands.
“The state of the industry is such that we’ve had to build all of our infrastructure to support all this, but we’re not happy about it,” said Kvamme.
Though AI promotes new business opportunities, Kvamme acknowledges agent technology isn’t ready to fully replace human judgment and decision making just yet. For now, Pathlight’s passive analysis drives value by problems no team could feasibly handle alone through constant monitoring of conversations.
Moving forward, Pathlight aims to introduce limited automated corrective actions if agent networks detect systemic issues requiring immediate response, like adjusting misleading marketing campaigns. In the meantime, supervision remains crucial to ensure AI augments rather than replaces human oversight.
Through continually developing custom AI infrastructure and its iterative agent frameworks behind the scenes, Pathlight ensures the intelligence of machines expands key facets of customer understanding far above what’s humanly possible. Its agents take on analytical responsibilities no team could achieve to fuel critical business conversations.