Cast Insights, a startup that’s building a real-time competitive intelligence layer for ephemeral speech, launched its platform today after closing on $4.5 million in a pre-seed funding round.

The investment was led by Abstract Ventures and saw participation from HF0, Village Global, Max Ventures, Embassy Ventures and Stratus Ventures.

The San Francisco-based startup, which emerged out of HF0 Residency, a startup program for repeat founders, uses artificial intelligence to index what it calls “ephemeral speech data” that can help institutional decision-makers and others to understand market trends, policy issues, geopolitical shifts and risk factors before a mainstream consensus emerges.

Co-founder and Chief Executive Otávio Costa Miranda believes that speech is the world’s largest untapped data source. There are millions of public statements and conversations in the world that are broadcast by TV programs, radio stations and through podcasts, which can help to shape public opinion and market sentiments. But this speech is difficult to track and record, and much of what’s said vanishes the moment the person saying it stops speaking.

This is why traditional forecasting models rely on lagging indicators: The tools to capture and analyze public speech simply haven’t existed before. But Miranda says these models are inadequate, based on incomplete, out-of-date information. “Most of what gets said in the world every day across radio stations, legislatures, podcasts and local broadcasts is ephemeral by nature,” he said. “It surfaces publicly, it influences people and then it vanishes.”

Cast Insights is trying to prevent these insights slipping through people’s fingers with an AI layer that continuously monitors thousands of global sources, including TV channels and radio stations, podcasters and more. It ingests spoken audio, transcribes what’s said, and then uses AI algorithms to identify the speakers, trace how narratives are formed and shaped, and then map how sentiment shifts in real time.

The startup demonstrated the power of its ephemeral speech intelligence layer in a case study monitoring signals from public discourse around the Strait of Hormuz, which is a key flashpoint at the center of the current conflict between the U.S. and Iran. Cast Insights tracked the issue from the moment it first popped up, showing how there were near-zero mentions before the conflict broke out. It quickly jumped to around 400 daily mentions, and peaked at 2,946 when the recent ceasefire between the U.S. and Iran collapsed.

According to Miranda, when the ceasefire was broken, Cast Insights illustrated how positive sentiment plummeted from 7.7% to 3.3% while tracking the market impact, such as surging war-risk insurance premiums for shipping companies. The platform also captured the rapid narrative shifts that occurred when U.S. officials proposed a 20% transit toll on all ships passing through the Strait, only to reverse that policy less than 24 hours later.

“We’ve always had this cultural obsession with forecasting the future, when the real problem is we barely understand the present,” Miranda said. “What Cast does is make the present legible in real time so that instead of guessing where things are going, our clients can see convergences forming and divergences emerging before they reach consensus. The signal is always there. It just slips through everyone’s fingers.”

While operating under the radar with early adopters, Cast Insights has already gained significant traction, processing more than 2.3 million hours of speech across 20 countries and 10 different languages. With the funding from today’s round, it’s planning to scale this to process 20 million hours, while monitoring 500,000 hours of live content each day. In doing this, it believes it will create what is by far the world’s biggest ephemeral speech dataset, covering major U.S. television broadcasts, radio stations and podcasts.

The startup was one of 10 to graduate from the latest HF0 cohort, compressing 12 months of business growth into just 12 weeks. HF0’s Evan Stites-Clayton said that when he first spoke to Miranda earlier this year, his reaction was one of disbelief that such a platform didn’t even exist at that time.

“We thought, ‘Wait, how does that not exist?’” he recalled. “Once Otávio showed us the platform’s features, like real-time alerts within 30 seconds of broadcast, speaker-level voice tracking and narrative curves across thousands of sources, it became very clear this team is building something the entire institutional world is going to depend on.”

Image: Cast Insights