NextLM Inc., which makes software that uses behavioral analysis and custom artificial intelligence models to help sales professionals identify and rank individual sales leads, is bringing its AI Prospecting Agent to Google LLC’s Cloud Marketplace and Gemini Enterprise.

The New York-based startup says its agent analyzes behavioral signals, identifies individuals it believes are researching a customer’s product and shows sellers the activity behind each recommendation. Customers can purchase the service using existing Google Cloud spending commitments and invoke the agent within Gemini Enterprise.

The distinction between company-level and individual-level intent is central to NextLM’s pitch. Business-to-business sellers typically have little visibility into who is researching their company and market. That makes prospecting a fishing expedition as salespeople try to identify the people behind a visitor’s IP address.

NextLM says its data engine processes 35 billion behavioral signals daily and spans more than 300 million companies and 370 million professional profiles. Chief Executive Chris Anzalone tol SiliconANGLE the company draws on advertising platforms and public databases to connect activity with individuals. Its model looks for combinations of behavior that may indicate an impending purchase.

“It’s not the specifics of the signals. It’s the way they interrelate with each other,” Anzalone said. Someone browsing a real estate site, for example, may simply be curious. Research into schools and restaurants in the same area provides stronger evidence of a likely home buyer.

The agent presents only prospects in the top 10% of its scores, Anzalone said. Each customer also gets a private adaptation of NextLM’s model that learns from its own sales wins and losses. The company says it doesn’t use customers’ deal data to retrain the shared base model.

Custom model

The service uses Savant, a lightweight AI model built on top of Nvidia Corp.’s Nemotron architecture that NextLM has fine-tuned for prospect scoring. NextLM says it trained the model on Nvidia DGX Spark systems and serves it on Nvidia A100 graphics processing units hosted on Google Cloud. Sellers can use the agent directly in Gemini Enterprise or feed its results into workflows built with Google’s Agent Designer.

In a company-published benchmark, NextLM reported that Savant delivered a 2.45-fold improvement in capturing eventual buyers among its highest-scored prospects compared with nine general-purpose AI models. It also reported a scoring cost of between 3 cents and 11 cents per thousand prospects when running on an owned GPU, which it said is an order of magnitude less than frontier models.

Anzalone said early customer results suggest that about one in four recommended prospects either closes or advances to an open opportunity that appears likely to close within 60 to 90 days. That measure combines completed sales with opportunities still in progress, and Anzalone said NextLM has not amassed enough data to assess how consistently the result holds across all customers.

Identifying named people from online behavior naturally raises questions about accuracy and privacy. Anzalone acknowledged that NextLM cannot match every person to the activity it observes. “There’s lots of people we can’t identify,” he said. “The ones we can identify with a high degree of certainty are the ones that we surface in the platform.”

NextLM applies an internal compliance checklist, relies on assurances from its data partners about consent and excludes users covered by Europe’s General Data Protection Regulation. The company only passes along the name of prospects who demonstrate a 97% likelihood to buy based on its scoring criteria.

The company’s lowest-priced plan starts at $599 a month without a long-term contract, with credits consumed as customers request prospects and related research.

Anzalone said Google integration makes NextLM easier to obtain and use within a broader workflow. While the company is open to working with other cloud providers, “With how good Google has been to us, we don’t have any intentions to move our infrastructure elsewhere,” he said.

Image: SiliconANGLE/Microsoft Designer