NextLM Puts Its AI Prospecting Agent in Google Cloud Marketplace and Gemini Enterprise

A sales team using Google Cloud can now buy NextLM’s AI Prospecting Agent through Google Cloud Marketplace or use it within Gemini Enterprise. NextLM says the agent does more than produce a list of companies: it names people who appear to be researching a seller’s product and shows the activity behind each lead.

From online activity to a person’s name

NextLM says it links behavior observed across the open web to named individuals. Each lead comes with the activity that flagged the person and a confidence score. The company says its system processes 35 billion behavioral signals a day and draws on records covering more than 300 million companies and 370 million professional profiles. Those figures describe the scale NextLM claims for its data operation; they do not, on their own, show how often a named lead turns out to be interested in buying.

A separate model for each customer

The agent runs on NVIDIA’s Nemotron model. NextLM says each customer gets a private version fine-tuned on that customer’s won and lost deals. Fine-tuning means adapting an existing model using examples from a particular business; here, the aim is to make lead scoring more relevant to that team rather than give every seller the same ranking. NextLM says it trains on NVIDIA DGX Spark systems and serves the agent on NVIDIA A100 GPUs on Google Cloud.

Distribution is the immediate change for buyers. Customers can procure the agent through their existing Google Cloud accounts and apply cloud-spend commitments toward it. In Gemini Enterprise, sellers can invoke it directly or pass its output into Agent Designer workflows for further work.

2.45×Reported lift in top-decile buyer capture

NextLM reports a 2.45-fold lift in top-decile buyer capture for its Savant system in its benchmark. The figure is a company-reported result, not an independently verified customer outcome.

What the benchmark can—and cannot—show

NextLM says it tested its fine-tuned Savant system against nine general-purpose language models on prospect scoring. It reports a cost of roughly $0.003 to $0.011 per thousand scores while running on a single owned GPU. That comparison addresses the cost of scoring prospects in NextLM’s test, not what a customer will pay to buy the agent through Google Cloud.

For a sales team, the consequential test is whether the evidence attached to a named person is useful enough to act on. A confidence score gives sellers a way to weigh a lead, but NextLM’s announcement offers company-run benchmark results rather than demonstrated results from customers using this new Google Cloud route.