Unify GTM Growth System using OpenAI o3, GPT-4.1, and CUA

Unify has developed an AI-driven go-to-market (GTM) system that treats sales growth as an engineering problem, utilizing OpenAI o3, GPT-4.1, and the Computer-Using Agent (CUA) to automate prospecting and hyper-personalized messaging. This system currently generates 30% of Unify's own pipeline and supports hundreds of millions in pipeline for its customers annually.

Transforming GTM into a Scalable Search Problem

Unify redefines the go-to-market process as a search problem over unstructured, semantically rich data rather than a traditional acquisition problem. By replacing manual research and outreach with an agentic architecture, GTM teams can identify companies and individuals with specific problems their product can solve at a scale and speed previously impossible with manual labor.

Unify's AI product stack consists of three primary components:

  • The Observation Model: Powered by OpenAI o3, this model continuously researches total addressable markets to detect high-signal events, such as stack changes or new hires, using multi-agent workflows to surface strategic insights.
  • The Research Agent: This component handles open-ended questions and outbound copy generation. It utilizes GPT-4.1 for planning, CUA for dynamic browsing, and GPT-4o for synthesis.
  • Copywriting: Using agentic research and GPT-4o, this layer transforms surfaced data into hyper-personalized email drafts tailored to specific leads.

Strategic Model Allocation for GTM Tasks

Unify optimizes its system by pairing specific OpenAI models with the tasks they are best suited for, focusing on reasoning, tool use, and interaction capabilities.

OpenAI o3 for High-Signal Detection

Unify deployed OpenAI o3 into its Observation Model after validating its ability to reason through complex upstream decisions. The model is specifically used for two- to three-turn reasoning, which powers the early-stage logic required for signal detection.

GPT-4.1 and CUA for Research and Planning

GPT-4.1 and the Computer-Using Agent (CUA) are integrated into the product's decisioning layer to unlock research and planning tasks. Specifically, CUA enables UI-level interactions—such as navigating Trust and Safety pages or review sites—that static scraping cannot support.

GPT-4o for Synthesis and Classification

GPT-4o serves as the default model for synthesis and reply classification due to its fluency and support for structured outputs.

Evaluating Reasoning Quality in Real-World Scenarios

Unify evaluates models based on reasoning quality in real-world GTM scenarios rather than relying solely on accuracy or latency. This approach is critical for upstream stages like directive planning and signal classification, where the initial model output determines the success of all subsequent steps.

"Reasoning quality really matters in those early steps, and we needed models that could perform near human-level reasoning, understand nuance, and adapt across diverse use cases." — Kunal Rai, Software Engineer at Unify

By benchmarking OpenAI o3 against open-source alternatives, Unify determined that o3's superior reasoning capabilities were essential for the complex logic required in the Observation Model.

Sources