For Anatoly Shilman, one of the biggest risks in building a startup occurs before a product is ever launched. It happens during early customer discovery, when founders rely on surveys, interviews, and research questions to validate an idea, often without realizing those questions may be shaped by cognitive bias — an unconscious mental shortcut that can lead to inaccurate thinking or decision-making.

After years of working as a serial entrepreneur and startup mentor, Shilman noticed a recurring issue: teams spending months building products based on feedback that was unintentionally influenced by the way questions were framed.

“Very few products fail because the technology doesn’t work,” said Shilman. “They fail because there was no market need. Often, that traces back to early research where founders heard what they expected to hear, rather than what the market was actually saying.”

That observation led Shilman and his team to create CogBias AI, a decision-intelligence startup developing tools to identify cognitive bias in written communication used for research, surveys, and customer discovery.

Turning Behavioral Science Into a Practical Tool

CogBias AI focuses on the earliest inputs that shape decisions: the questions themselves. The platform analyzes written prompts, such as survey questions or research language, to identify patterns associated with cognitive bias and suggest alternative phrasing aligned with the user’s intent.

During early testing, the team analyzed customer feedback surveys from large organizations and found that a significant portion of commonly used questions did not produce actionable insights because they were leading, ambiguous, or misaligned with the decision being made. In those cases, revising the questions changed the clarity and usefulness of responses.

According to Shilman, the platform is designed to reduce the time teams spend rewriting and reworking questions before sending them out, while improving confidence that the responses they receive reflect what respondents actually mean — stripping out the bias.

“Bias is often invisible, which is what makes it hard to catch,” he said. “The goal is to help teams see where judgment may be influencing outcomes before decisions are locked in.”

The approach draws from behavioral science and decision theory and is being developed using expert-informed models rather than relying solely on general-purpose AI outputs. The company is currently refining the platform based on early use cases rather than formal customer case studies.

The Role of the ATDC Ecosystem

CogBias AI began scaling with the Advanced Technology Development Center (ATDC), Georgia Tech’s statewide startup accelerator. Shilman, who previously served as a mentor with Georgia Tech’s former Flashpoint accelerator, said ATDC provided a structured environment to develop a complex idea that does not fit neatly into existing markets.

“ATDC gave us the space to test assumptions, pressure-test decisions, and work through uncertainty without rushing to force the product into a traditional mold,” he said.

The accelerator supports founders from across Georgia, including companies that are not based on licensed Georgia Tech intellectual property. For CogBias AI, participation in the ATDC community has provided access to coaching, peer feedback, and operational support during a stage when the company is still refining both its technology and positioning.

Looking Ahead

CogBias AI is preparing for its first formal press announcement while continuing to develop the platform ahead of a broader commercial release. Shilman said the focus remains on building a responsible tool that improves how organizations gather and interpret information before making high-stakes decisions.

Through ATDC and Georgia Tech’s broader entrepreneurship ecosystem, founder-led startups like CogBias AI illustrate how early-stage ideas can be supported as they move from observation to application, even before they are fully in market.