Key Takeaways

  • Trajectory raised $45 million to help subject-matter experts train AI models through feedback rather than machine-learning code.
  • The 18-person company is targeting a persistent enterprise skills gap as AI adoption expands beyond technical teams.
  • Trajectory’s opportunity will depend on combining accessible customization with data controls, evaluation, and human oversight.

Trajectory has raised $45 million to expand an approach that lets subject-matter experts train and customize AI models without deep machine-learning expertise. The 18-person company is focused on models that learn from user feedback and improve at specialized tasks over time.

That pitch addresses a practical gap in enterprise AI. Businesses may have capable foundation models and extensive internal data, but the employees who understand a particular workflow, regulation, customer interaction, or technical process are rarely the people building the underlying AI systems.

Enterprise research from S&P Global shows the scale of that disconnect. Average AI adoption across 38 enterprise use cases reached 50%, yet only 22% of projects targeted fully autonomous outcomes. Most deployments remain augmentative, with AI assisting employees rather than operating independently.

Skills are part of the explanation. Some 59% of organizations report moderate or severe gaps in machine-learning and AI development that affect their initiatives. Trajectory is effectively betting that companies can narrow that gap by moving part of the model-development process closer to the people who already know the job.

While business expertise differs from AI engineering, it provides critical training input. A claims specialist can identify a flawed insurance assessment. A compliance professional can flag an incomplete review. A manufacturing engineer may recognize an impractical recommendation immediately. Capturing that feedback in a form a model can learn from could shorten the distance between a promising prototype and a useful production system.

Trajectory enters a category that overlaps with no-code and low-code AI, model customization, and enterprise machine-learning platforms. DataRobot and Google Vertex AI are other examples of the broader effort to make model development available to more users while preserving centralized controls.

The distinction will come down to execution. How easily can experts translate their judgment into reliable model behavior? A simple interface does not remove the underlying challenges of representative data, measurable performance, access controls, and changing business conditions.

Those concerns are substantial. S&P Global enterprise respondents identify data privacy and security (51%), response accuracy (46%), and data quality (38%) as leading limitations on generative AI models. Making customization easier could increase experimentation, but it could also multiply poorly documented models if governance is treated as an afterthought.

The NIST Generative AI Profile offers one useful reference point. Published in 2024, it identifies 12 generative AI risk categories and organizes mitigation work around governance, mapping, measurement, and management. For user-built models, that translates into questions about data provenance, privacy, testing, human review, and dependencies on third-party components.

Other guidance points in a similar direction. The OECD AI Principles emphasize transparency, accountability, robustness, and responsible stewardship, while the Stanford Institute for Human-Centered AI tracks the widening gap between rapid AI adoption and the operational work involved in deploying systems responsibly.

That said, governance cannot become so cumbersome that it recreates the bottleneck Trajectory is trying to remove. Enterprises will likely need graduated controls: low-risk internal experiments may receive lighter review, while models influencing financial, legal, medical, or employment decisions could face more rigorous evaluation and approval.

The $45 million raise gives Trajectory more room to develop its product and compete in a crowded enterprise market. Funding alone does not settle whether expert-led training will produce dependable models at scale. Still, the underlying demand is clear. Companies want AI grounded in their own processes, and many do not have enough machine-learning specialists to customize every application.

Trajectory’s larger test will be whether it can turn everyday professional feedback into measurable, auditable model improvement. If it can, the company may help shift enterprise AI development from a centralized technical function toward a more collaborative process involving the people who understand the work firsthand.