Navan Deploys NVIDIA Model to Power AI Travel Agents in 3 Weeks

Navan Launches NVIDIA-Powered AI Travel Agents

Navan has deployed NVIDIA Nemotron 3.5 Super VL to power several of its AI travel agents in production, the business travel and expense platform announced Thursday from its Palo Alto headquarters1

1

. The post-deployment announcement marks a remarkably swift implementation timeline, with Navan training and deploying the NVIDIA model just 3 weeks after receiving early access2

2

. This rapid turnaround demonstrates the platform's technical capability and commitment to integrating cutting-edge AI technology into customer-facing operations.

Superior Performance Through Fine-Tuning

Before deployment, Navan conducted rigorous internal evaluations of NVIDIA Nemotron 3.5 Super VL against its existing production baseline and competing models1

1

. The company used consistent prompts, tools, travel scenarios and criteria to ensure fair comparison. After fine-tuning with NVIDIA NeMo AutoModel VLM LoRA training recipe, the fine-tuned model delivered the strongest overall performance, matching or exceeding competing models on six of seven measures while achieving the lowest latency2

2

. The model is now operational and handling customer interactions at what Navan described as lower cost, providing greater value to users while maintaining high-quality service standards.

Strategic Self-Hosting for Control

Navan has chosen to self-host the NVIDIA model rather than rely on third-party infrastructure, a decision that reflects the company's priorities around performance optimization, cost management, and reliability1

1

. This self-hosting approach allows Navan to maintain direct control over how the model operates in production environments and adapt it specifically to travel-related tasks. "Building reliable agentic travel systems involves much more than choosing a model, but continuously iterating to improve and solve real-world problems," said Ilan Twig, co-founder and CTO of Navan1

1

. His statement underscores that successful AI implementation requires ongoing refinement beyond initial deployment.

Open-Model Strategy for Travel AI

The deployment exemplifies Navan's open-model strategy, which involves evaluating models for specific travel tasks, adapting them with travel-specific AI expertise, and self-hosting them for optimal control2

2

. Navan combines both open and closed models to build AI travel agents designed to understand traveler preferences, follow corporate policies, and execute booking tasks efficiently1

1

. This hybrid approach gives the company flexibility to select the best-performing models for different use cases rather than committing to a single vendor or technology. The strategy positions Navan to rapidly adopt emerging AI capabilities while maintaining the specialized knowledge required for complex travel scenarios.

What This Means for Business Travel

This deployment signals a broader shift in how business travel and expense platforms are leveraging AI to automate complex workflows. Watch for increased competition among travel platforms to deploy similar agentic systems that can handle multi-step booking processes, policy compliance checks, and personalized recommendations. The emphasis on cost reduction through self-hosted models may prompt other enterprise software providers to reconsider their AI infrastructure strategies. As these AI travel agents mature, expect to see expanded capabilities beyond basic booking to include itinerary optimization, real-time rebooking during disruptions, and proactive expense reconciliation. The success of Navan's rapid deployment timeline could accelerate the pace at which competitors bring their own AI agents to market.