Travel and Finance Firms Reject LLM Plausibility in Favor of Deterministic Systems

Two of the world's largest consumer-facing enterprises are publicly distancing themselves from reliance on large language model outputs, choosing instead to anchor AI strategies in deterministic data pipelines and structured experimentation.

Expedia CEO Ariane Gorin told stakeholders that customers demand certainty over plausibility when booking travel, framing the company's operational data infrastructure as a competitive moat against conversational AI. "People want trust. They don't want an LLM that is plausible. They want something that is certain when it comes to travel," Gorin said. Expedia updates 65,000 property records and attributes daily, a cadence the company believes generative models cannot match without introducing hallucination risk.

Lloyds Banking Group and the University of Glasgow launched a four-year applied research program to evaluate agentic AI systems powered by LLMs within the bank's software and data engineering workflows. The collaboration embeds academic researchers directly with engineering teams serving 28 million customers, funds a PhD, a Masters by Research, and a postdoctoral role, and will run recurring experiments to measure output quality, development speed, and operational scaling. The partnership aims to produce evidence that guides responsible deployment rather than adopt generative tools at scale without validation.

The travel sector is fragmenting into distinct AI stack strategies. Some online travel agencies are investing in the model layer, training domain-specific LLMs on proprietary booking and pricing data. Others are prioritizing the orchestration layer, building runtime routing and prompt engineering pipelines across heterogeneous API providers. A third group is focusing on the product layer, embedding AI into customer-facing features with strict schema validation and API-level confirmations. A fourth infrastructure layer, focused on logging, provenance, and explainability, is emerging to satisfy regulators and partners.

(Luminai, a San Francisco-based AI workflow automation platform serving healthcare systems, raised $38 million in Series B funding led by Peak XV Partners, bringing total capital to $60 million. The company automates revenue cycle, prior authorization, and claims processing tasks by handling unstructured inputs and embedded clinical context across disconnected systems.)

The divergence reflects a broader tension in enterprise AI adoption. Generative models excel at natural language and planning but produce plausible-sounding errors without easy verification paths. In verticals where transactions entail payments, cancellations, and real-world logistics, system designers are choosing retrieval-augmented generation architectures that pair LLM reasoning with authoritative data stores. This approach reduces hallucination risk by grounding model outputs in verified inventory, strong schema validation, and explicit API confirmations.

The legal technology sector is also navigating multi-model strategies. Inquisita announced a generative AI-powered document analysis platform on March 30 that leverages multiple LLMs to perform semantic search, structured analysis, and privilege review without locking users into a single model provider. The platform aims to provide context and navigate document management at scale, reflecting a similar orchestration-layer bet seen in travel and finance.