AI Agents for Corporate Travel and Expense Teams: What GBTA's 2026 Data Shows Is Actually Live

Corporate travel management has always been a workflow problem disguised as a booking problem: search inventory, apply policy, get approval, file the expense, reconcile it against a receipt weeks later. That's exactly the kind of multi-step, rules-bound process AI agents are best suited for — and the travel industry's own trade association now has the numbers to show it's happening, not just being discussed.

What GBTA's own polling actually found

The Global Business Travel Association's latest industry poll, covered directly on GBTA's site, put real numbers on agentic AI adoption in the industry for the first time: 49% of suppliers and travel management company (TMC) professionals and 33% of corporate travel buyers report their companies are already experimenting with autonomous AI agents — up sharply from a GBTA poll just a year earlier, in which only 34% of buyers said they even planned to apply AI "in significant ways" in 2025.

The most concrete number in the poll is about expense, not booking: 51% of buyer respondents are planning to use agentic AI for expense reconciliation specifically — the single most-cited planned use case among buyers. On the booking side, the top applications suppliers and TMCs report are customer service, traveler personalization, and automated itinerary planning.

The infrastructure is showing up at the same pace as the intent

This isn't just survey sentiment. At the GBTA Convention 2026, enterprise travel-and-expense platform TripGain announced a new MCP (Model Context Protocol) server specifically built to let AI assistants execute travel and expense workflows directly — booking policy-compliant travel, filing employee expenses, managing vendor expenses, and approving pending requests "through natural conversation," in the company's words, while its existing policy engine, approval workflows, and financial controls stay in force behind the scenes. The point of building on MCP, an open standard, is that enterprises don't have to hand-build individual integrations for each AI assistant that wants to touch the travel system — the same standardization pattern other back-office software categories have been racing toward all year.

GBTA's own AI Pathways initiative backs this up from the buyer-education side: the association has shipped agents to help travel managers build supplier RFPs and translate program data into leadership-ready narratives, explicitly framed as "practical starting points" that keep a human reviewing the final output rather than fully autonomous decision-making.

Adoption is real, but so is the caveat

GBTA's poll didn't just report enthusiasm — it flagged data privacy concerns as a real drag on faster agentic AI adoption in the same breath as reporting the adoption numbers above, and a separate GBTA/Radisson Hotel Group study on hotel-program sourcing found a similar pattern: only about a third (32%) of programs actually used AI in their most recent hotel RFP cycle, even though more than two-thirds (69%) expect to use it in the next one. That gap between "expect to" and "actually did" tracks with a pattern seen across other back-office categories this blog has covered — intent runs well ahead of production use, at least for now.

What this means for travel and expense teams going forward

The travel-and-expense function has three characteristics that make it a strong early candidate for agentic AI: it's high in transaction volume, governed by clear written policy, and ends in a reconciliation step that's tedious for humans and mechanical enough for a well-scoped agent to handle end-to-end. GBTA's own numbers — a third of buyers already experimenting, and over half planning to hand expense reconciliation to agents specifically — suggest travel teams are further along this curve than the "will we ever trust AI with our expense reports" framing usually assumes. The teams moving fastest aren't the ones granting agents blanket access to book and spend; they're the ones, like TripGain's approach shows, keeping the existing policy engine and approval workflow in force and simply giving an AI assistant a standardized way to operate inside it.