Accomy
简体
EN简体繁體
返回部落格

AI in Corporate Travel: Hype vs What's Actually Shipping

Most AI in corporate travel is still slideware. Here's what actually books, approves and reconciles trips today — and what's worth waiting for in 2026.

2026年8月24日Ervin Loke

Every travel-tech vendor now has an "AI agent." The pitch is autonomous booking, self-servicing trips, and expense reports that file themselves. Then you talk to the people running actual travel programmes: in a March 2026 GBTA survey, 58% of travel buyers said AI has had little or no impact on their programme to date (GBTA, via Business Travel Executive, 2026). That gap — between the marketing and the desk — is the real story of AI in corporate travel right now. This guide separates what's genuinely shipping from what's still a slide deck, and what a buyer should realistically expect. It matches what we hear from buyers: plenty of curiosity, far fewer who have actually changed how a trip gets booked.

It's written for the ops, finance, or travel-manager audience trying to cut through the noise before making a decision.

Key Takeaways

• There's a wide gap between AI interest and AI impact: buyers want it, but most say it hasn't changed their programme yet (GBTA, 2026).

• What's genuinely shipping: AI support assistants, expense/receipt automation, and assisted disruption rebooking.

• What's still mostly hype: fully autonomous booking. Today's "agents" keep a human in the loop, and vendors design them that way on purpose.

• For a buyer, the useful question isn't "do you have AI?" — it's "which functions are genuinely new, and what stays under human control?"

The gap between the pitch and the programme

Start with the honest picture, because it reframes everything else. Interest in AI is genuinely high — in the same GBTA research, buyers wanted AI for policy recommendations (95%), automated disruption rebooking (89%), traveller support (85%), and conversational booking (83%). But wanting isn't having: only around 14% currently use AI in their programmes (up 8 points year on year), and a majority report little tangible impact so far (GBTA, 2025–2026). GBTA attributes the lag to unconsolidated data and weak cross-region visibility — AI can't do much with spend it can't see.

So the demand is real and the deployment is early. That's the frame to judge any "AI travel" claim against.

What's actually shipping today

Plenty is genuinely live — it just isn't the autonomous agent the ads imply. The mature, in-market categories:

  • AI support assistants. Navan's assistant "Ava" has run at scale since 2023, handling support chats, cancellations, seat changes, and disruption resolution (Navan, 2025). This is the most proven category: AI answering and resolving common travel-support requests, with humans for the rest.

  • Expense and receipt automation. SAP Concur's Receipt Analysis Agent reached general availability in late 2025, filling missing fields from receipt content and flagging anomalies a human approver would catch; Concur also added AI-generated-receipt detection and anomaly/fraud tools (SAP Concur / SAPinsider, 2025). Expense is where AI is quietly delivering real time savings today.

  • Assisted disruption rebooking. Proactively spotting a disruption and offering a rebooking is increasingly live (Ava does it today), though buyers still rank fully automated disruption handling as a wanted future capability — so treat it as shipping-but-maturing.

  • Conversational booking (rolling out). In March 2026, Amex GBT announced a conversational "Egencia AI" that books and manages travel through chat inside tools like Microsoft Teams, with a Concur Expense integration — explicitly designed with humans in the loop, not full autonomy (Amex GBT, via Skift, March 2026).

A note on status: some capabilities are live and proven (Ava, Concur's receipt agent), others are announced and rolling out (Egencia AI), and others still — Amadeus's Cytric AI Assistant among them — are positioned as maturing rather than fully shipped. When a vendor demos an agent, the fair question is which bucket it's in.

Vendor / feature

What it does

Status (as of mid-2026)

Navan "Ava"

AI support assistant — chats, changes, disruption resolution

Live since 2023

SAP Concur Receipt Analysis Agent

Expense/receipt automation + anomaly and fraud flags

Live (GA late 2025)

Amex GBT "Egencia AI"

Conversational booking in chat tools, humans in the loop

Announced / rolling out (2026)

Amadeus "Cytric AI Assistant"

Conversational travel assistant

Maturing / not fully shipped

What's still mostly hype

Three claims deserve more scepticism than they usually get.

"Autonomous" is really "assisted." The headline promise is an agent that books and manages travel without you. The reality is a human in the loop almost everywhere. Skift's research found only about 2% of people are willing to give an AI tool full autonomy over bookings without oversight (Skift, 2026), and vendors build to that — Egencia's agent keeps humans in the loop by design. Analysts describe a staged path (augmentation → supervised automation → managed autonomy), not a switch that's already been flipped (McKinsey; Sabre, 2025–2026).

Some "AI" is relabelled automation. Receipt OCR, rules-based policy checks, and anomaly flags existed long before the "agent" branding. Some of what's now marketed as AI is genuinely new LLM capability; some is last year's automation with a new name. The buyer's job is to tell which is which.

Accuracy and trust are real constraints in a compliance setting. Travel and expense is a domain where a confident wrong answer costs money or breaks policy. It's telling that Navan markets "built-in safeguards" and prioritising "zero critical hallucinations" — that language exists because the risk is real. SAP Concur's work on detecting AI-generated fake receipts shows the double edge: AI is simultaneously a new control and a new threat vector. Data quality and trust, not model cleverness, are the current ceiling.

What should a buyer actually expect from "AI" today?

Set expectations to the evidence, not the demo. Realistically, in 2026 a corporate buyer can expect AI to:

  • Take load off support and expense — resolve common support requests, auto-populate and audit expenses, flag anomalies. This is where the value is real today.

  • Assist, not replace, booking — conversational and policy-aware help, with a human confirming anything consequential.

  • Depend entirely on your data — AI does little with travel spend scattered across tools it can't see; consolidation comes first, intelligence second.

And three questions worth asking any vendor: Which of these functions are genuinely new AI, versus renamed automation? What stays under human control? And what happens when the model is wrong — the answers separate a serious AI story from a marketing one.

We're candid about our own position here: Accomy today leads with a human support layer — real people on the channels travellers use — with AI-assisted booking on our roadmap rather than shipping. In a field this full of overclaiming, we'd rather name what's live than market what isn't. The broader point stands regardless of vendor: in 2026, the right posture on AI in corporate travel is interested, not credulous.

Where this is actually heading

The direction is clear even if the timeline isn't. Analysts describe AI in travel moving through stages rather than arriving all at once: first augmentation (AI helps a human who decides), then supervised automation (AI acts, a human approves), and eventually managed autonomy (AI acts within tight guardrails, humans handle the exceptions) — a path both McKinsey and Sabre lay out (2025–2026). Corporate travel is arguably better suited to this than consumer travel, because the guardrails already exist: policy, approval thresholds, and preferred vendors give an agent clear rules to act within.

Two things will decide how fast it moves. The first is data — AI is only as useful as the spend and itinerary data it can see, which is why consolidation keeps coming up as the precondition. The second is trust, earned slowly through accuracy in a domain where a wrong answer carries a cost. The sensible read for a buyer: expect the "assist" stage to get materially better over the next year or two, and treat any pitch of full autonomy today as running ahead of both the technology and the appetite for it.

Frequently Asked Questions

Is AI actually used in corporate travel yet?

Interest is high but deployment is early. Around 14% of travel programmes currently use AI (up 8 points year on year), and in a March 2026 GBTA survey 58% of buyers said AI had had little or no impact on their programme so far. The most-used live capabilities are AI support assistants and expense/receipt automation; more ambitious "autonomous" use cases remain mostly aspirational.

What AI features are genuinely available today?

The proven, in-market categories are: AI support assistants (e.g. Navan's "Ava," running since 2023 for support, changes, and disruption resolution); expense and receipt automation (e.g. SAP Concur's Receipt Analysis Agent, generally available from late 2025, plus AI fraud/anomaly detection); and assisted disruption rebooking. Conversational booking assistants (e.g. Amex GBT's Egencia AI, announced March 2026) are rolling out, with humans in the loop.

Can AI book business travel autonomously?

Not really, and not by design. Today's travel "agents" keep a human in the loop — vendors build them that way, and buyers want it that way (only about 2% are comfortable with full AI autonomy over bookings, per Skift, 2026). Analysts describe a staged path from augmentation to supervised automation to eventual managed autonomy, so autonomous booking is a direction of travel, not a current reality.

Is "AI" in travel tools just rebranded automation?

Sometimes. Receipt OCR, rules-based policy enforcement, and anomaly flags predate the current "agent" branding, so some marketed "AI" is genuinely new large-language-model capability and some is renamed existing automation. The useful buyer question is to ask a vendor which specific functions are genuinely new AI versus longstanding automation with a new label.

What should a company look for in AI travel tools?

Set expectations to the evidence: expect AI to reduce support and expense load and to assist (not replace) booking, and remember it depends entirely on consolidated data. Ask three questions of any vendor — which functions are genuinely new AI, what stays under human control, and what happens when the model is wrong. In a compliance-sensitive domain like travel and expense, guardrails and human oversight matter more than autonomy.

Quick Takeaways

  • Interest in AI is high; impact so far is low — 58% of buyers report little/no change to their programme (GBTA, 2026).

  • Genuinely shipping: AI support assistants, expense/receipt automation, assisted disruption rebooking.

  • Mostly hype: fully autonomous booking — today's agents keep a human in the loop by design.

  • AI depends on consolidated data; without one view of spend, there's little for it to work on.

  • Buyer's test: which functions are genuinely new AI, what stays human-controlled, and what happens when it's wrong?

About Accomy — Accomy is an APAC-first travel operating system for SME and mid-market companies. Today it leads with a human support layer; AI-assisted booking is on the roadmap. More at About Accomy.

Related reading:

Sources (retrieved 2026-07-13)

立即開始

親眼看看這個網路與它的覆蓋。

預約示範