How this category works
Every major travel and expense platform now has an AI story. That makes “does it have AI?” one of the least useful questions you can ask.
Navan has agentic AI across T&E. SAP Concur has Joule. Ramp has finance agents. Perk connects travel and spend data into assistants such as ChatGPT and Claude. Amex GBT and Direct Travel are adding conversational and predictive intelligence. ITS has AI Chat, Bob and Ana, and agents working behind the scenes.
The better question is: where does the AI stop?
The AI can explain a policy, surface a trip, answer a question, or summarize spend. Useful, but this is the baseline now.
It can suggest a flight, flag an out-of-policy option, recommend an approval action, or identify an expense that needs attention. Better — but a human still does the next step.
It can submit, approve, route, modify, book, reconcile, or otherwise complete a defined action after the user gives permission. This is where conversational AI starts changing the workflow rather than only the interface.
The AI can use context from several parts of the program and move through multiple steps: the trip, policy, card charge, receipt, expense, event, or traveler profile. This is where “one data model” becomes more than an architecture claim.
Some work should still move to a person. A good AI system should recognize uncertainty, preserve the context, and hand the traveler or program manager to a human without making them start over.
ITS is building across several of these layers at once.
The conversational entry point: ask across travel, card, expense, meetings, and the broader program without hunting through products.
A digital-human experience for people who would rather interact visually and by voice.
Automated workflows work behind the conversation to handle repetitive work and keep processes moving.
People remain part of the travel experience when an exception needs judgment or hands-on servicing.
“Show me the Chicago trip, the charges tied to it, and what still needs to be expensed.” If the answer requires three separate assistants, you have learned something.
Do not stop at “what is our meal policy?” Ask it to take the next appropriate action.
A changed flight. A hotel folio with an unexpected charge. A receipt that does not match. AI looks best on the happy path; exceptions reveal the operating model.
Agentic should not mean uncontrolled. You should know when the system can act autonomously, when it needs confirmation, and what it records for audit.
See whether the person receives the context or makes the traveler repeat everything.
Roadmaps move quickly in AI. Separate what is generally available from beta, pilot, announced, and planned functionality.
Most mature programs will use more than one.
Best when the organization wants travel and expense context available without exposing every workflow to another enterprise assistant.
Best when the company has already standardized on ChatGPT, Claude, Copilot, or Gemini and wants business data available where employees already work.
Best when the goal is deep automation of repeatable finance, audit, support, or reconciliation work.
Best when travel has enough exceptions that automation cannot be allowed to become the final support layer.
FAQ
It should not be. The useful distinction is whether the system can use context, decide what step comes next, take permitted actions, and continue through a workflow instead of only returning text.
There is no useful universal answer. Navan, SAP Concur, Ramp, Perk, Amex GBT, Direct Travel, BCD, Engine, and ITS are solving different parts of the problem. Test the workflows that matter to you.
That depends on your enterprise AI strategy, security model, and what actions the connection supports. External assistants are convenient; native assistants often have deeper first-party context. The best architecture may use both.
That is a governance decision, not a marketing one. High-cost changes, unusual policy exceptions, sensitive traveler issues, and ambiguous financial decisions often need confirmation or human review.
It changes what they spend time on. Routine questions and transactions can move to automation. Disruptions, judgment calls, complex ticketing, and unusual traveler needs are exactly where a good human handoff becomes more valuable.
Bring a real workflow. Ask. Act. Break it. Escalate it. Then decide whether the AI is actually removing work.
Category overview based on public vendor product information reviewed August 2026. AI capabilities change particularly quickly; verify current availability, permissions, data access, and action support directly with each vendor.