2026 Buyer Agenda: AI Prompt Engineering

Amazon CEO Andy Jassy, in a June interview with CNN, said AI “will change how we all work and live,” including “billions” of AI agents “across every company and in every imaginable field…. Many of these agents have yet to be built, but make no mistake, they’re coming, and coming fast.” 

Corporate travel management, of course, is not immune. During a BTN Hot Topics 2026 webinar, travel buyers emphasized that effective communication with AI—known as AI prompt engineering—is a crucial skill for travel managers who want to remain relevant.

“Learning prompts [and] learning how to interact with AI is chief,” said ServiceNow senior manager of global travel and expense Heather Allegrina. “Being able to ask questions of data and get answers without having to do downloads and pivots… that has been a gamechanger for us.” 

For travel managers looking to integrate AI into their programs, Allegrina recommended restructuring travel policy information it into a Q&A format “so an AI bot can read and answer questions… that will help when employees are asking native questions”.

At the GBTA Europe Conference in Hamburg in November, PredictX CEO Keesup Choe led an AI session titled 'Rise of the Agents', where he described prompt engineering as the foundation for mastering AI agents.

“Prompt engineering is how to communicate to large language models to get them to do what you want,” said Choe. “It’s more a language than a code.”

He shared the following techniques: 

Zero-shot Prompt Engineering

When you ask AI to handle a task without providing any prior examples, training or detailed instructions. “It’s like asking a colleague to write a report on a topic they’ve never covered before,” Choe explained. Clear instruction is therefore critical and should comprise: 

• Subject: State main topic or focus of task (e.g. hotel spend)
• Action verb: Use a clear command to direct the AI, such as ‘analyze’ or ‘list’ instead of simply saying “give me”
• Task description: Specify what needs to be done with the subject (e.g. provide key financial metrics)
• Context (optional): Add details like audience, purpose or scope (e.g. for a business analyst preparing a report)
• Tone/format: Define the tone and output format (e.g. bullet points, table)
• Constraints (optional): Include limits like word count or specific requirements (for the period 2024-2025)

For example, instead of simply writing “tell me about my hotel spend”, Choe suggested the following:

I need you to analyze our hotel spend and give me these metrics for YTD 2025 only: total spend with hotels, the total number of room nights. Then a table of the top 5 hotels by spend, show me the spend, number of room nights and the average daily rate.

“The more specialized AI gets, the more capable and better it is,” he said. While OpenAI’s Chat GPT model was “designed to do everything as well as possible, and to be creative,” Choe advised that in business, AI models should be constrained to be “as literal as possible”.

Prompt Stacking

Another technique that gives AI a list of step-by-step instructions to follow, like a recipe. Each step builds on the one before it, like stacking blocks to build a tower.

Example: Using hotel property spend data from 2204 and 2025, first provide a year-over-summary of total spend and room nights. Then, analyze how the average negotiated rate and average daily rate have changed over these years. Finally, calculate the average spend per traveler for each year and identify which year had the highest efficiency in spend per traveler relative to room nights. Please present all findings in a clear table format and summarize key insights.  

“This is how you might instruct an intern or junior member of your team who may not know exactly what to do,” Choe explained. “This is a good technique to get the answers you need and reduce hallucinations.”

Role Playing Prompt Engineering

This approach involves asking the AI to act like a specific person or character, such as a teacher, or manager or expert to response to the question in a way that fits the role. This technique should equip the AI with a particular perspective or expertise in order to provide a tailored answer. 

Example: Your role is a spend and compliance performance analyst. Your job is to analyze our KPIs and find ways to make savings. You can analyze travel data and output a short report. Perform this task.

“If you are a good manager, you are a good role-playing prompt engineer,” Choe explained. Like a job description, this technique should outline a clear task and goal, and a clear expectation of the outcome. 

“Asking the model to constrain and take this role means it can’t do other things, so it won’t be able to write a poem or give you a recipe for chicken, like ChatGPT can,” Choe said. 

The role-playing prompt is the “bridge” to an autonomous agentic AI configuration.

“Most of what people are calling ‘agentic’ is not really because it’s a single thing working in isolation. Agentic AI really has to be in collaboration because agents are programmed to be highly specialized. Alone they can’t do much but working as a team they can execute a goal based on your strategy,” Choe said.

According to Choe, overseeing a team of AI agents will become an increasingly critical responsibility for travel managers.

“The biggest application of AI moving forward is doing work that’s not being done at all right now because we don’t have the time or the people,” he said. “The workload is not just replacing humans—that will have the least impact—it’s about adding resources and capabilities… you need to be a manager of a virtual workforce. That’s the role.”

Some travel managers are already stepping up to the plate. McKinsey & Company’s director of travel and events technology Jamie Stewart was named BTN’s 2025 Travel Manager of the Year for introducing AI-powered, large language model booking technology into existing workflows for the company’s travelers and travel arrangers.

Salesforce senior travel manager for North America and JAPAC Ryan Pierce has also introduced a Slack-based AI bot that instantly addresses queries related to the software company’s travel program. This initiative, which he said lays the foundation for an agentic travel program, earned him BTN’s 2025 Best Practitioner recognition.

“We want to move away from managing trips to orchestrating outcomes,” Pierce said during the Hot Topics webinar. “In 2026, we’re focusing on automation—shifting [prompts] from ‘can you help me?’ and ‘assist me’, to ‘do this for me.’” The goal, he added, is to create frictionless traveler experiences and to redefine the travel team to become a strategic function rather than a service function. 

Similarly, ServiceNow’s Allegrina expects AI data analysis tools in 2026 to improve future predictions and strengthen her department’s role as “the trusted advisor” for corporate travel procurement.