Thoughtful Automation: Behind the Scenes of Request Assistant

Eighty-four percent of business travellers say that their company requires pre-trip approvals for at least some of their business trips, according to our 2024 SAP Concur Global Business Travel Survey. This appears to be a growing trend, as 67% have observed an increased focus on pre-travel approvals at their company this year.

Without the right measures in place, pre-trip approvals can be a substantial lift for both the employees seeking them and those approving them. Employees must first research and develop a planned itinerary, based on their travel preferences, company policies, and budget, to be run by the approver. Then, the approver must find time to review the itinerary from a policy and budgetary perspective, and either give the employee the green light to book or request changes. Provided the proposed itinerary is approved, the employee must search for their selected options again—hoping that prices have not changed substantially since the planned trip was developed—and book accordingly.

We introduced Concur Request around a decade ago to streamline the approval process. However, the itinerary development aspect of approvals remained a challenge. Determining whether a trip could be within budget, while accounting for the preferences of business travellers and company policies, continued to be very manual. It was always the goal to automate processes and add data-backed cost estimates in SAP Concur solutions, but the technology wasn’t quite ready—until the recent leaps and bounds made in generative AI.

Request Assistant was added as a feature to Concur Request in 2024. Using generative AI, it provides employees with cost estimates for their business trip, saving time and effort. It accounts for specifics including trip duration, destination, and services, as well as user preferences such as flight class, number of connections, and hotel ratings. It also works with other SAP Concur solutions.

When integrated with Concur Travel, the details gathered with Request Assistant carry through to booking to help business travellers save time. This aligns with what many business travellers are looking for; our survey found that 35% of business travellers would consider using AI-powered automation to help with shopping for or booking business travel, and 35% would consider using it to ensure compliance with their organisation’s travel policy.

Additionally, according to our survey, 37% of business travellers would consider using AI-powered automation to support capturing and reporting expenses. An integration with Concur Expense makes it possible to populate an expense report based on expected costs. If the final total matches the original estimate, the company can even configure Concur Expense to auto-approve, saving time for the employee’s manager and the finance team while ensuring quick reimbursement.

The wealth of SAP Concur data, combined with large language models (LLMs), provided the core framework for Request Assistant. The current solution is the first of many iterations as we plan to evolve Request Assistant further. We’re approaching it in phases, informed by customer feedback. Upcoming features may include increasingly personalised recommendations for users, mode of transport recommendations and trip planning, augmenting the cost estimates based on industry benchmarks and additional historical data, and more.

A phased approach is also important because, as those tracking generative AI news know, LLMs continue to evolve and approaching them with a degree of flexibility is necessary. One of the biggest challenges associated with building Request Assistant was the unpredictability of the generative AI technology. Each upgrade on the LLM side required adjustments to ensure retro compatibility and benefit from the latest innovation.

Another challenge arose from the fact that LLMs are not primarily designed for numerical computations. Considering the mathematical intricacies involved in estimating travel costs, our team devised strategic prompts to enable the LLM to more effectively compute the projected expenses of a business trip.

While we’ve largely addressed these challenges through testing, training, and fine tuning, time and experience with Request Assistant and other SAP Concur solutions remain critical. The user experience continues to be customised based on data, which in turn continues to strengthen the technology and its predictive capabilities. However, it’s important to note that user and customer data is secure and segregated—it is not shared with the LLM.

When booking travel, regardless of the tool used, it is entirely reasonable to spend an hour researching just one aspect of the business trip. Request Assistant, built on the foundation of SAP Concur data and generative AI technology, makes it possible to reduce that time spent to as little as five minutes, making the booking and approval process much simpler from beginning to end.