Thoughtful Automation: The Past, Present, and Future of ExpenseIt

ExpenseIt has undergone several iterations since it was introduced more than 10 years ago. It was originally an “automagical” application in its own right, until it became part of the SAP Concur mobile application in 2018. And interestingly, in its original form, it wasn’t just the app’s technology making manual expense reports a thing of the past for ExpenseIt users. When a picture of a paper receipt was captured, a combination of people behind the scenes and technology would create expenses, then match them to categories and credit card transactions.

Today, ExpenseIt is an add-on solution that is embedded in the SAP Concur mobile app and Concur Expense on the web. With the mobile app, you can capture an image of your receipt and then ExpenseIt automatically creates, categorizes, and itemizes an expense entry for you. Or, on the web, you can simply drag and drop a receipt into Concur Expense, and ExpenseIt will do the work. It now operates very differently from past iterations, due in part to its incorporation of new AI technologies, specifically generative AI.

When you take a picture of a receipt, ExpenseIt's AI and machine learning models will analyze the image and text derived through optical character recognition (OCR). Candidates for key values like vendor name, location, currency, amount, and expense type are predicted, and the most probable option is selected based on a confidence score.

These models continue to improve. For instance, we cross-reference data points to make better predictions. Rather than looking for a currency symbol alone, our technology also considers location. So, if the expense happened in Japan, ExpenseIt knows that it is unlikely to be in U.S. dollars. We also take historical predictions into account, too, by looking into how often certain predictions have required correction and using that information to strengthen the model.

In the latest generation of ExpenseIt, we are using generative AI to further automate expense report development. To launch this functionality, we wanted to start with a use case that would be especially helpful for business travelers: hotel folios.

Hotel folios are among the more complicated, time-consuming expense report scenarios, spanning anywhere from a few line items to 60-plus, depending on the length of stay and what services are used. An itemized hotel receipt associated with a two-week stay takes a very long time to manually enter, and tools like our itemization wizard have difficulty assisting when the rate changes over the course of a stay. Folios also vary in format based on hotel chain and in language based on location.

When an e-receipt is not available and you upload an itemized hotel receipt into Concur Expense, ExpenseIt starts with the usual OCR. It then runs the resulting text through a privacy filter to remove personally identifiable information. Since receipts can include employee names, phone numbers, home addresses, and even payment information, we’ve applied specialized open-source software to quickly, effectively, and securely remove these details while retaining the necessary data from the hotel stay. This software helps crop receipts, focus on the line items, and obscure anything that can’t be removed.

We then analyze the anonymized data using a large language model that returns critical information for an expense report, like dates, categories, amounts, etc. What the model sends to us is what is displayed to you, the user.

As we’ve explored generative AI for hotel folio itemizations, we’ve also run into a few challenges. For instance, companies take different approaches to how they categorize expenses. Organizations with many expense types tend to see less accurate predictions than those with broader categorizations. We’ve made progress in addressing these challenges through training the technology and working with our generative AI partners, and this is something that we’ll continue to focus on as we expand to additional expense categories.

However, it’s important that we note: Neither our partners’ generative AI models nor our technology is trained with your personally identifiable information.

ExpenseIt has come a long way since its early days, when SAP Concur had dedicated team members behind the scenes, reviewing every receipt submission and correcting issues to ensure accuracy. With the introduction of machine learning and now, generative AI, we’re taking significant steps toward a future where the expense report writes itself.

Learn more about ExpenseIt and the SAP Concur mobile app.