For decades, expense automation relied on a simple premise: If the machine can read the text, it can do the work. But anyone who has ever tried to scan a crumpled, smudged, or sun-bleached receipt from their pocket knows that reading isn’t enough. When key data is missing, such as a city name or a clear date, the machine halts and the burden falls back onto the user for manual entry.
To close this gap, where traditional Optical Character Recognition (OCR) fails, SAP Concur’s engineering team set out to break new ground. While much of the industry was still focused on the design of conversational interfaces, SAP Concur foresaw a bigger shift. They recognized early on that the next leap in efficiency wouldn’t come from better scanning, but from intelligent reasoning.
The result is an agentic AI upgrade for ExpenseIt, moving automation beyond simply reading text to solving messy logic puzzles, significantly reducing the need for manual intervention. Now, travelers can simply snap photos of their receipts as they receive them, upload digital scans, or forward receipts as emails, and ExpenseIt instantly transforms them into accurate expense entries with no date entry or itemization required.
Bringing this next-generation system called for a partner who could push the boundaries of innovation while matching the ambition to execute at startup speeds. SAP Concur fused its visionary roadmap with Google Cloud’s full-stack AI power, partnering with the only provider that co-designs every layer, from custom silicon and data platforms to world-class models and agents. Together, the teams engineered a true breakthrough in cost management — an AI agent that not only captures the receipt but intuitively understands the business traveler’s reality.
Speed, scale, and ingenuity
Standard expense automation is great at seeing what is on receipts but can’t see what is not there. SAP Concur saw the emergence of AI agents as an opportunity to create systems that could reason, decide, and act.
Suppose you upload a lunch receipt from “The Main St. Café,” which doesn’t include the address. In the past, this missing information would completely derail the automation and require you to manually enter this data to continue.
Agentic capabilities enable analyzing contextual clues, such as a vendor’s name, expense types, and trip itinerary data, to fill in the gaps. SAP Concur wanted to create an AI agent that could think like a human assistant: “I see ‘Main St. Café.’ I also see this transaction coincides with a business trip, where the user has a flight to Dallas and a hotel in Greenville, Texas. Therefore, this vendor is probably the restaurant located near the hotel in Paris, Texas — not Paris, France.”
To solve this challenge, the teams approached the problem with a dynamic, startup-style mindset. Instead of a lengthy development cycle, the collaboration was defined by rapid prototyping and bold problem-solving.
Utilizing Google’s Gemini models, they built the Receipt Analysis Agent, underpinned by a cognitive architecture.
Here’s how it works:
-
Ingestion: The user snaps a photo in the SAP Concur mobile app, uploads a digital scan, or forwards a digital receipt as an email.
-
Deterministic core: SAP’s foundational technology, refined over decades of processing global expenses, applies finely tuned logic to lift the visible text on receipts with high precision.
-
Intelligent rRouting layer: If the scanned receipt data is clear, there’s no need to trigger additional actions. If the data is ambiguous (e.g., “Missing location”), the routing logic dynamically directs the task to the Receipt Analysis Agent.
-
Contextual reasoning: Built with Gemini models, the AI agent doesn’t just guess — it uses tools and grounding to infer missing information. ExpenseIt feeds the partial receipt data to the agent, alongside grounding data like the user’s travel itinerary and business calendar.
-
ReAct (Reason and Act framework): The Receipt Analysis Agent connects the dots, validating the vendor against the location history, and then completes the expense entry.
