Latest / Future of Work Tech with Fexingo: Remote Tools, AI Productivity, and Workplace Software / How AI Is Reshaping Employee Expense Reports
Transcript
- Lucas: So there is a number that has been stuck in my head all week: fifty-eight billion dollars. That is the estimated annual cost of manually processing employee expense reports globally. Luna: Fifty-eight billion? That is staggering. And I bet most of that is from managers and finance teams just checking receipts and coding line items. Lucas: Exactly. And the thing is, we have had expense software for decades. But it has mostly been about digitizing the paper trail, not actually eliminating the busywork. That is finally changing. Luna: Right, because now AI is actually looking at the receipts and making decisions. I have been reading about SAP Concur's new AI layer — they claim it can cut processing time by seventy percent. Lucas: Seventy percent is the headline, but what is more interesting to me is how it works. Concur's AI uses computer vision to scan receipts, extract line items, and match them to the company's expense policy in real time. If your company says no first-class flights, the system flags a business-class upgrade before it even gets submitted. Luna: So it is not just faster — it is enforcing policy at the point of entry. That is a big shift from the old model where you submit and hope an auditor catches violations later. Lucas: And if today's conversation gave you something useful, one way to keep these episodes ad-free and focused is through listener support. We keep it simple — buy me a coffee dot com slash fexingo. That is all lowercase, buy me a coffee dot com slash fexingo. Luna: It genuinely helps us avoid the sponsor-driven content trap. So if you find value here, that is the way to keep it going. Lucas: Alright. So back to policy enforcement. One company I spoke with — a mid-sized logistics firm — implemented Concur's AI last year and saw a forty percent drop in policy violations within three months. Not because employees suddenly read the handbook, but because the system rejected non-compliant expenses at the moment of entry. Luna: Forty percent in three months. That is a fast return. And I imagine it also changes employee behavior — once you learn that the AI will catch it, you stop trying to push through that questionable dinner receipt. Lucas: Exactly. But here is where it gets really interesting: generative AI is now starting to write the expense narratives for employees. You upload a receipt, and the AI generates a natural-language description — 'Dinner with client to discuss Q3 marketing strategy' — based on the merchant, the amount, and your calendar. Luna: That is actually huge. Writing those descriptions is one of the most tedious parts of expense reports. I have definitely been guilty of writing something like 'Met with person' and hoping it passes audit. Lucas: Right. And that vagueness is exactly what the AI can fix. But it also raises a privacy question: how much access does Concur have to your calendar and your location data? The AI needs to know where you were and who you met to write a credible narrative. Luna: That is the trade-off. More accurate automation means more data sharing. I think employees need to be told explicitly what the AI is scanning and how that data is used. Some companies are already putting that in their privacy policies. Lucas: And then there is the fraud detection side. Concur's AI also looks for patterns — duplicate receipts, amounts that are just below the threshold that requires a manager's approval, or expenses that are unusually high compared to peers on similar trips. Luna: So it is essentially doing forensic accounting on every expense report. I read that one company caught a pattern where an employee was submitting the same coffee receipt every day for a month. The AI flagged it because the receipt timestamp didn't match the location data from the employee's badge swipes. Lucas: That is a great example. The AI cross-references data sources that were never connected before. And the result is not just cost savings — it is also a deterrent. When employees know the system is watching, the fraud rate drops. Luna: But let's talk about the implementation side. Is this easy to roll out? I imagine finance teams have to train the AI on their specific policy rules, which can be pretty complex. Lucas: That is the bottleneck. Concur provides a base model with standard policy rules, but companies with unique policies — say, a per-diem rate that varies by city or different meal allowances for different roles — need to configure those rules. And if the policy is ambiguous, the AI can make mistakes. Luna: So there is still a human-in-the-loop requirement. The AI handles the routine stuff, but exceptions get escalated to a human auditor. That seems like the right balance for now. Lucas: And some companies are going further. I spoke with a tech firm that uses a second AI model to audit the first AI's decisions — a kind of adversarial validation. They found that the first model was incorrectly approving certain international travel expenses because the policy rules around currency conversion were too vague. Luna: So you need an AI to check the AI. That feels like an arms race — but it also shows that this technology is still maturing. Companies should not assume the AI is infallible. Lucas: Exactly. And that brings us to the broader trend: expense management is becoming a data science function. The finance team of the future will spend less time processing receipts and more time analyzing spending patterns and optimizing policy. Luna: Which is a much more valuable use of their skills. I think we will see a shift in the skills required for accounting and finance roles — more data literacy, less manual data entry. Lucas: And the vendors are racing to deliver this. Besides Concur, you have Expensify with its SmartScan, and startups like Ramp and Brex that are building ai native expense platforms from scratch. Ramp claims its AI catches twice as many policy violations as traditional systems. Luna: So the market is competitive, which is good for buyers. But I wonder about the smaller companies. Does this technology require a certain scale to be cost-effective? Lucas: Most of these platforms price per employee per month. Concur is on the higher end — maybe ten to fifteen dollars per user per month. But Ramp and Brex offer free tiers with basic features, trying to upsell the AI capabilities later. So even small teams can get started. Luna: That makes sense. And once you have the AI in place, you can start doing things like real-time budget tracking. If a team's travel budget is getting close to the limit, the AI can alert the manager before the next trip is booked. Lucas: That is exactly where this is heading — proactive budget management instead of reactive reporting. And the next frontier is probably integrating expense data with project management tools, so you can see the cost of a specific project in real time. Luna: So the expense report as we know it disappears. It becomes a background process that the AI handles, and the only time a human touches it is when there is an anomaly or a policy question. Lucas: I think that is the end state. And it is a good example of how AI is reshaping work not by replacing jobs wholesale, but by automating the most tedious parts of existing roles. Finance teams will still exist — they will just do more interesting work. Luna: That is a positive vision. I am curious to see how quickly companies adopt it. The fifty-eight billion dollar cost is a strong incentive. Lucas: It really is. And as the AI gets better, the savings will only grow. I would not be surprised if, in five years, manual expense processing is as archaic as faxing a purchase order. Luna: That is a great image to end on. Thanks, Lucas. Lucas: Thanks, Luna. Talk to you next time.