Latest / Future of Work Tech with Fexingo: Remote Tools, AI Productivity, and Workplace Software / How AI Is Reshaping Your Expense Reports
Transcript
- Lucas: So it's late June 2026, and I just found myself standing in the kitchen last night, staring at a crumpled receipt from a coffee meeting I had three weeks ago. And I thought — why am I still doing this? Luna: You manually entered an expense report in 2026? Lucas, that's almost nostalgic. Lucas: It was a moment of personal shame, yes. But it got me thinking about how dramatically expense reporting has changed on the corporate side. While I was squinting at a coffee stain, most finance teams are now using AI to process receipts before the employee even submits them. Luna: Right, I've been seeing more about this. Platforms like Expensify and SAP Concur have been adding machine learning layers for a few years, but it sounds like the technology really hit an inflection point recently. Lucas: Exactly. And the key shift is from reactive to proactive. Older systems would just scan a receipt and pull out the date and amount. Now, the AI is categorizing expenses in real time, flagging policy violations before reimbursement, and even predicting which departments are about to blow their travel budgets. Luna: I want to get into the predictive piece. But first, give me a concrete example of the automation in action. Lucas: Sure. I spoke with the CFO of a mid-sized SaaS company — about 800 employees — who switched to an ai native expense platform last year. Their old process: employee submits a report, manager approves, finance audits. Average time from submission to reimbursement was about 12 days. Luna: That's pretty typical, right? Lucas: Typical, yeah. After implementing AI — machine learning models trained on six months of historical data — that dropped to under four days. And here's the number that got me: they reduced the finance team's manual review time by 70 percent. The AI now auto-approves about 60 percent of all expenses without human touch. Luna: Seventy percent is massive. What's the catch? I assume the AI isn't perfect. Lucas: The catch is the edge cases. The same CFO told me their AI initially flagged a $200 client dinner as a potential policy violation because the restaurant name included 'bar and grill.' The algorithm learned that 'bar' often correlates with personal alcohol expenses, but in this case it was a legitimate business dinner. Luna: So the model was over-indexing on keyword associations. That's a classic machine learning pitfall. How did they fix it? Lucas: They added a human-in-the-loop review for flagged items under $500. And they retrained the model on context: not just the merchant name, but the attendee list and the time of day. A dinner at 7 PM with a client is different from a solo bar tab at midnight. Luna: That makes sense. But I'm also thinking about the employee experience. If the AI is auto-approving most things, does that mean less friction for the average worker? Lucas: In theory, yes. But there's a trust angle. Some employees feel uneasy knowing an algorithm is scrutinizing every coffee they buy. I've seen survey data — about 30 percent of employees say they'd prefer a human approval for any expense over $50, even if the AI is faster. Luna: Interesting. So the technology might be ready, but the culture isn't fully there yet. Let's talk about the predictive piece you mentioned earlier. Lucas: Right. This is where it gets really interesting for finance teams. Some platforms now use historical spending patterns to forecast departmental budgets. For example, if the sales team typically spends 120 percent of its travel budget in Q4, the AI alerts the CFO in September. It can even recommend preemptive caps or reallocations. Luna: That's a big step up from just processing receipts. It turns expense data into a strategic tool. Who's leading in this space? Lucas: Beyond the incumbents like Concur and Expensify, there are newer players. One I've been tracking is a startup called Ramp — they've built AI directly into their corporate card system. Every transaction is categorized in real time, and the AI can block a purchase if it violates policy before the card is even swiped. Luna: That's proactive to the point of being preemptive. Do employees find that intrusive? Lucas: Some do. But the companies that adopt it tend to frame it as a convenience feature — you never have to worry about accidentally expensing something non-compliant. And it reduces the back and forth with finance. The trade-off is transparency: employees want to know exactly why a transaction was blocked, and not all platforms explain their reasoning well. Luna: Explainability is a huge issue in AI generally. So what about fraud detection? I imagine the AI can spot patterns humans might miss. Lucas: Absolutely. One classic fraud pattern: duplicate submissions. An employee submits the same receipt twice, hoping one gets lost. AI catches that instantly because it compares every new submission against the entire history. More sophisticated models can flag anomalies like an employee who always submits expenses just under the receipt threshold — say $24.99 when the threshold is $25. Luna: That's a known red flag in audit circles. But the AI can do it at scale across thousands of employees. Lucas: Exactly. And it doesn't just flag the individual; it looks for patterns across teams. If one department has a significantly higher rate of 'just under threshold' expenses, the AI alerts the manager. It's a systemic view that's impossible for a human auditor to maintain manually. Luna: I wonder about data privacy, though. The AI is ingesting every transaction an employee makes. Where's the line between useful oversight and surveillance? Lucas: That's the billion-dollar question. The platforms I've looked at anonymize data at the model level — they train on aggregated patterns, not individual profiles. But the real-time monitoring is undeniably personal. Some companies address this by giving employees access to their own ai generated profile so they can see what the system 'knows' about their spending habits. Luna: Transparency as a trust-building measure. That seems smart. Do you think we'll ever get to a point where expense reports are fully automated — no human involvement at all? Lucas: I think we're heading that way for low-risk, low-dollar expenses. But for anything unusual or high-value, companies will want a human in the loop for the foreseeable future. The technology is robust, but the liability is still on the company. If an AI approves a fraudulent $10,000 expense, the company can't blame the algorithm. Luna: True. The accountability has to rest somewhere. So what's the next frontier? After receipts and approvals, what else can AI do in this domain? Lucas: I see two big areas. One is integration with travel booking. If the AI knows your company's travel policy and your personal preferences, it could book flights and hotels that are automatically compliant — no expense report needed afterward. The other is real-time tax optimization. For international expenses, the AI could apply the correct VAT or GST treatment at the point of sale, rather than during reconciliation. Luna: That second one would save finance teams a ton of headache during tax season. It's like moving from retrospective accounting to embedded compliance. Lucas: Exactly. And that shift — from reactive to proactive to embedded — is the real story here. Expense reporting used to be a backward-looking chore. Now it's becoming a forward-looking data stream that feeds into broader financial planning. Luna: Before we wrap, I want to come back to something you said at the top. You mentioned you manually entered a receipt last night. If you could wave a magic wand, what would the ideal expense experience look like for you? Lucas: Honestly? I'd love a system where I don't think about expenses at all. The company card auto-categorizes everything. I get a weekly summary: 'Here's what you spent, here's what's approved, here's what needs explanation.' I only intervene when something's flagged. That would save me maybe 15 minutes a week — which doesn't sound like much, but multiplied across the workforce, it's enormous. Luna: Fifteen minutes per employee per week. For a company of 1,000, that's 250 hours a week. That's real productivity. Lucas: And that's the promise. The technology is already there for most of that vision. The remaining hurdles are cultural — getting employees comfortable with the AI, and getting companies to trust the AI enough to reduce human oversight. Luna: Speaking of trust and value, I think we should mention something that keeps shows like this going without ads or corporate sponsors. Lucas: Yeah, you're right. We've been talking about how AI can take friction out of everyday tasks. And in a similar spirit, listener support is what keeps this podcast friction-free — no sponsored segments, no ad breaks. If you find value in episodes like this one, you can support the show at buy me a coffee dot com slash fexingo. Luna: It's a small way to ensure we keep digging into topics like automated expense reporting — or whatever the next workplace tech shift is — without any outside influence. Lucas: Exactly. And back to that ideal expense experience — I think within three years, we'll see the first major corporation announce a 'zero-touch' expense policy, where 95 percent of expenses are auto-approved with no human review. That's the benchmark I'm watching. Luna: That would be a milestone. For now, I'll settle for not having to squint at coffee-stained receipts. Lucas: Amen to that. Thanks, Luna. Luna: Thanks, Lucas.