Smart Software Can’t Save Broken Infrastructure

It's Tuesday afternoon and your AI copilot has just finished categorising last month's expenses in under a minute. Great. Except the bank feed it pulled from is three days behind, a quarter of the credit card transactions were in breach of the staff expense policy, and an employee who paid for a client dinner is still $480 out of pocket while they wait for reimbursement.

The AI did exactly what it was asked to do. The finance function still failed.

This is the gap most CFOs aren't really talking about. We've spent the last couple of years bolting AI onto every process we can find - drafting board updates, summarising meetings, automating reconciliations - and we've started to convince ourselves that we're building a 2026 finance function. But underneath all of it, the platforms are the same as those we were running on in 2012. And no model, however capable, can sign a paper bank form for you.

The illusion of 'modern' finance

The great hope is that AI is going to eliminate the boring parts of finance. Claude drafts the commentary on variances to budget. An agent codes the transactions. A workflow handles the monthly journal postings. The CFO finally gets to be a strategist instead of a processor.

It's a good idea, but it isn’t materialising in reality.

In the real world, finance teams are still bogged down by everyday fires. Think about the disruption when a director leaves and their corporate card expires, suddenly cutting off software access for three entire departments. Or the tedious reality of sitting there manually approving a mountain of expense reimbursements after an employee returns from an interstate client trip.

Suddenly the future breaks, and you’re back to being reactive.

The AI layer may be fast, but if the infrastructure layer is not up to task, then that becomes a costly time drain for your finance team. Hindering them from being a forward-thinking function to, once-again, focusing attention on what’s already in the rearview mirror.

From tools to infrastructure

Real modernisation isn't an AI layer sitting on top of a broken stack. It's rebuilding the stack so the AI has something worth working with. 

The problem the AI was originally being used to solve can simply stop existing. Take an expense claim, yes, AI can speed up processing it. But better infrastructure can mean that there isn't even a claim to process in the first place. 

The travelling employee was given a temporary-use card with a fixed budget to pay for the interstate trip, and the transactions were ready to be reconciled in the accounting system within seconds. No reimbursement queue. No out-of-pocket employee. No claim.

Or how about the company's monthly software subscriptions? These could instead be put onto a dedicated ‘Tech Subscription’ card which is controlled by the finance team, so it doesn’t matter if people come and go, the transactions stay sound. 

What this looks like in practice

You already have an invisible AI assistant. The next move is an invisible finance function: Infrastructure that handles policy, budgets, approvals and reconciliation in the background, and only surfaces the things that actually need a human in the loop.

That's what the travel and subscription card examples were really showing. The travel budget was set in advance, the employee’s daily card spend was visible in real-time, and no subsequent reimbursements were needed. The subscription cards roll-on in the background, not eating up any individual’s credit limits, and not being influenced by people’s departure or turnover.

All of it never went out of sync. The post-hoc workload disappeared, not because AI got faster at doing it, but because the infrastructure made it unnecessary.

The CFOs getting this right aren't waiting for their legacy providers to catch up. They're building on smarter infrastructure - providers like Archa - that were designed for how a finance function actually wants to run, not how banks were comfortable running them in 2012. The gain isn't marginal. It's the difference between a finance team that spends its month chasing receipts and reconciling feeds, and one that spends its month adding strategic value to the business.

If your team is still waiting days for bank feeds, processing expense claims that shouldn't have existed in the first place, or manually reviewing data before it's usable, you're paying an efficiency penalty to your legacy provider every month. A smarter prompt won't fix that - better infrastructure will.

‍