For most managers, approving expenses is not really part of the job. It is something that comes on top of it.
Between meetings, deadlines and managing a team, there is a queue of expenses waiting for attention. Each one needs to be opened, checked and approved. Receipt present? Amount reasonable? Within policy? Anything unusual?
Then on to the next one.
The problem is that the vast majority of those expenses are already correct. If around 90% are fine, asking a manager to manually review 100% of them means spending most of their approval time confirming what could already have been established.
It is inefficient, but there is a bigger issue. When every expense demands the same attention, the expenses that genuinely deserve a closer look can become harder to spot.
This is why the next step in expense automation should not simply be about helping people approve expenses faster. It should change what they need to review in the first place.
Instead of checking everything, managers should be able to focus on what actually needs their attention.
Traditional expense approval is built around a simple idea: a person reviews every expense before it is approved.
It sounds safe. In practice, more checking does not necessarily mean better checking.
Imagine having 50 expenses waiting for approval. If 45 of them are completely normal, the approver still has to work through all 50 to find the five that are not.
That creates repetitive work. And repetitive work invites shortcuts.
When managers are approving expenses alongside their actual responsibilities, it is understandable that they want to get through the queue quickly. The first ten expenses might receive careful attention. By number 30, the process can become more mechanical.
That is exactly when an unusual amount, a duplicate receipt or a policy violation can slip through. The problem then moves downstream. Finance has to identify the issue, investigate it and potentially send the expense back for correction.
A process designed to create control has instead created work for two people. There is a better way to think about approval: not as checking every transaction equally, but as directing human attention towards the transactions where judgement adds value.
Software has already made it easier to identify unusual expenses. Rules can flag missing information or potential policy violations and tell an approver to take a closer look.
But a flag alone still leaves work for the person reviewing the expense. Why was it flagged? What is unusual? How serious is it? What should I be looking at? Without that context, the approver still has to investigate.
There is also another side to the problem that is easy to overlook:
If software can explain why an expense looks suspicious, why should it not explain why an expense looks perfectly normal?
This matters because automation depends on trust.
Telling someone that an expense is safe to approve is not enough when that person is accountable for the decision. They need to understand what the assessment is based on.
That is where AI can change the approval process in a more fundamental way. It can help distinguish routine expenses from exceptions, while explaining the reasoning behind both.
This is the thinking behind Findity’s brand-new Compliance agent, an AI-powered feature that introduces a different way to approve expenses.
The AI reviews each expense before it reaches the approver. It looks at the receipt, the company’s expense policy, spending patterns and previous approval decisions, then combines those signals to assess how confident it is that the expense is ready to be approved.
This means the approver no longer starts with a queue of expenses that all require the same manual check. The Compliance agent has already analysed each one and helps show where attention is needed.
Every expense receives a confidence score from 1 to 100 and is given one of two labels:
Ready for approval
or
Needs review
But the important part is not simply the label. Every assessment comes with a reason.
When an expense is Ready for approval, the approver might see:
“Expense with valid receipt photo.”
“No deviations found and similar expenses are usually approved.”
“The amount is not unusually high compared to similar expenses.”
When an expense is marked Needs review, the reason could instead be:
“Expense without valid receipt photo.”
“The receipt appears to have already been submitted by this employee.”
“The amount deviates from what this employee usually submits.”
“The expense seems to contain alcohol.”
The approver can see not only what the AI recommends, but why, while still making the final decision on every expense.
Most approaches to automated compliance focus on exceptions. If something looks wrong, explain why it looks wrong. That makes sense, but it only solves half of the approval problem.
If an AI system says that 42 expenses are fine and eight need attention, the approver needs confidence in both sides of that assessment. Otherwise, they may simply open the 42 supposedly safe expenses and check them again.
At that point, automation has added information without really removing work. The explanation behind Ready for approval changes that.
There is a meaningful difference between seeing:
Ready for approval
and:
Ready for approval: No deviations found and similar expenses are usually approved.
The first is a recommendation. The second gives the approver a reason to trust the recommendation. That distinction becomes particularly important when approving multiple expenses at once.
With bulk approval, expenses marked Ready for approval can be approved in two clicks. But bulk approval is only useful if it feels controlled. Approving dozens of expenses without understanding why they have been classified as safe would simply replace manual checking with blind trust.
Showing the reason behind a positive assessment means the approver can see that the expense has been checked before deciding to approve it.
AI that explains itself gives approvers the confidence to trust its recommendations.
Consider a manager with 50 expenses waiting for approval. The Compliance agent assesses them and marks 42 as Ready to approve. The manager can see why. Some have a valid receipt photo. Others have no deviations and are similar to expenses that are usually approved. The amounts are not unusually high compared with similar expenses.
Rather than manually opening and checking all 42, the manager can review the AI’s assessment and bulk approve the Ready to approve expenses in two clicks.
Eight expenses remain marked Needs review.
Suddenly, the approval task looks very different.
Instead of working through 50 expenses with equal attention, the manager can spend their time on the eight where there is actually something to consider.
As an example, if manually reviewing those 50 expenses would previously have taken around 75 minutes, focusing on the exceptions could reduce the task to approximately 17 minutes. That is roughly 75% less time spent on approval in this example.
But the biggest change is not the number of minutes saved. It is what those minutes are spent doing.
Human judgement is valuable because it is human. A manager may know why someone travelled on a weekend. They may understand the context behind an unusually high amount. They can decide whether an expense that falls outside the normal pattern is reasonable given the circumstances.
That is where their attention belongs.
Checking whether the 36th ordinary receipt in a queue looks as ordinary as the previous 35 is not a good use of the same judgement.
AI is particularly useful when it can do the repetitive work consistently, look across patterns and surface the cases that deserve a person’s attention.
The result is not approval without control. It is a different kind of control.
For the approver, the workflow shifts from “I need to check everything” to “I need to understand the exceptions.”
For finance teams, there is a secondary benefit. If unusual or non-compliant expenses receive more attention before approval, fewer mistakes should need to be corrected later.
But the biggest change happens for the person with the approval queue.
They can stop treating every expense as if it carries the same level of risk.
Today, the Compliance agent helps approvers make faster, better-informed decisions, while keeping the final decision with them.
The natural next step is to remove some of those routine decisions altogether. Low-risk expenses will be approved automatically, so approvers can focus even more of their attention on expenses that require review.
It is a natural progression for expense approval. The objective is not to automate human judgement, but to recognise where human judgement is actually required.
If an expense has a valid receipt, follows company policy, fits established spending patterns and resembles expenses that are routinely approved, requiring a manager to perform the same check again adds very little.
When something deviates from that pattern, the person steps in. From checking everything, to checking what matters.
For years, expense technology has focused on making the process easier. AI takes it a step further by removing routine work altogether.
That is the idea behind invisible expense management: it works in the background, with people stepping in only when their attention is needed.
The new Compliance agent is a first step towards that future, with auto-approve taking it further.
Over time, the best expense process may be the one you barely notice.
Expense approval on autopilot does not mean giving up control. It means knowing where your attention is needed, and always knowing why.
Want to see the new Compliance agent in action?
Watch the video or book a demo.