How to auto-approve expense claims without losing control

At a logistics company in Surabaya, the operations manager approves the expense claims of 64 drivers, dispatchers and field supervisors. In October that was 520 claims: tolls, parking, fuel, meals on long routes and the odd tyre repair. Most were under Rp 200,000. The manager approves them on a phone, in batches, on Friday evenings, at a few seconds each.

One of them was a Rp 4,750,000 claim from a driver for a truck repair at a roadside workshop, with a handwritten receipt. It got the same few seconds as a Rp 15,000 parking ticket. Nobody asked why a driver had paid for a repair out of their own pocket, or why the workshop wasn’t one of the company’s vendors. When internal audit called the workshop in November, it had no record of the job.

The problem isn’t the manager. When one person signs 520 claims a month, approval stops being a decision. Auto-approval fixes that by giving the small, routine, clean claims to a written rule, so the manager’s attention goes to the hundred or so that need a person. Done badly, it’s the same rubber stamp without the manager. This guide covers what auto-approval is, what a safe rule contains, a worked example of that company’s October, how it goes wrong and how to set it up.

What is auto-approval?

Auto-approval means an expense claim is approved without a person because it meets conditions written down in advance: below an amount, in an allowed category, passing every policy check and raising no flags. Everything else goes to an approver as before. It belongs in your approval matrix as one more level of authority, given to a rule instead of a manager.

It isn’t the absence of control. The policy and duplicate checks still run on every claim, the rule is approved by whoever owns the authority it uses, every claim it approves is recorded as approved by the rule, and someone reviews a sample afterwards. What changes is what a person checks: the rule and a sample of its results, rather than every claim one at a time.

There are two ways to run it. One is a fixed rule in your approval system. The other is an AI agent working within an authority limit you set, which can also compare a claim with the employee’s history and record why it approved. Either way, the limit should be enforced by the system, not left to the judgement of whatever applies it.

What a safe auto-approval rule contains

• A limit per claim. Set from your own claims, for example Rp 500,000, rather than from what feels small.
• A limit per employee per month. For example, Rp 3,000,000 of auto-approved claims, so ten claims of Rp 450,000 don’t pass one at a time.
• Allowed categories. Routine ones that need little judgement: parking, tolls, fuel, local transport and meals within the cap. Not entertainment, gifts, repairs, equipment or anything that should have been a purchase request.
• Every policy check passed. Within the caps for the employee’s grade, a readable receipt attached, submitted within the claim window and coded to a valid cost centre. The checks come from your expense policy.
• No flags. No possible duplicate across claims, card lines and cash advances, no amount just under a limit, no claim dated on a day the employee was on leave and no merchant the company hasn’t seen before.
• Eligible employees. Not new joiners in their first months, not people serving notice and not anyone who can edit the rule.
• A record and an owner. Each approved claim is logged as approved by the rule, with the rule’s version, and a named person owns the rule. If a claim is edited after it’s approved, it goes back through the checks.

Approval doesn’t release payment. Auto-approved claims join the same reimbursement run as every other claim, and the people who release payments stay the same.

A worked example: one month of claims

Here is the logistics company’s October, sorted the way an auto-approval rule would sort it. The rule, only as an example: claims under Rp 500,000, in parking, tolls, fuel, local transport or meals, passing every policy check with no flags, up to Rp 3,000,000 per employee per month.

October claims	Claims	Rp ’000
Approved by the rule	412	61,800
Under the limit, held by a check	38	6,270
Over the limit or excluded	70	98,400
Total	520	166,470

The rule takes 412 of the 520 claims, nearly four in five, off the manager’s list, but only Rp 61.8 million of the Rp 166.5 million, about 37% of the money. The manager now reviews 108 claims worth about Rp 104.7 million, the ones where a person makes a difference. The Rp 4,750,000 truck repair is one of them: it’s over the limit, and repairs aren’t an allowed category.

The 38 claims under the limit that the rule held back:

• 12 without a readable receipt. Blurred photos, and handwritten receipts the rule can’t check.
• 9 from employees who had reached the monthly limit. Most from two drivers on long routes, which is worth a conversation in itself.
• 9 possible duplicates. The same amount and date as a card line or an advance settlement, which a person compares with the original.
• 8 from merchants the company hadn’t seen before. Most were genuine, and approving them teaches nothing; a person confirming them does.

Each week, finance reviews a random sample of the claims the rule approved, for example 5%, which is about 20 a month, and a report of auto-approved spend by employee. In October the sample found one claim coded to the wrong cost centre and nothing else. That finding, written down, is the evidence that the rule works, and what you show your auditors.

Why auto-approval goes wrong

• The limit is set from what feels small. A limit chosen in a meeting, rather than from the claims, is often too high. For a team whose typical claim is Rp 150,000, Rp 2,000,000 isn’t small.
• There’s no monthly limit. A limit per claim on its own invites splitting: four claims of Rp 450,000 for one Rp 1,800,000 dinner.
• The rule checks the amount and nothing else. Auto-approval only works on top of policy and duplicate checks. Without them it approves anything small, including the red flags a person might have caught.
• Late approvals are approved by default. “Approved automatically if the approver doesn’t respond within three days” isn’t auto-approval, it’s no approval. A late approval should be escalated to someone else, never decided by the clock.
• Nobody looks afterwards. Without a sample and a report, a rule that’s too loose is found by the auditors.
• The record doesn’t say who approved. Auto-approved claims carry the manager’s name, or no name, so nobody can tell later which claims a person actually saw.
• The rule never changes. Meal caps rise and categories change, but the limit is still the one set three years ago, by someone who has since left.

Auto-approval in Indonesia

• Write it into the delegation of authority. Auto-approval is authority delegated to a rule, so it belongs in the written delegation of authority or SOP, approved by the directors who own that authority, with the limit and conditions spelled out and a date to review them.
• Expect the auditors to test the rule. Instead of checking a signature on each claim, your auditors may test the rule as a control: what it says, who approved it, when it changed and whether it ran as written. Keep its version history and the results of each sample review.
• Handwritten receipts. A handwritten receipt (kuitansi) from a roadside workshop or a small shop is common and hard to check automatically. Send handwritten receipts above a small amount to a person, whatever the category.
• Tax documents don’t depend on who approves. Where you credit input VAT (PPN masukan) on a category, the claim needs a complete and correct faktur pajak in the company’s name, whether a person or a rule approved it. For those categories, make the faktur pajak a condition of the rule.
• Allowances aren’t claims. Fixed allowances, such as a monthly transport or phone allowance, are treated differently for PPh 21 from reimbursements of actual costs. Keep them out of the claim categories the rule approves, and confirm the treatment with your tax adviser.

Signs your expense approvals need a rule

You probably need auto-approval, or a better rule than the one you have, if any of these sound familiar:

• Managers approve hundreds of claims a month, mostly in batches.
• A Rp 15,000 parking ticket and a Rp 5,000,000 claim wait in the same queue.
• Claims wait a week for a signature nobody reads closely.
• Approvals pile up and are cleared in one sitting before the reimbursement run.
• Late approvals are approved automatically after a few days.
• Your auto-approval rule has an amount limit and no other conditions.
• Nobody can say which claims a person actually looked at.

Five hundred signatures a month isn’t a control. A written rule, applied to every claim and checked afterwards, is.

How to set up auto-approval

1. Start from your claims. Export three months of claims and sort them by category and amount. Find where the volume is and where the money is; they’re rarely the same claims.
2. Write the rule. The limit per claim, the monthly limit, the categories, the checks and the exclusions, with the numbers taken from your own claims.
3. Approve it as part of the matrix. By whoever owns the authority the rule uses, with a named owner and a review date.
4. Run it in the background first. For a month, let the rule mark what it would approve while managers still approve everything, then compare. Any claim the rule would have approved that a manager rejected shows what the rule is missing.
5. Switch it on for one team. Start with one team or entity and a lower limit, and widen it once the samples come back clean.
6. Sample every week. Review a random sample of auto-approved claims and a report of auto-approved spend by employee, and record what you found.
7. Review it every quarter. Check the limits, the categories and the sample findings, and update the rule in writing. While you’re there, make sure late approvals escalate to a deputy rather than approving themselves.

---

How Procapy helps

Proca Agents include a template for low-value approvals. You write the agent’s jobdesk in plain language: which transactions it may approve, the policy conditions, the amount limit and who it escalates to. ProcApprove enforces those limits, so the agent’s scope is set by Procapy, not by its judgement. A typical starting point is expense reports under a set amount that pass every policy check; everything else goes to a person, with the agent’s recommendation and the evidence attached. Payment release stays with the people in your approval matrix.

The agent approves claims that have already been checked. In Capy Expense, every line is checked against your expense policy before a report can be submitted: per-diem and meal caps, receipt thresholds, allowed categories and budget. Duplicate receipts are caught across employees, reports and card feeds. Every agent approval is logged as such, with its reasoning, in the same audit trail as your team’s own actions, and each run produces a summary of what the agent approved and what it escalated. When your team edits or rejects a result, the correction is applied to that agent’s next run. If a request is edited after it’s approved, its approval chain re-evaluates it, and approvers who still handle hundreds can act on a batch in one pass, with anything ProcAI has flagged held back for individual attention.

A good place to start is the agent doing a first-pass review of expense reports, then extending its authority to low-value approvals once the team trusts the run reviews. The approvals page covers the rest of the approval flow. Book a 30-minute walkthrough, bring your delegation-of-authority policy, and we’ll configure the first chain live.