The one place AI genuinely earns it
Supplier invoices arrive as PDFs, photos and email bodies, in a different layout from every supplier and often a different layout from the same supplier last month. That variability is why rules fail on them, and it is one of the few places where an AI step genuinely pays for itself rather than being sold to you.
A rule needs to know where the total is. Supplier invoices put it in a different place every time — and the same supplier changes their template without telling anyone. Then somebody photographs one on a phone in a van and it arrives rotated.
This is genuinely unstructured input, which is the precise test for whether an AI step is warranted. If a person is currently reading something and deciding, a model might help. If a person is following a rule, a rule will do it cheaper — that test is set out on the AI tools page.
Open the email. Read the PDF. Work out which supplier, which job, which account code. Type it into the accounting package. File the PDF somewhere it can be found. Check it against what was ordered. Chase the one that does not match.
The last two are where the time goes, and they are the ones nobody counts as a software cost. Most people estimate their AP time by thinking about the typing and miss the checking and chasing entirely — which is why the estimates are usually low by a factor of two or three.
Bills arrive at one address. The line items and total are read out. The likely supplier, job and account code are proposed. A human approves — or, above a threshold you set, a human approves and below it nothing needs approving at all.
The approval step matters more than the reading step. An AP automation with no human checkpoint is a way to post wrong numbers faster, and errors in accounts payable are both expensive and hard to spot, which is a combination specifically warned about.
A model call has a unit cost. A rule does not. So the design goal is to move as much of this as possible into rules and leave only the genuinely unstructured reading to the AI step.
In practice that means a supplier who sends a consistent format gets a rule after the third invoice, not a model call forever. That reduces what you pay monthly, which is the correct incentive and one worth asking any vendor about — it is on the questions list.
Under roughly twenty supplier bills a month, the sums do not work. The build cost is the same whether you process twenty or two hundred, and at twenty the manual version is cheaper than the automation plus its running cost.
It also does not pay if your problem is approval rather than entry. If bills sit for three weeks because nobody will sign them, automating the reading makes them sit unread faster. That is a process problem and no software fixes it.
There is no reliable published Australian figure for how long supplier bill processing takes in a small business, and we are not going to borrow one from a vendor whitepaper to make this page more persuasive. Most numbers you will see quoted for this were produced by companies selling the fix.
What we would do instead is count it on one of your jobs. Follow a single job end to end, note every time a human touches this step, and multiply by what that person costs loaded. That is a real number about your business and it takes an afternoon.
You can run the arithmetic yourself first — the leak calculator uses ABS, Average Weekly Earnings, Australia, May 2026 for the wage default and ATO, Super guarantee, current rate for super, and returns a range rather than a single confident figure. If the range comes back under our minimum engagement, that is a real answer and you should stop there.
Sometimes, which is why the approval threshold exists and why we will not build this without one. The question is not whether it errs but whether you would catch it — and if the answer is no, we would rather build the checking first.
Not published, because we do not have measured before-and-after numbers we can stand behind yet, and a case study without them is a story. Kindra AI was founded in 2025. When the measurements exist they will appear here with the sample size attached.
20 minutes. We find where the money's going. No pitch.
Book a free Leak Check