There is a task in audit that nobody brings up at conferences, yet it eats entire weeks of real work: the grant justification.
Justifying a grant is, at heart, an exercise in patience. You take each expense. You find its invoice. You find its payment. You check that the amount matches, that the description fits, that the date makes sense. And you repeat it, expense by expense, all the way to the end. One at a time.
I have seen it up close. I spent ten months inside a large audit firm, and this was one of those tasks everyone simply assumed had to be done by hand. Painstaking. Repetitive. And easy to sink with a single number that does not add up.
One error in cell 47 and the whole justification falls apart. The risk is not proportional to the size of the error.
The work no tool could solve
For years, this had no real technological answer.
Macros helped a little. Templates brought some order to the chaos. But in the end someone had to open the PDF, read the amount, go back to the spreadsheet and type it in. Someone had to decide whether that payment matched that invoice. The tedious part stayed human, not because it needed judgment, but because there was no other way.
And here is what struck me most from the inside: it is not a rare case. Documentary, repetitive, rule based work dominates the cost of an audit firm. Auditors spend close to 90% of their time on work that, to be honest, should not require a person.
The grant justification is the perfect example of that 90%.
How Jeff does it
That is why we built Jeff, our AI agent. And we recorded a demo doing exactly this, from start to finish.
The flow is simple, and that is the point. You hand it the documents and the client's expense spreadsheet. You tell it, in plain language, which grant you are justifying. No templates to fill in, no macros to configure.
From there, Jeff works.
It classifies each document. It extracts the amounts. And it matches every invoice to its payment and its description, expense by expense, the way a junior would, but in a fraction of the time. When it matches, the row turns green. When something is missing or does not fit, it flags it in red for the auditor to look at.
You even decide the color convention. If you prefer a different one to read the work more easily, you tell it and it adapts.
The part that really matters: traceability
If you have worked in audit, you know that speed is worth nothing without evidence.
A number that appears out of nowhere is not useful. It is a risk. And that is exactly what today's AI tools fail to give you: they hand you a result and leave you with no idea where it came from.
This is where Jeff is different. Every figure is traceable. You click a cell in the working paper and it goes back to its document, its page and its exact box, with the highlight over the original data. No black boxes. If a partner asks where that amount comes from, the answer is one click away.
That is what turns this into real audit, and not a pretty demo.
The auditor does not disappear. The chair changes
The takeaway is not that the auditor becomes unnecessary. It is the opposite.
With Jeff, the bulk of the justification is ready in minutes, not days. But the decision stays human. The auditor stops typing amounts and moves to reviewing, approving and sending the work back to the client. They stop executing and start supervising, which multiplies by four the work a single person can take on.
Demand for audit is not going to fall. Regulation grows every year, and with it the mandatory work of verification. What changes is the cost of producing it.
This is the first in a series. We will show Jeff task by task, with real product captures and synthetic data, so you can see exactly how it works.
If your last close looked anything like this, matching invoices and payments by hand well into the night, let's talk.
