Straight answers about AI you can audit
The real questions finance teams ask before they trust AI with money — about mistakes, cost, compliance, and proof. Answered in plain English by a former Controller, with nothing dressed up.
These are the actual objections I hear from accounting firm owners, fractional CFOs, and Controllers — not softball questions written to make me look good. If yours isn't here, the short version of every answer is the same: the AI's "I'm done" releases nothing on its own, a human approves the risky moves, and every step lands in a record nobody can quietly edit.
Want to pressure-test a tool yourself? Run it through the 12-point "will your AI pass an audit?" checklist. Want to know who's behind this and why I build it this way? Read the about page — the short story is twelve years as a Controller, then code.
No jargon. No dodging.
Will AI make mistakes with my money?
Yes — AI makes things up, and you should never trust it on that alone. That's exactly why, in what I build, the AI saying "I'm done" releases nothing. An independent check has to pass first, and anything risky stops for a human. The mistake isn't the real danger; not being able to catch it before money moves is. So the system is built to catch it before money moves.
What happens when the AI is wrong?
Every demo shows you the AI working. The real question is what happens on the exception. In an audit-safe build, when the AI hits something it isn't sure about — an odd invoice, a bank-detail change, an amount that doesn't match — it stops and hands it to a person instead of guessing and moving on. And every step is written to a record nobody can quietly edit, so you can always reconstruct what it did and who approved it.
We're too small and not technical — is this for us?
You don't touch the wiring. You get the busywork handled and a record you can pull up. The whole point is that you don't need to be technical: you tell me the task that drives your team crazy, I build the automation with the checks and approvals in the right places, and you stay in control with plain-English answers. Small teams are often the best fit, because one fraudulent invoice or one bad month-end hurts more when you're small.
$12K is a lot — where do I start?
You don't start at $12K. You start with a Process Audit, around $1,500, which is yours to keep and credits toward a build if you go ahead. It's a plain-English plan: what to automate first, what it'll take, and what you'll get back. That way you're deciding on a real plan with real numbers, not signing up for a big build on faith. Plenty of people do the audit and stop there with something useful in hand.
How are you different from other AI automation firms?
Most Utah AI firms sell you a chatbot and a 66%-savings promise. I sell automation your auditor will sign off on — verified work, human approval on risky steps, and a tamper-evident record. I was a Controller for twelve years before I started building this, so I build the part most firms skip: the part you can defend when someone asks "who approved this?" If a vendor can't answer that question, walk away.
Is this even compliant? What does EU AI Act Article 12 require?
Article 12 of the EU AI Act deals with record-keeping — it requires certain AI systems to automatically keep logs of what they do so their behavior can be traced and reviewed, with a key enforcement date of August 2, 2026. I build exactly that kind of tamper-evident logging today.
This is a practitioner's plain-English summary, not legal advice — check the official regulation and your compliance counsel for your specific situation.
What does "human in the loop" actually mean?
It means a real person is built into the process at the points that matter, with the power to approve or reject before anything happens. It's the most overused phrase in AI sales, so here's the honest test: ask to see the exact point where the system stops and waits for the human. A real loop has a line of code that halts the process until a person signs off. A fake one just means someone could log in and look around if they wanted to. A real loop stops; a fake one only watches.
Can I verify the audit log myself?
Yes — and if a vendor can't let you, that's a red flag. A real audit trail is tamper-evident: the records are linked together cryptographically, so changing one quietly breaks the chain and the tampering shows. You can run the check yourself, offline, and confirm the record wasn't edited after the fact. If your AI vendor can't hand you a way to verify the log independently, the "audit trail" is theater. Don't trust me — verify me.
Which departments can use this?
Any team with repetitive work that follows rules — sales, marketing, operations, HR, customer service, finance, and leadership. The strongest fit is finance, because that's where the stakes and the audit pressure are highest, but the same approach works anywhere: automate the busywork, keep checks before important actions, and keep a person on the risky steps. If it's repetitive and follows rules, it's probably a candidate.
How much does it cost?
Everything starts with a free intro call. A focused Process Audit starts around $1,500 and is yours to keep. A full Automation Build for a real process typically runs $12,000 or more depending on scope, and ongoing care is custom, month to month. No surprise pricing — the audit tells you the real number before you commit to anything bigger.
How does the fraud catch actually work?
The most common money scam is a fake bank-change invoice — "update our payment details to this new account." It looks routine, which is why it works and why a lot of AP automation pays it. The scanner checks for that pattern and refuses to change banking details without a separate, out-of-band confirmation, flagging the invoice for human review before payment. It's not a smarter AI that stops it; it's a control that won't move money on a bank-detail change without verification.
We tried an AI tool before and couldn't trust it with real work. Why is this different?
That's the most common story I hear: it worked in the demo, then you couldn't actually trust it with real client work. The reason is almost always the same — the demo showed the happy path and skipped the exception, the approval, and the record. I build those three things first: an independent check before anything releases, a human gate on the risky steps, and a tamper-evident log. Trust isn't a personality trait of the AI; it's something the system has to earn with evidence every time.
Do I have to take your word that it's safe?
No, and you shouldn't. I red-team my own code and write NO-GO verdicts on it when it fails — that habit is the whole product. On a call I'll run real working examples live, including an AI catching a fraudulent invoice and a tamper-evident record breaking when a single byte changes. You can also work through the 12-point audit checklist yourself against any AI tool you're considering, mine included. Don't trust me — verify me.
Still have a question I didn't answer?
Bring it to a free intro call. Bring the AI tool you're unsure about too — no jargon, no pressure, no pitch you didn't ask for.
Book a free intro call