Garbage In, Garbage Out - and Everyone Goes Back to Doing Nothing.
With Jez Rowe and Cam Lynch
"Garbage in, garbage out" gets said, everyone nods, and nobody does anything. Jez and Cam flip it: messy data isn't a reason to avoid AI, it's the first job to give it. Run the report you want, let AI show you where the gaps are, and clean up from there.
What this episode answers
Does my data need to be clean before I use AI?
No. Waiting for clean data just becomes a reason to do nothing, because business data is never completely clean. Cam Lynch, founder of Vibe CFO, starts the opposite way: build the report you actually want and see how big the mess is. If 5,000 transactions are missing a tracking category, you've found the real problem, such as timesheets being categorised half-heartedly. Fix that process, then clean up. Start with what you have.
Can AI audit messy data in Xero?
Yes. A general AI assistant like Claude can audit uncategorised Xero transactions when you connect it through the Xero API or a Xero MCP server, then help re-sync the fixes. An MCP, or Model Context Protocol, is a standard way to give an AI model access to another system's data and actions. Claude will happily say it can do the job, and there is usually more nuance than it first suggests, so check what each connection can actually reach.
What is the difference between a data error and a data gap?
A data error is a value that's wrong, and a data gap is a value that's missing. If a price column should hold numbers but one row contains a letter, quantity times price fails for that row, which is an error. If 50 of 5,000 rows in a custom field are blank because nobody entered anything, that's a gap. An audit picks up both, and AI makes that audit fast and cheap.
Why does AI hallucinate when business data is incomplete?
AI hallucinates most when it hits gaps in the data and tries to fill them instead of flagging them. Weaker tools act like an overly optimistic yes man, giving you a confident answer anyway. Better tools now say plainly where the gaps are. When you choose an AI tool for finance or reporting, test it on incomplete data and check whether it tells you what's missing. You still know your business best, so treat flagged gaps as your to-do list.
Is AI actually making businesses more productive?
Yes, AI is a productivity beast, and the boom is already happening. It lets you tackle the jobs you used to park for years, like auditing messy data or digging into problems you couldn't justify the time for. Prompt it and see what happens. The harder question isn't whether productivity rises, it's how you turn that productivity into a real return on investment.
Related: See what connects →
Episode notes
AI might be the cheapest data audit any business has ever had access to. Which makes garbage in, garbage out a strange reason to avoid it. Jez and Cam pull the objection apart and land somewhere most people don't expect: don't clean the data first, put it in and let the output show you what's broken. Data errors versus data gaps, why a good tool tells you where the holes are while a lesser one quietly makes something up, and what it looks like to have something willing to go down the rabbit hole you've been parking for two years. Stop treating your data like it has to be worthy of the thing.
Five minutes. One thing that matters. We ignore the rest, so you can too.
Vibe CFO is Australian founded. An AI CFO for small and medium business, out of the box, governed and secure.
Vibe in 5.
Transcript
Read the full transcript
Ah, yet another big week in AI. This is The AI Exhale, hosted by Vibe CFO with Jez and Cam. Five minutes, one thing that matters, or maybe two. We ignore the rest so you can too.
Jez: This week's theme on the AI Exhale is focused on the garbage in, garbage out concept, which has been around for a while in terms of data inputs and data outputs. And I just want to sort of frame this, Cam, before I pass to you. If I mention AI to a business owner, half the time some of them say garbage in, garbage out, and then everyone nods. And everyone goes back to doing nothing. When is data ever completely set, clean, perfect and 100%?
Cam: AI has the potential to clean up this data a little bit more, and also get your hands on reporting to see where the mess is. I have always gone the opposite to everyone. I don't say, oh, there's lots of garbage data, so I'm not touching it. I go, let's get the end result that we're looking at and see how big of a mess it is. You know, you put together a report and you notice, huh, I'm running this report where I wanted to use my tracking categories or my custom fields, and look, there's like 5,000 transactions that haven't been categorised.
All right, that's the problem. That means that we're entering timesheets and half-heartedly sometimes categorising it and sometimes not. How can we clean this up? That's my first port of call. All of a sudden, I've just cleaned my act up and realised where the problems are.
Jez: Maybe mess is good. You're going to see that it's, you know, 70% of what you're needing. Is AI arguably the cheapest data audit we've ever had?
Cam: Yeah. "I don't want mess, so I'm going to do nothing." I mean, that sounds like a flawed foundation in itself, doesn't it?
Jez: So really it's about, stop treating your data like it has to be worthy of AI, and just shove it in there. There are some platforms that offer data cleaning as part of the onboarding. I mean, really what you're saying here is that we can do it ourselves.
Cam: Definitely. Now, with OpenAI or a Claude subscription, there might be others off the shelf, but Claude is just this Swiss Army knife. Just through the chat, you could genuinely have this type of conversation, where you go to it and say, Claude, I have got 5,000 transactions that don't have the category applied to them. Here's my Xero MCP, or here's my Xero API.
Can you build a bridge to re-audit this and re-sync this? Claude will be like, "Sure." Claude will happily say that it can do it. There might be a little bit more nuance than how simple it seems to begin with, but Claude's a Swiss Army knife that can do this type of work.
Jez: And I think, you know, if I may pin this to hallucinations, people know about hallucinations in AI, and I think this comes into the fold a bit in this topic of conversation, because that's when there are gaps in data. And depending on the tool you use, it can be that rambunctious, overly optimistic yes man that hallucinates the answer for you, instead of what we're now seeing with better tools saying, "No, there's a gap here. There's a gap there."
Let's just finally cover a bit of a clarity point here. What would you say the difference is between a data error and a data gap?
Cam: A data error would be where you times two numbers from two different columns together, and the column has not got a data type on it. So it does 50 times a letter, right? So can I calculate that? That's like when you've got quantity times price, right? And you've got five times two. So each row does this calculation of quantity times price, except one of the prices doesn't have a number in it. It's got a letter. So then when you do the multiplication, it fails.
And then the other bit was the gap, right? The gap is like a custom field where you've got 5,000 rows, but like 50 of the 5,000 rows are blank or null because they didn't actually enter anything. So that's a gap in the data that needs to be filled. That's where an audit picks that up, picks up where you've missed holes.
Jez: You are the expert, the listener. You will read what is given to you and you can make sense of it. And that's back to the messy point. Messy is okay.
Cam: Yeah. We've got this ninja who can work with us at all hours of the day, go investigate things, and you use your head and your smarts to find the holes. You know your business better than this ninja, but this ninja is alive and welcome to go off and do these tasks. I feel like it's exciting. This is just such an exciting time, to be able to do these hard tasks that we used to avoid.
Jez: Hell, I like to say it, down the rabbit holes you wouldn't be bothered to go down.
Cam: It is. I even get that myself. I'm so deep in this, right? I have this project in my head and I go, oh my God, I'm not touching that tonight. And then I just go, Cam, just prompt it. See what happens. Stop being a baby. Get on with it. And all of a sudden you're solving this problem that you would have just parked for a couple of years.
Jez: And that's really back to the last comment here, quoting yourself, mate. It is a productivity beast, AI. That's what we're seeing and we'll see going forward. It is just an increase in productivity. And that's a good example there. So don't procrastinate. Get on with that project, give it a go, input the data and see what happens. Just take a leap.
Cam: Absolutely. There is undeniably, productivity is going to boom. It is booming. It's going to boom. It's unstoppable. What is hard is how you get a return on investment from that productivity. That's the real tricky bit.
Jez: Beautiful tease. That'll be, if not next week's episode, one of many to come.
And that's your five-ish minutes back. Vibe CFO, Australian founded. An AI CFO for small and medium businesses, out of the box, governed and secure. vibecfo.ai. Now you know.
Now you know.
The AI Exhale is made by VibeCFO. Australian founded, an AI CFO for small and medium business, out of the box.