We have been experimenting with AI tools for small business admin for a while now, and invoicing seemed like a perfect fit. Give the chatbot your business details, tell it what you want to charge, and out comes a clean, professional invoice. Lovely. Except it kept going wrong in ways we did not spot until someone actually queried one.
Here is what we learned the hard way.
1. The VAT number format was wrong
Irish VAT numbers follow a specific format (usually starting with IE, then seven or eight characters). AI tools often generate a plausible-looking number or leave a generic placeholder that looks fine at a glance. Always manually enter your actual VAT number and double-check the format matches what Revenue issued you. Do not trust the AI to know this.
2. It did not include the supplier's full legal name
Revenue requires your full legal name or the registered business name on every invoice, not just a trading name. If you operate as a sole trader, that means your actual name. A few of our AI-generated templates defaulted to the brand name only, which is not sufficient for a VAT invoice.
3. The invoice date and tax point were missing or confused
A valid Irish VAT invoice needs the date of supply (or tax point), which is not always the same as the invoice date. When we asked an AI chatbot to generate an invoice template, it included one date field and labelled it vaguely. Fix this by having two explicit fields: invoice date and date of supply. If they are the same, fine, but both should be there.
4. VAT was shown but not itemised correctly
Revenue wants to see the VAT rate applied, the net amount, and the VAT amount shown separately for each rate used. We had a template that lumped everything together and showed one VAT total at the bottom. Tidy-looking, but not compliant. If you charge at different rates (say, some items at 23% and some at 13.5%), each rate needs its own line.
5. Sequential invoice numbering was not mentioned at all
The AI gave us a template with "Invoice #001" and left it at that. What it did not flag is that Revenue expects your invoice numbers to follow a sequential series with no gaps. The tool has no way of knowing what number you are up to, so this is entirely on you to manage. Use a spreadsheet or accounting software to track it, and do not let the AI generate numbers automatically without a system behind it.
6. The unit of measurement and quantity fields were vague
For goods especially, Revenue wants to see quantity and unit of measurement clearly stated. AI templates often use generic placeholder text like "description" and "amount", which does not prompt you to add this detail. We added explicit columns for quantity and unit before using any template.
7. There was no retention of records guidance built in
This one is less about the invoice itself and more about what the AI did not tell us. Revenue requires you to retain records, including invoices, for six years. Not a single AI tool flagged this when helping us set up our invoicing process. It is worth building a simple folder system, either local or cloud-based, from the start, because retrofitting it later when you have a couple of years of invoices scattered around is not fun.
The AI tools are genuinely useful for getting a starting structure together quickly, and what Irish small business owners think of AI tools right now reflects that mix of enthusiasm and caution we have found ourselves. But invoicing is one area where you really do need to go through every field manually and check it against Revenue's own guidance rather than trusting the output blindly.