Supplier invoices IA
They arrive as PDFs, each in its own format. They get read, the amounts are extracted and entered into your accounting software already categorised.
Most ordered
Process automation
Stop typing the same thing twice
If someone on your team copies data from one place to another, sends the same email every week or types invoices into an Excel file by hand, that can be automated. It is not magic and you do not have to change software: what you already have gets connected.
If the saving does not cover the work, I will tell you and we will not do it.
What automating actually means
A real, boring example: what happens between an order coming in and the invoice going out. Each box is a step; the grey ones are done by a person.
Now ≈ 11 min per order
The order comes inemail or WhatsApp
Someone writes it in the Excelby hand, 3 min
Someone checks the stockinto another program
The customer is notifiedemail written by hand
The invoice is createdretyping the data
After ≈ 40 sec, and only if something goes wrong
The order comes inthe same as before
All that is recorded isand takes it off stock
Customer notifiedwith its real timescale
Invoice issuedin your software
You just reviewwhatever does not sound right
The last step is still yours on purpose. An automation nobody checks ends up introducing errors for months without anyone noticing.
Do the sums before you ring me
Automating costs money and does not always pay off. Move the three sliders and find out: if the number does not justify the work, better you know now.
Adding up everyone who touches it.
Gross hourly pay plus social security. If it is your own time, count what you bill.
It is almost never 100%: there is always the checking and the odd cases.
What it costs you a year
—
—
Hours you get back—
Per month—
It is paid in—
—
The sum is deliberately simple: hours × cost × 52 weeks × the automatable share. For the «pays for itself in» figure I use €1,200 as a reference, which is what automating one specific process from start to finish usually costs. The ones that touch several programs at once go to €2,000 or more. Your case is only known after looking at it, and I give it to you fixed and in writing before we start.
What actually gets automated
These are not ideas: they are the tasks I am asked for most. The ones marked with IA need a language model because free text has to be understood; the rest are rules and connections.
Orders and delivery notes
From the email or the website into your management software, with stock taken off and the customer told the real timescale.
With nothing retyped
Appointments and reminders
Confirmation on booking and a reminder the day before by WhatsApp. It is the single biggest thing for cutting no-shows.
Fewer empty gaps
Reports nobody wants to do
The Monday summary, the month-end close, the list for the accountant. They generate themselves and arrive in your inbox.
From Friday afternoon to automatic
Sorting and answering emails IA
They are read, sorted by type and urgency, and the repetitive ones get answered with a draft you approve.
The inbox, in order
Getting your programs talking to each other
Till with management software, website with stock, form with CRM. The bridge that today is a person copying things across.
Zero double typing
Repetitive quotes
If your quotes always follow the same pricing logic, they get generated from a form in a minute.
From half an hour to one minute
Alerts before it is too late
Low stock, an invoice unpaid at 30 days, a customer who does not come back. You find out while you can still do something.
Stop being the one watching
Summarising meetings and notes IA
From the recording or from scattered notes to a summary with tasks and who is responsible. Useful if you work job by job or visit by visit.
No writing notes up afterwards
Not everything needs artificial intelligence
This is where quotes get inflated most in this sector, so let me be clear: most of what a small business has too much of gets fixed with rules, and rules are cheaper, faster and do not make mistakes.
With rules
If this happens, do that. A hundred per cent predictable.
- Moving data between programs
- Generating invoices, reports and delivery notes
- Alerts by date, amount or quantity
- Appointment reminders and follow-ups
Cheaper, instant and with no cost per use. If your case fits here, here is where it stays.
With AI
When you need to make sense of text or images that arrive in a mess.
- Reading invoices and receipts in any format
- Sorting emails and messages by what they say
- Drafting text in the way you write
- Summarising long documents or recordings
It has a cost per use and gets things wrong now and then, so it always has a human review step. I never let it decide about money on its own.
What it connects to
If what you use has a way of connecting, it gets connected. If it has not, I check before promising anything — and I tell you no if it cannot be done.
HoldedFacturaDirectaOdooSage
A3 / Wolters KluwerExcel and SheetsGmail / Outlook
WhatsApp BusinessGoogle CalendarStripeRedsys
WooCommerceShopifyVerifactu / TicketBAICorreos and MRW
Fair questions
Am I going to have to let someone go?
In the small businesses I work with, no: what happens is that person stops doing the boring part and starts doing what is needed, which is nearly always looking after customers better or chasing what is owed. If your plan is to cut staff, I will not lie to you either: it will depend on how much of the job was typing. But it is worth saying before we start, because the team finds out and automation needs them to cooperate.
What if the automation gets it wrong?
It gets things wrong, sooner or later. That is why everything I build has three things: a log of what it did and when, an alert when something does not add up instead of carrying on regardless, and a final human check on anything that touches money. The dangerous part is not that it fails, it is that it fails quietly.
I have got an old program that does not connect to anything.
It happens a lot, and sometimes it can be done anyway. There are three routes: if the software exports to Excel or CSV, that is enough; if it has an accessible database, it gets read from there; and if there is nothing, what is left is automating everything around it and leaving that program as an island. What I am not going to do is sell you an impossible integration. Tell me which one it is and I will look before quoting.
What does it cost and how long does it take?
One specific, well-defined process is usually between two and four weeks. The price depends on how many systems have to be touched, so I start with a short, fixed-cost analysis that ends with a fixed quote in writing. If the saving that comes out of the calculator above does not pay for that quote in less than a year, my advice is not to do it.
Does my business data end up in the hands of an AI?
Only what is essential for the task, and I tell you exactly what. When a model has to be used, I work with providers that do not train on their paying customers' data, and for anything sensitive we can use a model running on your own server, though it costs more. And very often the answer is that no AI is needed at all, at which point the problem disappears.
Who maintains it afterwards?
Automations break when the software at the other end changes something, and that happens a couple of times a year. It comes with a month of adjustments included and, if you want, monthly maintenance that covers watching it and fixing it. It works without maintenance too, but the day it breaks you will see it before I do — which is why the alerts are part of the build.
I have only got one small task. Is it worth bothering you?
Yes, and in fact it is the best way to start. One small automation that works is worth more than a big plan that never gets going, and it lets us check whether we work well together before taking on something big.
Let us talk
Tell me what is done by hand
You do not need to know how it gets solved. If you describe the task, who does it and how long it takes, I can already tell you whether it can be automated and where to start.