Secure AI for businesses
Automatic Classification of Invoices, Requests and Emails
Every day, invoices, requests and emails arrive, and someone reads and sorts them by hand. AI can classify them on its own, and a person only reviews the doubtful cases.
- What arrives each day
- The AI classifies
- Review of doubtful cases
Who it is for
It is for you if…
- Companies that receive invoices, requests, emails or forms of all kinds. Today, a person reads them and sorts them by hand.
- Teams that classify, assign or distribute many cases by hand every month.
It is not for you if…
- Tasks that can be solved with fixed rules. In those cases, AI adds cost and uncertainty without adding any value. See the Systems integration and task automation page.
- Important decisions about people with no one reviewing them. We don’t do that with AI alone yet.
What you get
- Preliminary study: yes, no, or in part
Which part of the work the AI can do, with the expected accuracy, and which part still needs a person.
- Automatic classification of what arrives
The AI classifies or decides at the point where each case requires judgement. The rest runs on fixed rules.
- Human review where appropriate
A tray with the cases where the AI is not certain or the matter is important.
- Quality measurement and logging
A set of cases with known answers, a panel showing the AI's accuracy, and every decision is logged.
First, see what can be handled with rules and what can’t
Every job has a part that can be written as a rule and another part that requires thinking. The first is done automatically with rules. Only the second goes to the AI. Before building anything, a short study shows where that line is.
The AI decides, and a person reviews what’s doubtful
The AI classifies, extracts the data or proposes what to do. It also says how confident it is. If it’s very confident, the decision stays automated. If it’s unsure, or if the matter is important, a person confirms or corrects it. We agree with you how confident it must be to decide on its own. Everything else goes to the review tray.
Always measure, with data kept in Europe
Before launch, we measure how accurately the AI performs on cases where we already know the answer. Once it’s running, we keep measuring with a sample at regular intervals. Your data does not leave the European Union, and it is not used to train other companies’ AI.
What we build it with
The AI runs in European Union data centres. Your data is not used to train other companies’ AI. It is only fine-tuned with your data if you request it in writing, and then only for you. Personal data is minimised before it reaches the AI. The code is delivered to your repository, the place where it is stored along with its full history.
Glossary
- Classify
Deciding what type each item is and who should handle it. For example, whether an email is a complaint, an order or an invoice.
- Fixed rules
Instructions that are always applied the same way. "If the invoice comes from a new supplier, notify management."
- Review tray
The list of cases the AI has not decided on due to lack of certainty. A person reviews them.
- Training an AI
Giving it many examples so it learns from them. The providers we use don't do this with your data.
What problems it prevents
What can go wrong and how we prevent it.
- The AI makes a wrong decision and nobody notices
It only decides on its own when it is very confident. Doubtful cases go to a person. And we measure accuracy at regular intervals.
- Something that should not happen gets done automatically
The preliminary study will tell you so. And we'll say it even if the project ends up smaller.
- Your data leaving Europe or being used to train other companies' AI
Everything stays in the European Union. Providers do not use your data to train their AI. Personal data is minimised.
Frequently asked questions
The preliminary study is short and gives a clear answer: yes, no, or only in part. If you go ahead, the system is built in stages. We start with what reaches you most.
Yours. The code is kept in an account in your name, not ours, along with its full history. The data belongs to you and is not used to train other companies' AI.
An initial meeting to look at how it's done today and a sample of real cases. Afterwards, we send you the quote, with a fixed price broken down into stages. The preliminary study can be contracted separately.
It depends on the work and the data. We don't give a figure until we've measured it on real cases from your company.