You automate your invoice processing in Zenvoices by combining different technologies. This way, you stay in control of what you automate and you book according to your own booking policy. With many companies, there's an extra factor to consider, choices made at organisation level apply across all companies.

In this article, you'll read which parts belong to this process. You'll also read how they connect, and how you decide which technique to use when.
​

Are you working with a solution that only creates a booking proposal? Then that's not automation yet. Someone still reviews every invoice.

Automation goes a step further. An invoice comes in and is automatically routed to the correct journal. The system recognises and books the invoice. Only in case of a deviation does someone review the invoice. The automation rate therefore shows how many invoices are processed without manual action.

You'll see that difference reflected in the following three percentages on your performance dashboard. They look similar, but each measures something different..

The first two percentages show how well the digital processing works. The third percentage shows how much of that is actually processed automatically. Good recognition, therefore, doesn't automatically lead to a high automation rate. It is an important precondition though, the booking proposal needs to be correct first, before you can automate further.

The three percentages also help you determine where improvement is needed.

There's one key starting point when building automation, use the simplest method that's reliable enough. Only add complexity when volume, booking complexity or the required certainty calls for it.

Automating as much as possible doesn't mean you should configure as much as possible. Especially with many companies, an overly extensive setup can actually become difficult to manage. Hundreds of rules, templates and models require maintenance. They also make it harder to understand why a booking is processed a certain way.

That's why you should build the setup in layers.

The first three layers scale well. The management burden doesn't grow with them straight away. The fourth layer is different. You have to build and maintain every recognition field, model or specific setup yourself. So only use this layer when there's a clear reason to.
​

The route to automatic invoice processing consists of three steps, standardising, centralising and automating. Standardisation makes central setup possible. From that central setup, you then automate further.

Accounting firms often manage many companies. Over the years, these have been set up with different general ledger accounts, VAT codes and exceptions. If you set up booking logic per company and each employee uses their own approach, differences between companies will emerge. As a result, data quality becomes less predictable. The setup also becomes harder to scale.
​

Office costs are booked to account 4300 in one company and to account 4500 in another. If most of your companies use the same chart of accounts, a booking rule can refer directly to a general ledger code. That code means the same thing across all those companies.

If a company has a different chart of accounts, there are two alternatives, RGS or general ledger tags. You add these to the same booking rule, alongside the general ledger code. So you don't need to create a separate rule.

ESo a different chart of accounts doesn't have to stand in the way of central setup. You set up the deviating company once. After that, you can apply the same central rule there too.

Tip: record as much as possible at organisation level, this way the same setup applies to multiple companies. Only record genuine exceptions at company level. Also ask yourself whether an exception is really necessary. For example, does an invoice really need to be split across ten cost accounts? And do you actually need that level of detail in your reporting?
​

Zenvoices forms a central layer on top of your accounting package. Invoices arrive in one place. You decide centrally how they're processed, even if you work with multiple accounting packages or organisations.

The parts aren't isolated from each other. Each step uses information from the previous steps.

Note: a part later in the chain doesn't solve a problem that started earlier. An illegible document, for example, doesn't become clearer because of an extensive booking template. Missing booking logic also isn't solved by simply changing the recognition threshold.
​

You set the rules. Zenvoices provides the tools to apply them.

Recognition already works from the very first invoice. You start with a general model, trained on millions of invoices. If a processor corrects something, the model learns from that within the company. The recognition score shows how confident the system is about a recognised field.

Corrections during the initial period, therefore, help the model get to know the company better.
​

Note: bookings that are approved automatically don't train the model. The recognition model only learns from user input. If you book everything automatically too soon, the model gets fewer corrections to learn from.
​

Every document ends up in one of four situations, "Upload & read", "Not complete", "To review" or booked automatically. Two conditions apply for automatic booking, there must be enough booking information available, and the invoice must pass the configured checks.

A recognition score gives an estimate of certainty for each recognised field. For example, do you see an invoice number, invoice date or amount? Then you see not only the recognised value, you also see how confident the model is about it. You can use this score as a condition for automatic processing.

Two points matter here. The system provides the score for four basic fields, invoice number, invoice date, amount and VAT percentage. For a PDF, the maximum score is 99.9%, the model always allows for a margin of error.

A higher score means the model is more confident in its prediction.
​

The more you automate, the more important managing the setup becomes. The principles below help keep that setup clear and organised.

The performance dashboard shows you, per company, user, vendor and segment, where corrections or manual actions are still frequent. Tags help you with this.
​

Make a fixed working agreement for this too. Does a correction come up more than once? Then make a choice, do you set up a rule for it, or do you consciously accept it as an exception? This prevents recurring manual work from continuing without anyone having consciously decided on it.

Related articles: