2024 is the Year of Invoice Automation AI

Regarding the processing of purchase invoices, we are in a better situation than ever before. When we introduced artificial intelligence for automatic purchase invoice coding and routing for approval process in 2018, no one else was offering it. This also meant that there were no previous experiences in the market of utilizing AI in purchase invoice automation.

This allowed us to introduce a completely new service model. We heard a lot of feedback from the market about some of the system providers’ model, where a system is provided to the customer, and the customer is left on their own to build automation. Every service request to the provider incurs separate costs, and there is significant room for improvement in the quality of service.

We wanted to be a different supplier. We wanted to be paid for the added value we bring, i.e., higher purchase invoice automation. We do not charge customers separate hourly rates; all costs are included in the service. We knew that it was a significant risk for us to forgo charging for AI’s erroneous predictions, but it has been a clear added value for us, making our service the world’s best-functioning purchase invoice AI, with a large number of demanding customers.

We have spent a lot of time helping various organizations understand the possibilities and practical logic of AI because there was no previous experience. We are very grateful that we have been able to grow together with you towards the era of AI. Our smart customers have helped us steer the product in the right direction, and we have learned through shared experiences over the years.

As a reward for our efforts, we received significant recognition from Google when we won the Customer of the Year award in 2023.

Don’t settle for low-level purchase invoice automation

In some organizations, purchase invoices are still processed almost entirely manually. Did you know that every manually processed purchase invoice takes up to more than ten minutes per invoice for your organization?

This work involves the business side, the reviewer of the invoice, their supervisor, and the accounts payable personnel from the finance department.

Often, the responsibility for purchase invoice accounting lies with the business’s invoice reviewer. However, frequently, the correction work for accounting falls under the responsibility of accounts payable. This is simply because business personnel are not trained in purchase invoice accounting and often do it incorrectly. Accounts payable often corrects invoice accounting before transferring them to the accounting system, leading to double accounting for many invoices.

If accounting errors are not corrected by accounts payable, many invoices end up in the accounting system with incorrect accounting. This is also reflected as inaccurate information in procurement reporting and, for example, emissions reporting if invoice accountings are used for that purpose.

Rule-based automation 

Rule-based automation for purchase invoices is history. Anyone familiar with building purchase invoice automation using rules knows that achieving a truly high level of automation with rules alone requires a small miracle.

It is downright surprising that some still think that building rules alone can achieve a high level of automation. We have explored the purchase invoice processes and automation levels of a vast number of large companies, and no organization we know has built a functional automation system with rules alone. The main reason is likely that purchase invoices are not so repetitive and clear-cut that they can be automated with simple rules, only to a small extent.

Rules have their role in clearly repetitive purchase invoices, but they simply are not enough for building high automation. There are several reasons for this, which you can read about, for example, in this blog.

AI has brought a new logic to implementing purchase invoice automation

Purchase invoice AI is completely superior in terms of results compared to traditional rule-based automation. The logic of AI is the opposite of rule-based automation.

While in a rule-based model, rules are built first and based on a single data field, AI learns automation rules directly from the historical purchase invoice and accounting data. In other words, AI does not require any rules to be created for it; it learns the rules automatically from historical data.

Once AI is trained with historical data, it performs well immediately after implementation. Therefore, AI brings a tremendous leap in purchase invoice automation right after implementation, whereas rule-based automation is built slowly and painstakingly.

The capabilities of AI are continuously monitored through analytics to stay abreast of its prediction accuracies and detect if AI is making repetitive errors on certain invoices. This way, we are constantly aware of AI’s performance and see how well its capability improves over time from a good starting point.

What to consider when choosing AI services?

Some systems already offer some level of AI for purchase invoice accounting. Some companies have decided to adopt these in-house capabilities with varying results. Before adopting these, there are a few things to consider.

AI operating principle

• Is it genuinely an AI solution?
• what does AI do? Is the AI learning and evolving?
• Are AI models trained with historical data, and does it start off well?

Prediction accuracies

• How accurately can AI predict invoicing and circulation? Good prediction accuracy is in the range of 80% – 95%
• Can the AI service perform globally at a high level?

Testing AI capabilities before making a decision

• Does the supplier offer the possibility to test AI’s prediction accuracy with your historical purchase invoice data before making a decision? If it is a genuine AI service, this should be an easy task for the supplier.

Service scope

• Does the AI service only perform invoice accounting, or can it also handle invoice circulation?
• Does AI also utilize images of invoices for predictions? PDF invoice automation is not possible if AI cannot read invoice images and use the content for predictions. The file produced by a scanning service is not sufficient for high automation because it lacks a lot of data content.
• Can AI extract field information from invoices? For example, project numbers or other relevant information.
• Does the supplier provide real-time analytics for monitoring AI capabilities? This is critical for monitoring AI’s operation.
• Are there limitations on the number of predictable fields in the service?
• Does AI provide a confidence value for each prediction separately?
• Can you set threshold values for confidence scores dimensionally?

AI service pricing

The fairest pricing model for AI services is paying for correct predictions. This way, the supplier’s success also depends on the effectiveness of the service.

Maintenance of AI service

• Is adding new dimensions easy?
• Is removing dimensions easy?
• Changes in dimension values – e.g., changes in the chart of accounts or cost center structure. Can you update threshold values for confidence scores?
• New companies or companies that are no longer in use.
• Does your supplier charge separately for AI maintenance, or is it included in the service price?
• AI needs to be updated frequently during use, and if you have to pay for every small update, it will become costly for you.
• Does the supplier offer proactive maintenance, or do you have to create a ticket for every request?
• Traditional service models work poorly for AI service maintenance.
• If you decide to change your purchase invoice system, does the AI come with it to your new purchase invoice system?

Embark on the journey towards automation together

We are fortunate to serve a large number of customers. We process well over ten million purchase invoices annually, and the volume is growing rapidly. At the beginning of the year, we are set to bring a significant number of new Snowfox customers into production. This is exciting as we get to see how the purchase invoice automation of new customer companies reaches a high level finally.

Our customers may have built automation based on rules for years, and now in the era of AI, they can let AI handle a large part of automation. This is a relieving but trust-demanding journey for many of our customers. Fortunately, AI’s capability can be easily monitored through real-time analytics, maintaining a sense of control over the process from start to finish.

Finally, a big thank you to our brave first customers who trusted us when they understood the potential that AI brings. Without you, we could not have collectively taken purchase invoice automation into a new era. This journey is just beginning, and there is much more to come.