Artificial intelligence has brought the biggest leap in development for purchase invoice automation in the last decades. The change is so significant that it even seems unbelievable. I still often hear comments from our customers in the negotiation phase that 'this sounds too good to be true.'
What kind of change does artificial intelligence bring to purchase invoice automation?
Let's first briefly return to current automation methods. Traditionally, we've become accustomed to performing purchase invoice automation based on rule-based automation. This means that every automation rule is manually created in the purchase invoice system. For example, a coding model is created for invoices from a specific supplier, such as DNA, suggesting a predetermined coding for these invoices. If you're purchasing only one product or service from a single supplier, this rule might work. However, this situation is rarely the case, and the rule doesn't work as intended.
The more complex your environment is, the more and more complex rules you need to be able to construct to get automation to work to some extent. For example, a company with dozens or even hundreds of entities will likely need to build and maintain automation rules for each entity separately. This alone adds a tremendous amount of work and challenge to building traditional automation.
What factors make the purchase invoice environment complex? Complexity is influenced by factors such as:
Purchase invoice volume Number of entities Number of suppliers Number of tracking items / dimensions Number of values for individual dimensions (e.g., number of different General Ledger account values) Number of invoice approvers and approvers Additionally, for rule-based automation to work, purchase invoice data must be of very high quality, and certain reference details must be in their respective fields. For instance, in contract matching, the contract number must be in the right place on the invoice because the contract number triggers the automation rule. So, it's not enough to have entered the contract into your purchase invoice system; the sender of the invoice must also commit to placing the correct contract number in the right place.
The challenge with rule-based automation is that humans have to manually construct each automation rule and also maintain these rules. Building automation requires a highly capable individual and a significant amount of work hours to create the automation rules. Commonly recurring purchase invoices and the underlying logic need to be identified. Furthermore, exceptional cases of these invoices and even exceptions within exceptions need to be understood, so that there is no need to constantly intervene in the rule suggestions.
One of the most significant aspects related to the proliferation of AI is the fact that adopting AI is very straightforward. You don't need to create any automation rules for AI; rather, the purchase invoice AI learns processing logics from historical purchase invoices and their coding information automatically. Thanks to this logic, purchase invoice automation becomes automated for AI.
The adoption of purchase invoice AI begins with retrieving previously processed purchase invoices from 3 to 12 months. The AI goes through these invoices one by one, noting all the information from the purchase invoice message and the purchase invoice image, and observing how the invoices were coded and who reviewed and approved them. After going through all the invoices with their coding and reviewer and approver information, the AI begins to create logics on how the invoices should be coded. The AI recognizes correlations between the information on the invoice and the coding, considering even the smallest nuances.
For example, the AI easily recognizes that a specific system cost should be coded to specific dimensions and that the invoice should be sent for review and approval to a certain person. Notably, in the training of AI, it truly utilizes all the data from the purchase invoices. It understands, for instance, that the address information visible on the invoice could be the most important information for selecting the cost center, etc.
The purchase invoice automation brought by AI operates dynamically. This means that AI continuously modifies its automation rules automatically, adapting to changes in the environment. Whereas rule-based automation requires constant maintenance, AI maintains itself automatically.
Purchase invoice AI automatically receives feedback messages from the purchase invoice system for each invoice it processes at the stage when the invoice has gone through the approval process and is being transferred to accounting. Using the feedback message, AI compares its coding suggestion to the coding completed after the approval process and checks if any changes were made to the suggestion during the process. If all dimension values are the same as what the AI originally suggested when they transfer to accounting, this is a sign to AI that all dimension values were correct. However, if, for example, the cost center was changed during the process, AI understands that the suggestion for the cost center was incorrect and that a correction is needed. Therefore, AI automatically learns from its mistakes and improves constantly, becoming better and more confident.
For the reasons mentioned earlier, building and maintaining purchase invoice automation is becoming easier and more effortless over time. There's no need to spend as much time on constructing and maintaining complex automation rules because AI automates the construction and maintenance of automation.
It might still be sensible to handle the easiest recurring invoices with methods like contract matching, but it's much easier and more cost-effective to leave the more complex automation rules to AI. Moreover, based on the results, AI outperforms human automation builders by a wide margin.
Explore the possibilities of AI more thoroughly through the webinar we hosted