Top 5 Concerns About Invoice Automation AI and How to Avoid Them
The process of handling purchase invoices is notoriously time-consuming and frustrating. Most invoices are coded and routed for approval flow manually. This manual work causes delays in payment and often results in coding errors. Handling purchase invoices is rarely anyone’s favourite task in business, so it’s often postponed until the last minute.
Usually, the individual responsible for coding invoices is the employee from the business unit who initiated the purchase. When non-accounting professionals handle this task, the quality of the coding can be quite poor, as you might expect.
Quite often, the accounts payable team has to do a final check on the purchase invoices before they are posted to the accounting system, primarily to correct coding errors. This means that a large portion of invoices end up being coded twice: first by the business unit and then corrected by accounts payable.
Rule-based automation hasn’t solved the problem
If it were easy to automate the purchase invoice process with traditional rule-based methods, we wouldn’t have these manual work challenges. Everyone would have automated their purchase invoice processes. However, rule-based automation has only managed to automate a small part of the process.
There are several reasons for this, which I’ve detailed in a previous blog post. In short, complex and frequently changing invoices can’t be automated with simple rules. You can read more about it in my other blog.
AI is an excellent use case for automating purchase invoices
AI has proven to be an excellent technology for automating the coding and routing of purchase invoices. With AI, you can achieve over 90% automation. Unlike with rule-based automation, you don’t need to create automation rules for AI; it learns everything from historical data.
However, there are several common concerns about AI. We frequently meet with representatives from large companies’ finance departments and discuss the use of AI, encountering some recurring worries.
Here are the top 5 concerns about AI in purchase invoice processing and some thoughts on how to avoid them.
Top 5 Concerns About AI in Purchase Invoice Processing
1. “AI is a black box – I have no control.”
Concern: I have no control or visibility into how AI works. I don’t know how well it actually performs or why it makes certain coding decisions.
Solution: A reputable AI service for purchase invoices always includes transparent, real-time analytics. This allows you to track exactly how well the AI is performing and how it has improved over time. If an AI solution is offered without analytics, it’s worth reconsidering.
2. “AI makes mistakes.”
Concern: AI makes mistakes in coding. If these mistakes go unnoticed, incorrectly coded invoices could end up in the accounting records.
Solution: It’s a common and understandable concern. The fact is, AI does make mistakes, but it makes significantly fewer mistakes than humans. AI bases all its predictions on historical data and doesn’t lose focus like humans can. AI also provides a confidence score for each prediction, indicating how certain it is. You can filter out predictions based on these confidence levels to ensure only the most certain predictions are used in your accounting.
3. “How do I choose the best AI solution?”
Concern: There are many AI options available for automating purchase invoices. How do I choose the best one for our needs?
Solution: While many solutions market themselves as AI, their capabilities can vary. Most AI solutions focus on extracting data from PDF invoices and mapping it. When it comes to AI that automatically codes and routes invoices for approval flow, there are fewer options, but they do exist. For smaller invoice volumes (under 10,000 per year), an internal AI solution within your purchase invoice system may suffice.
For larger volumes, a specialized, system-independent AI service is typically more accurate and capable. Providers that focus solely on developing AI services are more specialised and have broader resources compared to those that are developing and maintaining an entire purchase invoice workflow system. This specialisation results in more accurate AI predictions and broader capabilities. Another clear advantage of specialised AI modules is their system independence. If you switch your workflow system, the AI module can move with you from one system to another, ensuring a high level of automation.
When comparing AI modules, the best method is an AI capability test. This test uses your historical purchase invoice data, typically around 6 months’ worth, along with corresponding coding details and invoice approval chain information, to evaluate the AI’s prediction accuracy. You can find more details about this in another blog of ours.
4. “Yet another IT project?”
Concern: Implementing AI for purchase invoices is seen as an IT project, requiring significant time and resources.
Solution: Choose a packaged AI solution with ready-made interfaces for your purchase invoice system to ensure smooth implementation without an extensive IT project. The best solutions are straightforward to set up, requiring only the training of the AI with your historical data and connecting it to your system.
“Customers don’t need to worry about Snowfox.AI implementation being a massive undertaking. For us, it was a straightforward and short-lived project.” – Hanna Van Enk, CFO, Lumon
- “Do we have to adopt another new system?”
Concern: Will implementing AI mean introducing a new system we have to adopt?
Solution: Ensure that the AI solution you choose integrates seamlessly with your existing purchase invoice system. This way, the handling of invoices continues within your current system, but the process is automated with AI, without needing a new interface.
I hope these insights help you on your journey towards automating your purchase invoice process!