My company is looking at deploying a Gen-AI model to increase the rate of touchless invoices (Invoice-to-Pay). The main area of Gen-AI adoption is to automate as much as possible the completeness of the invoice post-OCR & derive the taxation (VAT code, VAT rate etc.) Does anyone have any experiences with such approach? Lessons learned?

Yes, we’ve done this before. If the invoice formats are consistent, accuracy improves significantly; otherwise, you’ll need to train the model for better results. With Gen-AI, there’s always a risk of hallucination, so I strongly recommend keeping a Human-in-the-Loop for such projects. Poor scans and complex layouts can pose major challenges.
VAT code and rate derivation works best when using contextual AI. Data quality is critical—poor scans or inconsistent supplier formats can still cause errors, so invest in pre-processing like image enhancement and skew correction.
Tax compliance adds complexity since VAT rules vary by region; Gen-AI models must be trained on localized tax data and updated regularly. Avoid over-reliance on AI—full automation without governance can lead to compliance risks, so maintain audit trails and exception workflows.
Finally, don’t overlook change management. Employees may fear job loss, so clear communication and training on AI-assisted roles are essential.

1. Ensure you have a consistent process and procedure for validation for those documents or fields. Inconsistent human in the loop validation can negatively impact the solution’s performance and training.

2. Ensure you understand the underlying model the solution is using and that it is a LLM based approach. Many vendors now are developing such an approach, but many have solutions that claim Generative AI when it’s not.

3. Ensure your business partners have updated their procedures to accommodate the solution now doing this work.