1. How is AI used at Yokoy?

AI models at Yokoy are used to simplify and improve processes within the spend management context. Currently, the primary use cases include:

• • Information Extraction: AI models automatically extract relevant information from expense receipts and supplier invoices during upload. This information is then used to prepopulate submission forms, reducing the need for manual data entry by employees.

• • Information Complementation/Matching: In cases where certain information is not directly available in the uploaded documents, AI models utilize client-specific historical data or data provided by clients to infer and prepopulate fields. Examples include selecting predefined expense categories and matching suppliers for invoices.

2. What data will be extracted by the AI?

AI models are solely used to extract objective information from documents (e.g. for expenses this includes country, currency, total amount, tax items and more) and not to extract or collect personal information about app users.

3. No decisions or judgements about the user made by Yokoy’s AI models

The AI models do not make any sensitive decisions or judgements about app users. If Yokoy decides in the future to implement AI models that assess user behavior or are trained on personal data of users, such models will only be activated with the explicit and documented consent of our customers.

4. No training of AI models with personal data

The term “training of AI models” refers to the process of providing historical input and output data to an AI model, enabling it to establish its own logic and rules for predicting the most likely output for a given input. AI models are developed and trained by Yokoy’s in-house AI team.

Yokoy’s AI models do not receive or use personal data as input, nor are they trained on personal data. During preprocessing, digitized images that may contain personal data are converted into purely technical, fully anonymized features. The AI models are trained and operate solely on these anonymized data/technical features.

Information about the submitter of an expense is provided only as a randomized user ID, which is used exclusively to enhance extraction accuracy based on previously submitted documents.

5. Information about the logic involved in the AI

Yokoy currently uses two types of in-house AI models: “pure” AI models and heuristic models. While the heuristic models are based on human/business logic (e.g. for recognizing the date or the time) and are “easy” to understand by humans, the “pure” AI models’ internal logic is not easily comprehensible by humans, which is why they are often referred to as “black boxes”.

To provide trustworthiness in the pure AI models, every model undergoes extensive backtesting (e.g. tested against historical data and in manual tests) before going live. Regular recalibrations and monitoring of the AI models ensure they remain up-to-date and functional. Root-cause analysis of erroneous model predictions is conducted as part of regular reviews to improve model performance and acknowledge limitations.

6. Usage of external Large Language Models

Yokoy may utilize third-party Large Language Models (LLMs) from trusted providers (e.g. Google Gemini on VertexAI), with robust terms and conditions for data handling in specific use cases. Yokoy prioritizes providers who classify model inputs and outputs as client data, strictly prohibiting their use for training or improving external LLMs.

To be clear: customer data is not used by third-party providers to train or enhance their AI models or services. Yokoy also applies stringent fine-tuning and recalibration rules, ensuring that no external LLMs are fine-tuned or trained on client data.

7. Strong collaboration between Yokoy’s legal and AI-team

Yokoy has established regular meetings and mandatory involvement processes between the in-house AI team and the legal team, which closely monitors legal developments. For critical decisions, advice from external counsel is also sought to ensure an unbiased outside view.

8. Continuous improvement of AI Models and features

Yokoy is committed to the ongoing development and enhancement of its AI models and features. This continuous improvement process ensures that the AI capabilities remain effective, accurate, and aligned with best practices in data protection and compliance. The AI Policy may be updated from time to time by publishing the new policy on the company’s website.

9. Conclusion

At Yokoy, our commitment to data protection, compliance, and responsible AI use is crucial. We continually assess and improve our AI models and practices to ensure they align with the highest standards of security, transparency, and user control.

By prioritizing best practices in data handling and offering robust customer controls, Yokoy aims to deliver innovative AI-driven solutions that empower businesses while maintaining trust and compliance. Download our AI policy here.

Last updated: April 2025