How we use AI: Pleo's Spending Guidelines Feature

Spending guidelines enables admins to share a company’s spend guidelines with employees directly via Pleo. The guidelines are based on a company’s existing spend categories, like “Meals and Drinks” or “Travel”. The feature is powered by generative artificial intelligence (genAI) and allows admins to make use of this technology when writing the guidelines. They can also write the guidelines manually. Furthermore, it allows employees to question the guidelines in natural language (Example: what can I spend on client dinner?) and generative AI  will source the answer from the guidelines to the asked question. If the answer generated isn’t satisfactory, there’s a feedback mechanism to share this back to us at Pleo. This allows employees to get questions about how they can spend money, without having to bother the admins/finance teams. Guidelines can always be edited or deleted by admins, to ensure they reflect the company’s up-to-date spend policy.

What is the Value of Spending Guidelines for the Company?

Spending Guidelines offer numerous benefits:

• Clear Guidelines: Employees receive clear, role-specific guidelines for acceptable spending, minimising ambiguity and potential non-compliance issues.
• Consistency: Ensures that spending policies are consistent, up-to-date, and tailored to the company’s specific needs.
• Error Reduction: Defined guidelines reduce the likelihood of spending errors and financial mismanagement.

In addition, AI-Enabled Spending Guidelines offers the following benefits:

• Streamlined Policy Creation: Pleo automates the creation, management, and updating of company-specific expense policies, reducing the workload for administrators.
• Natural Language Q&A: AI-enablement provides natural-language answers to employee questions regarding Spending Guidelines, improving communication and clarity. 

How is AI Used?

Pleo leverages Microsoft Azure’s OpenAI LLM (large language model) to provide AI functionality within the platform. The LLM is not trained on company data, but provides insights based on what is available within the Customer’s instance.  For this purpose we share company information, company name, company location. To generate guidelines we use the list of categories and accounts for each category that the company has.

We share that information as part of the prompt we use to generate text from the LLM. For production data we only use Azure OpenAI, and the version we’re using is deployed within the EU in Sweden. The data is used for the LLM to generate the result and it isn’t used to train the LLM.

AI models can be employed to automate the generation of spending guidelines. When doing so we do the following:

• Data Processing: AI processes company-specific data (e.g., accounting categories, size, location) and role data*. 
• Automated Decision-Making: The AI models generate recommendations, simplifying the policy creation process while ensuring accuracy and relevance.
• Admin is in control: the admins will always be the ones who publish the guidelines, and can decide whenever to make use of the AI to edit, update or publish a new guideline.
• Secure Storage: All data is stored securely using Pleo’s cloud infrastructure on AWS (Amazon Web Services), ensuring confidentiality and protection from misuse.
• Personal Data: No personal data is being processed by default. Since the feature is based on free-form text input, we will process what the user inputs.

This approach helps businesses maintain effective and efficient spending policies while leveraging the power of AI to adapt to changing organisational needs.

*Role data does not include personal data from employees, only job titles and team names.