AI Transparency Policy

1. Our Commitment

ProcureLabs uses artificial intelligence to help procurement teams work more effectively. We believe you should understand exactly how AI is used in the platform, what data it processes, what decisions it can and cannot make, and what controls you have. This policy explains all of that.

2. AI Features

The platform includes an AI-powered procurement advisor called Pala. Pala provides:

• Spend analysis and category insights
• Supplier risk assessments and recommendations
• Contract analysis and clause extraction
• Transaction integrity and value leakage detection
• Sourcing strategy recommendations
• Natural language Q&A over your procurement data

Pala carries out analytical and preparatory work (analysing, scoring, ranking, drafting) on its own. It does not approve spending, create purchase orders, make payments, sign contracts or send them for signature, or send external communications unless a person confirms the action.

3. AI Providers

We use multiple AI providers to deliver different capabilities:

4. Data Handling

When you use AI features, relevant context from your procurement data is sent to AI providers for processing. Here is how that data is handled:

• OpenAI processing OpenAI supports chat, analysis and bounded Decisions judgements for tool selection and procurement evidence. Subject to current customer permission and the agreed regional configuration, Decisions receives minimised, redacted context and questions for the permitted purpose. OpenAI API data is not used for model training unless explicitly opted in. Decisions can retain encrypted prompt-cache state for up to 24 hours and standard abuse-monitoring logs for up to 30 days, subject to applicable security, safety or legal exceptions.
• PII redaction. Before prompts are sent to AI providers through the platform's AI orchestrator, which carries the assistant and most AI features, the platform automatically redacts email addresses, phone numbers (US formats, and any number written with a + country code), US Social Security numbers, payment card numbers, and labelled bank account, routing and passport numbers. Personal names are not detected, and some background jobs and whole-document extractions send content to providers without this step.
• Permission for Decisions processing Decisions processing requires current customer instructions and permission for the specific purpose, account, processor and US transfer. Disabling AI or withdrawing the applicable authority prevents new Decisions requests. Consequential actions remain subject to the required safeguards and independent review.

Retention follows the applicable data category, customer instructions, legal holds and provider terms. Chat history, derived content, audit records and provider logs have separate retention conditions. Deletion requests are handled under the DPA and applicable law.

5. Autonomy and Human Oversight

ProcureLabs operates on a graduated autonomy model. The current platform setting restricts AI to analytical and advisory actions only:

No AI feature can approve spending, execute payments, sign contracts, or send external communications without a human confirming the action. This is enforced at the platform level and cannot be overridden by users or by your organisation's administrators under the current platform setting.

6. Accuracy and Limitations

AI-generated content is probabilistic and may contain errors. You should:

• Treat all AI outputs as recommendations, not decisions
• Verify numerical claims, supplier assessments, and risk scores against source data
• Apply professional procurement judgment before acting on AI suggestions
• Report inaccurate answers from the AI assistant using the feedback buttons on each of its responses

We do not guarantee the accuracy, completeness, or fitness for purpose of AI-generated content. See our Terms of Service for liability limitations relating to AI features.

7. Bias and Fairness

AI models can reflect biases present in their training data. We run statistical fairness checks (the four-fifths rule, chi-square and Kolmogorov-Smirnov tests) on supplier performance scores, comparing groups of suppliers by size, region, country, years in business and diversity classification where your organisation has recorded those attributes. Text written by the AI assistant is not automatically checked for bias. If you observe outputs that appear biased (for example, systematically disadvantaging suppliers from certain regions or of certain sizes), please report it to support@procure-labs.com. We investigate all reports and adjust our systems where bias is confirmed.

8. Audit Trail

Every decision the platform makes on an AI-proposed action (carried out, sent for approval, or logged in shadow mode) is recorded with the action, the autonomy level in force, the person accountable (the user who asked), and hashes of the decision's input and output. Records cannot be edited or deleted, and your administrators can view them in AI Governance > Decision log. We store digests rather than full text by design: it gives you provenance you can audit without a second copy of your data.

9. Your Controls

• Opt out entirely: Switch AI processing off in Settings → Account — for yourself, or for your whole organisation if you are an administrator. The cookie banner does not control AI processing.
• Review AI activity: Administrators can view the log of decisions on AI-proposed actions in AI Governance > Decision log.
• Request deletion: Request deletion of AI interaction logs by contacting support@procure-labs.com.

10. Regulatory Compliance

We designed the platform's AI governance with the EU AI Act, Singapore PDPA, and GDPR Article 22 (automated decision-making) in mind. The platform does not make legally or significantly impactful decisions about individuals without human involvement. For further detail, see our Privacy Policy and Security page.