AI Principles

AI, built for your mission

Nonprofits run on trust. The people you serve, and the donors who fund the work, share some of their most personal information with you, and that trust extends to us. We build AI to help your team do more good in less time: to surface insight, draft the first version, and reduce busywork, with you in control. We also believe AI should never come at the cost of the trust your mission depends on. This statement sets out the principles we hold ourselves to, and the practical commitments behind each one.

Our principles

1. Your data, protected to the same high bar

Commitment: Your data is used to support your organization, not to train outside companies' AI models, and AI is held to the same security and compliance program as the rest of Velora.

We follow a privacy-by-design approach: customer and donor data is protected at every step and governed by the same security and compliance program as the rest of Velora, including our SOC 2 Type II controls, so AI never lowers the protections already in place. What we retain, who can access it, and where it's processed are set out in "How we handle your data" below.

2. You stay in control, and we stay accountable

Commitment: AI assists, suggests, drafts, and acts within approved workflows, but you stay in control, and we review every AI feature before release and stay accountable after launch.

Where available, our AI features let you see what they did and step in, reviewing or overriding what any feature or agent does, and a person is always accountable for weighty decisions about a donor, gift, grant, or dollar. Every feature is reviewed and approved internally before launch, then monitored and updated to stay secure, reliable, and useful, and we give reasonable notice before a material change to these principles, though the models or providers behind a feature may change quickly, sometimes without notice, to keep it secure and available.

3. Fairness for everyone you serve

Commitment: We test for obvious failure modes and monitor AI features so they do not create unfair outcomes.

We review AI features for clear risks in their inputs, outputs, and workflows. We test for obvious failure modes, monitor for regressions, and use human review for higher-risk workflows. Our goal is to help every organization and community we serve use AI safely and fairly, without creating outcomes that unfairly disadvantage the people behind the data.

4. Transparency you can pass on

Commitment: We explain what our AI does, what data it uses, who helps power it, and what controls are available.

We tell you in plain language what our AI features do, what data they use, and what controls are available to manage them. We keep a current AI Subprocessor List for the subprocessors that help power our AI, maintained separately so it stays accurate as providers and tools change, and we notify customers before adding a new AI subprocessor where our product terms or subprocessor process require it.

5. We use AI where it genuinely helps

Commitment: We use AI to make useful workflows faster and better, not to add AI for its own sake.

We believe AI should be proportional to the problem it solves. We use it where it can reduce manual effort, surface useful insight, or make an existing workflow meaningfully better. We do not believe every problem needs AI, and we will keep evaluating whether each AI feature is useful, safe, and worth the trust customers place in it.

What we mean by "AI feature"

An AI feature is any Velora capability that uses a large language model (LLM, the kind of AI that powers chat assistants and generative tools), a generative model, or a machine-learning model to produce user-facing content, recommendations, insights, decisions, or actions.

This includes in-product assistants, intelligent suggestions, agentic workflows, MCP servers, and integrations with third-party AI tools. It does not include basic rules-based automation or traditional software logic unless that capability uses a model to generate, recommend, classify, or act on customer data.

AI features come in two shapes:

• Velora-hosted AI: features we run for you. For most of these we operate models provided by others; for some Keela and Raisely features we also build and train our own machine-learning models. Examples include the Aplos and Raisely assistants, Raisely Intelligent Amounts, and Keela's intelligent suite.
• Third-party AI you connect: you bring your own AI tool, such as Claude or ChatGPT, and connect it to a Velora product through a Velora MCP server (Model Context Protocol, an open standard for connecting AI tools to software). Today this means the Raisely MCP server. For these connections, our commitments cover what Velora logs and stores on our side. What your chosen tool does with the data is governed by your own agreement with that provider.

How we handle your data

• At the model layer, providers do not keep your data. Velora-hosted AI features route through a secure gateway configured for Zero Data Retention (ZDR). That means our model providers do not store your prompts or outputs, or use them to train, fine-tune, or improve their models, and your prompts and outputs are not retained at the model-provider layer once the request completes.
• On Velora's side, we keep only what we need to operate and improve the service. Like any software service, we may retain prompts, outputs, logs, and feedback so we can operate, debug, secure, support, and improve Velora AI features. This data is managed by Velora and authorized subprocessors under our security controls.
• Access is restricted and purposeful. Access to your AI prompts, outputs, and logs is limited to authorized roles that need it to support, secure, debug, or improve the service. Access is logged where technically supported and retained for a limited period based on the feature and system involved.
• Some features learn from your feedback. When you correct an AI feature, for example by recategorizing a transaction or giving feedback on an answer, we may use that correction to give you better results in your own account.
• We use aggregated usage signals to improve features. We look at overall, aggregated patterns in how features are used and perform so we can make them better. When we review actual conversation or prompt data to debug or improve a feature, it stays within Velora and our authorized subprocessors under the access controls above, and we never sell it or use it to train outside companies' models.
• We never expose one organization's data to another. Your records, donors, and conversations are never shared with another organization or used to give anyone else access to your private information. To operate and improve features we may analyze usage at an organization level, but this is never used to reveal one organization's data to another.
• Where your data is processed. Delivering AI features relies on specialized providers, and some AI processing may take place in the United States or other regions rather than only in your organization's home country. We choose subprocessors that meet our security and contractual standards, and we list them on our AI Subprocessor List so you can see who is involved and where they operate.

Accuracy and human judgment

AI output can be wrong, incomplete, or out of date, so please review it before relying on it for any financial, donor, legal, or operational decision. Always check important output against your own records, which remain the source of truth, and make sure a person makes the final call on decisions that carry weight.

Your controls

Controls vary by product and feature. Where available:

• Manage access. Your account administrator controls which AI features are available to your organization, and per user where we support it.
• Opt in or opt out. Some AI features are opt-in. Where supported, you can disable or opt out of an AI feature.
• Review activity. Where available, you can review AI activity history, outputs, and actions.
• Access, correction, and deletion. Requests about AI-feature data follow the same paths as the rest of your data. Some operational data we use to run and improve features, which is not tied to an individual person, may be retained as described above. See your product's Privacy Policy.

Using AI responsibly

We ask customers and users not to use Velora AI features to:

• make consequential decisions about donors, beneficiaries, or staff without meaningful human review;
• break fundraising, donor-protection, financial-reporting, or privacy law;
• submit special-category data unless the relevant product note supports it;
• reverse-engineer, jailbreak, or extract training data from a feature; or
• generate infringing, harmful, or unlawful content.

Full acceptable-use terms live in each product's Terms of Service.

Changes and notice

We may update these principles as our AI evolves. We'll give customers reasonable notice before a material change to these principles takes effect. As noted above, the models or providers behind a feature may need to change quickly to keep features secure, reliable, and available, and those changes may not always come with advance notice. Clarifications and minor fixes take effect when published.