How ImpactBI protects people
Every row is a person.
ImpactBI is built to help nonprofits serve people better, and to protect them while it does. Here’s how we handle privacy, ethics and access.
Privacy
Names and identifying details stay out of answers.
- Set aside the moment a file arrives. Names, contact details and ID numbers are separated from the data Data Assistant analyzes.
- Names never reach the AI. Even quotes from surveys and notes are anonymized before Data Assistant reads them.
- Revealing someone is deliberate. Only staff who are allowed can show a person’s name inside ImpactBI, and every reveal is logged.
- Downloads follow your role. Files with personal details can only be exported by admins, leads and editors, never by viewers.
Ethics
Sensitive topics follow a built-in ethics guide no organization can turn off.
Data Assistant checks itself on every question. The guide is grounded in established data-equity and responsible-data practice and current US civil rights and privacy law, and it’s reviewed every quarter.
- You ask a questionIn plain language, about your own programs and the people you serve.
- It checks its reasoningBefore and while it answers, Data Assistant holds its work up against ImpactBI’s ethics guide.
- People stay protectedThe answer protects the people in your data, and tells you when the guide shaped it.
Protected in every answer.
- Sensitive groups stay anonymous. Labels like race appear as Group A and Group B until a staff member who’s allowed reveals them.
- It never guesses. Sensitive details come only from what people chose to share about themselves.
- It looks for causes, not traits. Gaps between groups are explained by conditions like cost, distance or access, never by who people are.
- People decide, not the AI. Data Assistant informs your staff’s judgment. It never makes eligibility or service decisions.
Some questions it won’t answer.
Data Assistant refuses to respond in ways that could endanger a group of people, or reduce them to a single trait.
- It won’t sort, rank or exclude people because of who they are.
- It won’t flatten a group of people into one simple story.
- Questions that could put people at risk stop before any data is touched. No one can override a stop; it can only be escalated to administrators.
Extra care where data has done the most harm.
The guide gives Data Assistant specific care for the communities most often misread, exposed or left out by data.
- Race and ethnicity
- Immigration status
- Disability
- Health and mental health
- Survivors of violence
- Children and youth
- LGBTQ+ people
- Religion
- People with criminal records
- Older adults
- Language communities
- Neighborhoods
Access
Each person sees only the files they’re allowed to open.
- Invitation only. No one can sign themselves up. An administrator invites each person and chooses their role.
- Roles and teams. Admins, leads, editors and viewers each see and do only what their role allows. Files can be shared with everyone at your organization or with specific teams.
- Data Assistant follows the same rules. It can only read the files the person asking is allowed to open.
- Sharing is careful. A shared insight only appears for people who can open every file behind it. To reach anyone else, an administrator publishes a snapshot with quotes removed.
Security
Protections built into the product, not just its promises.
- EncryptedData is encrypted in transit and at rest.
- Enforced at the sourceAccess rules live in the database itself, on every record, so no screen or shortcut can get around them.
- LoggedSensitive actions, like revealing names, publishing insights and escalations, are recorded.
- Not used for trainingYour organization’s data isn’t used to train AI models.
Want to see how it works with your own programs?
This page explains our approach. It isn’t legal advice.