ResourcesWhite paper
What AI actually needs from your data
Most AI pilots in associations and nonprofits do not stall on the AI. They stall on a member who exists three times, a status field nobody uses the same way, and a renewal history that was overwritten every year. This paper explains what AI does with your records, what it needs from them, and what to fix first.
In short
- AI reads, predicts and retrieves. Each one fails on a different data problem.
- "One record" is a set of matching rules you write down, not a feeling.
- Inconsistent picklists and free text defeat counting and prediction alike.
- A model learns your definitions. If they changed, it learned two things.
- Permissions and consent belong in the data, not in the instructions to the AI.
- Fix what your first use needs, at the point of entry, then clean backwards.
ForAnyone planning AI, or the data work behind itTime to readAbout 25 minutesUse itBefore you fund an AI pilot
- Why pilots stall
- One record, in practice
- Fields AI can read
- Definitions and history
- Retrieval, privacy and consent
- What to fix, in what order
- The AI data-readiness grid
Rather talk it through?
A 30-minute conversation about your systems and what's getting in the way. You leave knowing what would help.