GOODGRAPH FOR SOCIAL IMPACT

Know what your AI is doing, and what your organization is learning.

For mission-driven organizations, the question is whether AI is improving program delivery, reporting, grantmaking, and operations without losing the local context and staff judgment that shape the result.

ContextPolicyHumanjudgmentCorrectionOutcomeLearningAI workflow

BUILT FROM THE FIELD

Nearly two years working alongside grassroots and mission-driven organizations.

GoodGraph has been shaped through discovery, pilots, and product development alongside food-security and community-based organizations.

We saw two things get lost repeatedly:

Meaning: the same measure or term could mean different things across reports and organizations.

Human judgment: staff added exceptions, interpretations, and local context but that knowledge could remain trapped in a conversation or workflow.

Preserving both means important knowledge doesn't disappear when staff leave or AI models change.

MEANING

Same words. Different meaning.

Reports often contain measures that look comparable but aren't.

TWO REPORTS · ONE MEASUREExample

ORGANIZATION A

People served

3,800

Definition
Unique individuals receiving assistance
Source
Annual program report

ORGANIZATION B

People served

5,200

Definition
Service visits during the reporting period
Source
Quarterly program report

NOT DIRECTLY COMPARABLE

3,800 unique individuals 5,200 service visits

GoodGraph preserves meaning and provenance so AI doesn't silently turn different evidence into the same thing.

PRESERVING HUMAN JUDGMENT

What happens when a human corrects the AI?

SCENARIOA staff member checks program eligibility

Example workflow

STAFF MEMBER ASKS

“Can a family outside our county receive emergency assistance?”

AI ANSWERS

“No. Services are limited to county residents.”

HUMAN CORRECTS

“Emergency assistance is an exception.”

QUESTION

What happens to that correction?

TODAY, WITHOUT GOODGRAPH

The correction may stay inside one conversation or workflow.

The next person, or AI system, may get the outdated answer again.

WITH GOODGRAPH

  1. CONTEXT USED

    Eligibility Policy v2

  2. HUMAN CORRECTION CAPTURED

    “Emergency assistance is an exception.”

  3. CURRENT SOURCE IDENTIFIED

    Emergency Assistance Policy

  4. GOODGRAPH

    Flags a potential context change

    Needs review

  5. HUMAN / PROGRAM OWNER

    Reviews / approves

    Not every correction becomes policy.

  6. APPROVED ORGANIZATIONAL CONTEXT

    Eligibility Policy v3: emergency exception recorded

  7. APPROVED AI WORKFLOWS

    Use the current exception

When human judgment is reviewed, approved, and retained, a correction can become organizational learning.

ONE STEP FURTHER

Permissioned Governance and Sharing.

Once an organization retains its context and learning, it can decide what is approved, who owns it, and what can be reused or shared.

Organizations keep their underlying context under their own control, while choosing what approved evidence to share with funders or partners.

hand over its context or use identical measures.

  1. ORGANIZATION

    Keeps its context and learning

  2. APPROVED EVIDENCE

    Shared selectively

  3. FUNDER / INTERMEDIARY

    Learns across a portfolio

HOW TO WORK WITH US

Start with one AI workflow.

Choose one recurring AI-supported workflow where staff still review, correct, or interpret the output.

We'll look at the context the AI depends on, where human judgment enters, and what evidence should be captured or retained to understand whether the work is improving.