AI Matching

AI matching helps your team find the best fit between animals and the people who want to adopt or foster them. Instead of manually combing through applicants, you get a ranked list of compatible candidates for any pet; or a ranked list of compatible pets for any applicant. Each detailed result includes a calibrated score, factor breakdown, confidence and readiness context, supporting evidence, and a quick read on what works and what to review.

What makes it different from a keyword search is that it genuinely understands context. "I work from home and have a big fenced yard" registers as a strong fit for a high-energy dog that needs space, not because the words overlap, but because the AI understands what they mean together. It factors in household setup, lifestyle, experience level, what the applicant is looking for, and what the animal needs to thrive.

The AI is a decision-support tool. Your team always makes the final call.

What staff see

When you open an eligible pet's profile, an AI Matches section shows adopters and fosters ranked by compatibility. Adopter and foster profiles show matching animals in an AI Matches panel. Every match includes:

• A calibrated compatibility score from 0–100%
• Strengths: short labels explaining why the match works (e.g. "Fenced yard", "Works from home", "Dog breed experience", "No children")
• Concerns: short labels for things worth reviewing (e.g. "Apartment living", "First-time pet owner", "Limited space")
• Match factors for preferences, household, experience, schedule, and care needs when enough evidence exists
• Confidence and profile gaps, including the best next profile detail to collect when information is missing
• A separate readiness status such as Ready now, Needs information, Team review, or Blocked
• A Show details button that opens the available written explanation or generates and caches one when needed
• Compare controls for reviewing two or three candidates side by side
• Feedback and progress controls that help the organization learn from staff review and placement outcomes

Matches are always presented as suggestions. Nothing is sent, recommended, or acted on automatically.

How it works, end to end

1. Enable AI matching (admin only) at Settings → AI Features.

2. Choose which form fields to include. No applicant form fields are shared with the AI unless your team explicitly enables them. See Controlling what the AI sees.

3. Let PawPlacer build profiles. After matching is enabled, PawPlacer builds the initial matching index and queues refreshes when relevant pet, applicant, or enabled form-field data changes. Admins can clear generated matching data from Settings → AI Features if they need to rebuild from a clean state.

4. Quick estimates appear automatically. Once profiles exist, eligible pet and applicant profiles show AI Matches with quick compatibility estimates. These are lightweight comparisons based on overall profile similarity; good for a fast scan.

5. Click "Find Matches" for the full picture. The Find Matches button triggers the AI to individually evaluate the top candidates and return a scored, ranked list with strengths, concerns, and explanations. Results are cached; viewing them again is instant and doesn't re-run the AI.

6. Review factors and compare candidates. Use the factor bars, confidence, readiness, and missing-evidence guidance to understand why candidates differ. Select Compare on two or three matches to review their tradeoffs side by side.

7. Expand a match for a written explanation. Click Show details to open the explanation. If an explanation is not already available, PawPlacer generates and caches one on demand.

8. Teach the ranking system. Mark a result as helpful or not quite right. Negative feedback can include a reason such as household, schedule, experience, care, preferences, or missing information. Use Progress to record events such as shortlisting, meet-and-greets, trial placements, completed fosters, declines, or withdrawals.

9. Keep results fresh. PawPlacer queues profile refreshes when relevant records change. If the source data is newer than the current detailed results, the Matches area marks them as stale so an authorized team member can run Find Matches again.

Where matches appear

On a pet's profile

The AI Matches section appears on eligible in-care pet profiles when matching is enabled. It has two views:

• Adopters: ranked adoption applicants compatible with this pet
• Fosters: ranked foster applicants compatible with this pet

The Fosters tab is only visible to users with foster coordinator access or higher.

On an adopter's profile

The AI Matches panel shows ranked animals for that applicant. It is visible to users with adopter matching access.

On a foster's profile

The AI Matches panel shows ranked animals for that foster applicant. It is visible to users with foster matching access.

Who and what matching applies to

Animals

Matching only includes pets that are active and available. A pet is automatically excluded if it has been deleted, is no longer available for adoption, or its status is an outcome (adopted, transferred out, deceased, lost, etc.). These exclusions take effect instantly when a status changes.

People

Matching includes adopters and fosters with active or pending status. Inactive, suspended, or blocked applicants are automatically excluded; volunteers are not included in matching.

What the scores mean

The final score blends the available factor scores with semantic and keyword similarity. Organization learning may make a small adjustment as your team records more feedback and outcomes. A limited-confidence estimate is labeled clearly when detailed AI scoring is unavailable or cannot be verified against the structured profile facts.

Scores reflect the data that exists. A well-filled-out profile produces a more confident score than a sparse one. Missing factors are shown as missing information instead of being silently treated as an average score.

Score and readiness are separate

Compatibility answers "could this be a good long-term fit?" Readiness answers "what must staff confirm or do next?" A candidate can have a strong compatibility score while still needing information or team review.

• Ready now: the available evidence supports moving to staff review and normal next steps
• Needs information: one or more important profile details should be collected
• Team review: a known concern deserves an explicit staff conversation
• Blocked: a documented blocker must be resolved before the placement should proceed

The card also shows a confidence percentage and the number of profile gaps. These indicators describe evidence quality; they are not a replacement for staff judgment.

Organization learning

Matching learns at the organization level from three kinds of evidence:

• Staff ratings on individual matches, including optional reason categories
• Placement progress such as shortlisting, meet-and-greets, trials, completed fosters, declines, and withdrawals
• Confirmed placement outcomes and follow-up signals

Learning starts conservatively and uses sample safeguards so a small number of clicks cannot dramatically reshape rankings. Admins can review signal totals, average confidence, confirmed placements, top-three success, and learned factor priorities under Settings → AI Features. Reset learning removes feedback and learned weighting while preserving generated profiles and current match results.

Automatic safety checks

Before scoring anything, the system runs hard checks on each candidate pair. These can disqualify a candidate entirely; no score, not shown in results:

• Species mismatch: An applicant who indicated they only want dogs won't see cats, and vice versa
• Child safety: A pet flagged as not good with children won't appear for households with children
• Household animal conflict: A pet with known aggression toward cats or dogs won't appear for homes that have them
• Foster at capacity: A foster with no remaining capacity won't appear in results

These checks run on the pet's core profile data and on what the AI can determine from the applicant's form answers. If your forms don't ask about household children, the children safety check won't fire; the AI only evaluates what it has.

Controlling what the AI sees

There are two separate layers of control: what your team explicitly chooses to share, and what is automatically protected regardless.

What you choose to share

No custom form fields are sent to the AI by default. Your team opts each field in individually.

To include a field: open any form in Settings → Forms, click the gear icon on a field, and turn on Include in AI matching. If that toggle is off, that field is never used for matching; even if matching is enabled globally.

The AI settings page also has a bulk-enable option: click Preview compatible fields to review eligible fields across your forms, deselect anything you do not want included, and enable the selection in one step. PII-like fields are labeled and cannot be selected.

After updating fields or records, PawPlacer queues matching profile refreshes automatically. Run Find Matches again when the UI marks detailed results as stale.

What is always included

Regardless of field settings, the following pet attributes are always part of matching:

• Species, breed, size, age, sex, spayed/neutered status
• Temperament
• Good with / bad with (children, cats, dogs)
• Health notes and special needs
• Medical complexity (active conditions or prescriptions)

For adopters and fosters, the system also looks at which pets they've linked, favorited, or previously adopted to infer species preference; no form field needed.

What is automatically protected

PII (personally identifiable information): names, email addresses, phone numbers, home addresses; never reaches the AI, regardless of your field settings. Three layers of protection run server-side:

1. Field type blocking: Email, phone, address, and file upload fields are excluded at the source and cannot be enabled for matching
2. Identity column blocking: Any field that syncs to a core identity column (name, email, phone, address) is blocked regardless of label or type
3. Content scanning: All free-text answers are scanned for PII patterns before being sent. If an applicant typed their phone number into an open-ended question, it's caught and replaced with [redacted]

The AI never knows who a specific applicant is; only compatibility-relevant information about their lifestyle, household, and preferences.

Availability by region

AI matching is not available for organizations in the EU/EEA, United Kingdom, or Switzerland due to data protection regulations (EU GDPR, UK GDPR, and Swiss FADP respectively) that place specific requirements on AI-based automated processing of personal data.

If your account's country is set to one of these regions, AI matching cannot be turned on and the settings page explains why. If you believe your region was set incorrectly, contact support.

Other AI features (AI descriptions and AI data mapping) are not affected by this restriction.

Compare Candidates

Open More > Match Lab to place up to two saved adopter or foster matches side by side. See Match Lab for its filters, comparison details, and read-only shortlist tools.

Privacy and data handling

• Processed by OpenAI: Matching uses OpenAI's API, so data is sent to OpenAI's servers as part of this
• Not used to train AI models by default: OpenAI states that API data is not used for training by default. Limited retention for abuse monitoring may apply unless a Zero Data Retention arrangement is in place
• You decide what's shared: No form field data is sent unless your team explicitly enables it field by field
• No automated decisions: Matching produces suggestions your team reviews. No emails are sent and nothing happens without a staff action
• Data can be cleared anytime: Admins can delete all matching profiles and results instantly from the AI settings page