Experienced adoption counselors get good at spotting which homes might suit which animals. That knowledge is valuable, but it can be hard to apply when you have dozens of open applications and many animals in care. PawPlacer's AI matching is an optional feature that helps your team find promising matches to review.
This post explains what staff see, what checks the system runs, how it handles private information, and how it's built. Matches are suggestions only. Nothing is sent or decided automatically, and your team makes the final call.
What staff see
When matching is turned on, an eligible pet's profile includes an AI Matches section with ranked adopters and fosters. Adopter and foster profiles show the same thing in reverse, with a ranked list of animals.
Each match includes:
- A compatibility score from 0 to 100%
- Strengths, such as "Fenced yard" or "Works from home"
- Concerns, such as "First-time pet owner" or "Limited space"
- Scores for factors like household, experience, schedule, and care needs, when there's enough information
- A confidence level and any details still missing
- A readiness status, such as Needs information or Team review
- A written explanation you can open, and a way to compare two or three candidates
Quick estimates appear once profiles have been built. Choosing Find Matches has the AI evaluate the top candidates individually and return more detailed results.
A compatibility score and a readiness status answer different questions. The score suggests whether this could be a good long-term fit, and readiness shows what staff should confirm next. A candidate can score well and still need more information.
Safety checks come first
Before anything is scored, the system runs fixed checks on each pairing. A candidate who fails one of these won't appear in the results:
- Child safety: a pet marked as not good with children won't appear for households with children.
- Other animals: a pet marked as not good with cats or dogs won't appear for homes that have them.
- Declined care: a pet with medical, special, or behavioral needs, or one that must go with a bonded partner, won't appear for an applicant who clearly said no to that kind of care.
These checks rely on the information you collect. If your form doesn't ask about children in the household, for example, that check has nothing to work with. It's a good idea to review your application questions before you turn matching on.
How private information is handled
Matching starts turned off, and an admin has to turn it on in Settings > AI Features. No applicant form field is shared with the AI unless your team explicitly enables it.
Names, email addresses, phone numbers, and street addresses aren't part of the information used for matching. The system filters direct identifiers and blocks or redacts fields and values that look like personal information before anything is sent. See Security and Data Ownership for more about how PawPlacer handles your data.
How it's built
This section is more technical, so feel free to skip it.
Profiles. PawPlacer builds a compatibility profile for each animal and applicant. Animal profiles come from structured details such as species, age, temperament, medical complexity, and enabled form fields. Applicant profiles summarize the answers your team has chosen to include. Profiles refresh when relevant data changes.
Embeddings. Each profile is converted into a 512-dimensional vector using OpenAI's text-embedding-3-large model and stored in Postgres with pgvector. Similar profiles end up close together even if they use different words. "I work from home and have a quiet house" can land near "needs a calm home with someone around," although the two phrases share few words. This provides the quick estimates.
Scoring. After the fixed safety checks, Find Matches asks the AI to score the top candidates on the available factors. The final score blends those factor scores with semantic and keyword similarity.
Missing information. Missing details are shown as gaps, rather than being treated as an average score. Confidence depends on the less complete of the two profiles, so a detailed pet record can't make a sparse application look well supported.
Why not only use rules? Rules work well for fixed requirements, so we use them for safety checks. They're less useful for judgment calls, such as whether someone's experience and daily routine suit a shy dog. Combining the two lets each approach do what it's good at.
Setting it up
- Go to Settings > AI Features and turn on Pet Matching. You'll need admin access.
- Choose which form fields to include.
- Let PawPlacer build the profiles. It keeps them up to date as your records change.
- Open a pet's profile, review the quick estimates, and select Find Matches.
Better records lead to better matches, so it's worth filling out pet profiles and asking clear questions on your forms. Full details are in the AI matching guide.
