Shelter AI Assessment

Smiley Animal Welfare Center

Fictional example. Smiley Animal Welfare Center is fictional. This page shows a saved Shelter AI Assessment result generated through the same workflow used by the Workbench. It is read-only. Viewing it does not call a model.

What was submitted

Show

The fictional inputs used to generate this saved result.

Organization name

Smiley Animal Welfare Center

How animals enter the shelter

We take in about 2,500 animals a year, roughly 1,150 dogs and 1,350 cats. Around 40% come from animal control under our municipal sheltering contract, about 30% are owner surrenders booked by appointment, about 20% are transfers in from rural partner shelters that are over capacity, and the rest are litters born in our care. Stray hold is five days without identification and seven with. Owner surrender appointments are booked by phone and email, and the intake coordinator confirms them by hand; if somebody arrives without an appointment we take the animal anyway and sort the paperwork afterwards. Cat intake roughly doubles between May and September. We know it is coming every year and we still end up rearranging the building in June. Intake paperwork is completed on paper at the desk and typed into the shelter system afterwards, usually the same day. Two people run that desk on different shifts and they do not always record surrender reasons the same way, which we have known about for a while.

How animal status is tracked

The shelter management system holds the animal record — location, medical history, vaccination, spay/neuter status, holds, and outcome. That part works. Behavior observations do not fit. The free-text field in the system is too small for a real observation, so the behavior team keeps a shared spreadsheet instead, one row per observation. Kennel staff write observations on the paper kennel card during rounds twice a day and someone types the important ones in later, when there is time. Medical holds and behavior holds are set in the system and they do control whether an animal shows as available, so that much is reliable. What is not reliable is the picture. To know where an animal actually stands today you look at the system record, then the behavior spreadsheet, then the kennel card, then you ask whoever worked the last shift. New staff and weekend relief staff get this wrong, and it is the single most common reason an adopter is told something that is out of date.

How adoption inquiries are handled

Inquiries arrive four ways: the website application form, which emails a PDF to the adoptions mailbox; direct email to that same mailbox; phone; and people walking onto the adoption floor. We get somewhere between 180 and 220 a week across all four, more after a social media post does well. The mailbox has no owner. Whoever opens it first answers it, which means some get answered twice and some sit. A good number are the same handful of questions — is this animal still available, what is the fee, do you adopt out of state, what is your policy on dogs with young children, can I meet an animal today. A real inquiry becomes an application, then a counselling conversation, then a meet-and-greet, then approval and go-home. The counsellor reads the animal's notes, checks holds and restricted placement conditions, and talks the household through it. That conversation is the job and it should stay that way. The retyping is the part that grates. Applicant details come in on a PDF and get typed into the shelter system by hand at approval. We are slower to reply than we want to be, and adopters do go elsewhere while they wait.

How volunteers are trained

We have about 200 active volunteers, counting anyone who has taken a shift in the last 90 days. New volunteers attend a two-hour orientation, then do role training with the coordinator or an experienced volunteer, then shadow two shifts before working alone. The materials are a mess. There is a volunteer handbook as a PDF, role sheets in a SharePoint folder, some older documents on the shared drive that people still send because they have the link, and printed sheets in the volunteer room that nobody is certain match the PDF. When a procedure changes, we update one of those and usually not the others. Scheduling runs in a separate volunteer tool with its own login. It works, but the roles and shift types in it were never standardized, so pulling hours by role at month end takes the coordinator a couple of hours of cleanup. Ongoing guidance is mostly a group email and word of mouth on shift.

How fosters are supported

We have around 120 active foster homes. They are the reason we get through kitten season and we lose some of them every year because we are slow to answer. Fosters reach us by text to the coordinator's work phone, by email to the foster mailbox, and in a group chat that grew organically and that we do not officially support. Questions are mostly the same set: feeding amounts, weaning, whether a symptom needs a vet today or can wait, when the next vaccine is due, how to get more supplies, how to book a spay appointment, and what happens when their foster animal is adopted. Capacity lives in a spreadsheet the coordinator keeps. It reflects what she remembers and is updated after the fact, so on a Monday in June it is genuinely out of date. The medical questions are the ones that worry me. A foster asking whether a kitten that has stopped eating needs to be seen is asking a clinical question, and that has to reach a technician or a vet, quickly. It cannot be answered by anything else.

Repeated staff questions

The ones I hear most: is this animal available, and if not, why not. What are the current restricted placement conditions on this dog. What is our policy on adopting to households with children under 12. What is the fee this month, and does the senior discount still apply. Where is the current version of the intake procedure. Who approves a transfer out. What do I tell a foster whose animal has diarrhea at 9pm on a Sunday. Which vaccine is due and when. How do I book a community clinic appointment for a member of the public. What counts as an animal served for the county report and does it match what the family foundation means by it. Most of these have a written answer somewhere. The problem is that the answer is in a binder, a PDF, an email from the Director eighteen months ago, or in somebody's head, and the four versions do not always agree.

Documents used today

Standard operating procedures for intake, stray hold, cleaning and disease control, and euthanasia. Adoption application and adoption contract. Foster agreement, foster manual, and a set of one-page foster care sheets by age and species. Volunteer handbook and role sheets. Kennel cards. Behavior observation forms and the behavior review notes. Medical protocols and vaccination schedules. Transfer agreements with partner shelters. Community clinic consent forms. Pet food pantry intake sheet. Grant reporting templates, one per funder. Board reporting pack template. Incident report form. They live in SharePoint, on an older shared drive, in printed binders at the intake desk and in the volunteer room, and as PDFs attached to old emails. There is no index. Version control is the file name and the file name is not always right.

Software used today

A commercial shelter management system, hosted, which is our system of record for animals. Microsoft 365 for email, Teams, documents and spreadsheets, with SharePoint and an older shared drive that we never finished migrating. Four shared Outlook mailboxes for adoptions, foster, volunteers and general inquiries. A separate volunteer scheduling tool with its own login. A donor and fundraising platform. An accounting package. A website CMS with an online adoption application form that emails a PDF. A clinic scheduling workbook in Excel. A lot of other spreadsheets — foster capacity, transport runs, behavior observations, pantry visits, and a grant reporting workbook. None of these talk to each other. The only integration we have is a monthly export from the donor platform into accounting, and a person does that. We are about 40 staff. IT is a half-time operations coordinator and an outside IT company for laptops and email. We have no developers, no test environment, and no data or analytics staff. Wi-Fi in the kennels is unreliable, which is why the kennel card is still paper.

Manual or repetitive tasks

Typing intake paperwork from the paper form into the shelter system. Typing adopter details from the application PDF into the shelter system at approval. Typing kennel card observations in after rounds. Answering the same adoption and foster questions by hand, several times a day. Confirming surrender appointments one at a time. Rebuilding the foster capacity spreadsheet. Cleaning up volunteer hours by role every month because the roles were never standardized. Reconciling donation records against adoption fee records when development needs a household view. Reporting is the worst of it. We produce a board pack monthly, a contract report to the county quarterly, and eleven grant reports a year across six funders. Every one is assembled by hand from the shelter system's reporting module into a spreadsheet and then into a document, and every funder counts things slightly differently, so the numbers have to be rebuilt rather than reused. Two of them define an animal served differently and we have never settled which one we agree with.

Sensitive data handled in workflows

Describing categories only, not entering any of it here. Adopter and applicant contact details and the contents of adoption applications, including household composition and references. Foster carer contact details and home information. Volunteer records, including any incident reports. Donor identity and giving history. Owner details on surrender and stray records, including surrender reasons that are frequently personal — illness, eviction, bereavement, domestic circumstances. Animal medical records. Bite and behavior incident records, which can end up in an insurance or legal context. Staff notes and internal decisions, including euthanasia records. Some records relate to minors, and some surrender conversations involve people in crisis. Anything that leaves our control needs to be a deliberate decision, not a default.

Decisions that must stay human-owned

Euthanasia. That decision is the veterinarian's, with the Executive Director where the grounds are not purely medical, and it is not something any system should have an opinion about — not a score, not a ranking, not a suggested list. Medical diagnosis and treatment, including whether a foster animal needs to be seen today. That is a clinician's call. Behavior risk. Bite history, restricted placement conditions, and whether a dog is safe in a particular household are decided at behavior review by people who have watched the animal. Cruelty and neglect. We house animals and we report what we see; the investigation and any determination belong to the municipal officers, not to us and not to a system. Final adoption approval, and the match between animal and household. A counsellor makes that call in conversation. We are not scoring applicants and we are not letting anything decline an application automatically. Foster placement of medically fragile or behaviorally complex animals. Volunteer eligibility and anything following an incident. Anything with a safety consequence for a person or an animal. And anything we say to a funder, the county, the press, or the board goes out over a named person's signature. We are open to help with the paperwork around all of these. The decisions themselves stay with people.

Executive Summary

This assessment outlines potential AI opportunities for Smiley Animal Welfare Center, focusing on streamlining repetitive tasks, improving data consistency, and enhancing communication, while strictly adhering to a human-centered approach for critical decisions. The primary goal is to free up staff and volunteer time, improve data accuracy, and enhance the experience for adopters and fosters, without automating sensitive judgments. A phased roadmap is proposed, starting with a pilot focused on standardizing information access and automating routine inquiries.

Draft Workflow Map

Animal Intake

Animals enter the shelter from various sources: animal control, owner surrenders, transfers from partner shelters, and litters born in care. Owner surrenders are booked by phone/email, confirmed manually. Walk-ins are accepted without appointments. Paperwork is completed manually and then typed into the shelter system.

Actors

  • Intake Coordinator
  • Intake Desk Staff
  • Animal Control Officers
  • Owner Surrenderers
  • Partner Shelter Staff

Pain points

  • Manual confirmation of owner surrender appointments is time-consuming.
  • Inconsistent recording of surrender reasons by different staff.
  • Manual data entry from paper forms into the shelter system is repetitive and prone to error.
  • Seasonal surge in cat intake (May-September) causes operational strain.

Animal Status Tracking

The shelter management system (SMS) tracks core animal data. Behavior observations are kept in a separate shared spreadsheet due to SMS field limitations. Kennel staff record observations on paper cards, which are later typed into the behavior spreadsheet. Medical and behavior holds are set in the SMS and reliably control availability.

Actors

  • Shelter Management System
  • Behavior Team
  • Kennel Staff
  • Medical Staff

Pain points

  • Behavior observations are fragmented across the SMS (limited field), a shared spreadsheet, and paper kennel cards.
  • Manual transcription of kennel card observations is time-consuming and often delayed.
  • Inconsistent and incomplete picture of an animal's current status, requiring cross-referencing multiple sources and asking staff.
  • New and relief staff frequently provide outdated information to adopters.

Adoption Inquiries & Process

Inquiries arrive via website form (PDF email), direct email, phone, and walk-ins. The adoptions mailbox lacks an owner, leading to duplicate responses or delays. Many inquiries are repetitive questions. Approved applications lead to counselling, meet-and-greet, and go-home. Applicant details from PDF forms are manually typed into the SMS.

Actors

  • Adoptions Staff
  • Adoption Counsellors
  • Applicants
  • Website CMS
  • Shared Outlook Mailbox

Pain points

  • High volume of repetitive inquiries consumes staff time.
  • Lack of clear ownership for the adoptions mailbox leads to inefficiencies (duplicates, delays).
  • Manual data entry of applicant details from PDF to SMS is slow and error-prone.
  • Slow response times lead to adopters going elsewhere.

Volunteer Training & Management

New volunteers undergo orientation, role training, and shadowing. Training materials are fragmented across PDF, SharePoint, shared drive, and printed sheets, with poor version control. Scheduling is done in a separate volunteer tool with inconsistent role definitions. Ongoing guidance is via email and word of mouth.

Actors

  • Volunteer Coordinator
  • Volunteers
  • Experienced Volunteers
  • SharePoint
  • Shared Drive
  • Volunteer Scheduling Tool

Pain points

  • Disorganized and inconsistent training materials across multiple platforms.
  • Lack of version control for training documents.
  • Inconsistent role and shift definitions in the scheduling tool make reporting difficult.
  • Manual cleanup of volunteer hours for reporting is time-consuming.

Foster Support & Management

Fosters communicate via text, email, and an unofficial group chat. They frequently ask repetitive questions about care, medical symptoms, supplies, and appointments. Foster capacity is tracked in a manually updated spreadsheet, which is often out of date. Medical questions require quick escalation to clinical staff.

Actors

  • Foster Coordinator
  • Foster Carers
  • Medical Staff
  • Shared Outlook Mailbox
  • Text Messages
  • Unofficial Group Chat
  • Spreadsheet

Pain points

  • High volume of repetitive foster questions consumes coordinator's time.
  • Critical medical questions need immediate and reliable escalation to clinical staff.
  • Fragmented communication channels for foster support.
  • Inaccurate and outdated foster capacity tracking leads to missed opportunities or over-commitment.
  • Losing foster homes due to slow response times.

Information Access & Reporting

Staff frequently ask repetitive questions about animal availability, policies, fees, and procedures. Answers are scattered across binders, PDFs, emails, and staff knowledge, often with conflicting versions. Reporting for the board, county, and funders is manual, time-consuming, and requires rebuilding data due to differing definitions of 'animal served'.

Actors

  • All Staff
  • Executive Director
  • Development Staff
  • Operations Coordinator
  • Shelter Management System
  • Spreadsheets
  • Printed Binders
  • SharePoint
  • Shared Drive

Pain points

  • Difficulty for staff to quickly find accurate and up-to-date information.
  • Inconsistent policies and procedures due to poor version control and scattered documentation.
  • Manual assembly of all reports (board, county, grant) is highly labor-intensive.
  • Inconsistent definitions of key metrics (e.g., 'animal served') require manual data manipulation for each report.
  • Lack of integration between software systems necessitates manual data export and reconciliation.

Top AI Opportunities

AI-Powered Knowledge Base & Chatbot for Staff and Public

Readiness: Medium. Requires consolidating and standardizing existing documentation first. Technology for a basic chatbot is readily available.

Develop an internal and external knowledge base, populated with standardized, version-controlled policies and FAQs. An AI chatbot could then answer common questions from staff, adopters, and fosters, directing complex or sensitive inquiries to human staff.

Potential impact

Significantly reduce time spent by staff answering repetitive questions (e.g., 'is this animal available?', 'what's the fee?', 'foster feeding amounts'). Improve consistency of information provided to all stakeholders. Enhance adopter and foster experience through faster responses. Free up staff for higher-value tasks.

Risks

  • Ensuring accuracy and up-to-date information in the knowledge base.
  • Risk of providing incorrect or misleading information if not properly trained and monitored.
  • Maintaining a human touch for sensitive interactions.
  • Initial effort to centralize and standardize disparate documents.

Intelligent Document Processing for Intake & Adoptions

Readiness: Medium. Requires integration with the existing shelter management system and careful configuration for form layouts.

Implement AI/OCR (Optical Character Recognition) to extract data from paper intake forms and adoption application PDFs, automatically populating relevant fields in the shelter management system. This would reduce manual data entry.

Potential impact

Eliminate significant manual data entry for intake and adoption processes. Improve data accuracy and reduce transcription errors. Speed up the adoption process, potentially reducing adopter attrition.

Risks

  • Accuracy of OCR for handwritten or varied forms.
  • Integration challenges with the existing shelter management system.
  • Need for human review and validation of extracted data.
  • Data privacy and security considerations for sensitive applicant information.

Automated Communication Triage & Scheduling Support

Readiness: Medium. Requires natural language processing capabilities and integration with email/text platforms. Initial training on common inquiry types is needed.

Utilize AI to analyze incoming emails (adoptions, foster mailboxes) and texts, categorize them, and suggest responses for common queries or flag urgent medical questions for immediate human review. Could also assist with scheduling owner surrender appointments by checking availability and drafting confirmations.

Potential impact

Reduce response times for adopters and fosters. Ensure critical medical questions are escalated promptly. Improve efficiency of shared mailboxes by reducing duplicate efforts and ensuring all inquiries are addressed. Streamline appointment booking.

Risks

  • Misinterpretation of urgent or nuanced communications.
  • Maintaining a personalized and empathetic tone in automated responses.
  • Integration with various communication channels (email, text, unofficial chat).
  • Ensuring human oversight for all suggested actions.

Enhanced Data Aggregation & Reporting Assistant

Readiness: High. Primarily involves data integration and scripting, rather than advanced AI. Requires clear definition of reporting metrics.

Develop a tool that can pull data from the shelter management system and other spreadsheets, apply standardized definitions (e.g., 'animal served'), and generate draft reports for various stakeholders (board, county, funders). This would not automate the final report, but automate the data assembly.

Potential impact

Drastically reduce the manual effort and time spent on report generation. Improve consistency and accuracy of reported metrics. Allow staff to focus on analysis and narrative rather than data compilation.

Risks

  • Ensuring all data sources are accessible and compatible.
  • Defining and standardizing reporting metrics across all stakeholders.
  • Risk of misinterpreting data if definitions are not precisely coded.
  • Requires ongoing maintenance as reporting requirements evolve.

First Pilot Recommendation

Standardized Knowledge Base & Internal Staff Chatbot

Rationale

The most pervasive pain point is the difficulty for staff to find accurate, up-to-date information, leading to repeated questions and inconsistent answers. Addressing this first will immediately improve internal efficiency and lay the groundwork for external-facing AI. It's a contained scope with high potential impact and lower risk than automating external interactions.

Scope

Phase 1: Consolidate and standardize all existing policies, procedures, and FAQs (e.g., animal availability, restricted placement conditions, adoption policies, fees, intake procedures, foster care sheets, vaccine schedules) into a single, version-controlled knowledge base (e.g., within SharePoint). Phase 2: Implement a simple internal chatbot (e.g., within Microsoft Teams) that can query this knowledge base to answer staff questions. The chatbot would clearly state its source and direct to a human for complex queries.

Success metrics

  • Reduction in the number of repetitive questions asked to supervisors/colleagues (measured via staff survey or informal tracking).
  • Increased staff confidence in finding accurate information (measured via staff survey).
  • Improved consistency of information provided by staff to the public (measured by spot checks or feedback).
  • Completion of a centralized, version-controlled knowledge base for key operational areas.

30 / 90 / 180 Day Roadmap

First 30 days

  • Form a small, cross-functional working group (Operations Coordinator, key staff from Adoptions, Foster, Intake, Volunteer) to champion AI initiatives.
  • Identify and prioritize the top 5-10 most frequently asked staff questions and their current answer sources.
  • Begin consolidating and standardizing existing documentation for the pilot knowledge base, focusing on the most critical and frequently referenced policies/procedures (e.g., adoption policies, fee schedule, animal availability rules).
  • Research simple, low-cost internal chatbot options compatible with Microsoft 365 (e.g., Teams integration).

First 90 days

  • Launch the internal, centralized knowledge base (e.g., in SharePoint) with initial standardized content.
  • Pilot a basic internal chatbot within Microsoft Teams, trained on the initial knowledge base content, for staff to ask questions.
  • Gather feedback from staff on the usability and accuracy of the knowledge base and chatbot.
  • Begin standardizing volunteer role definitions and shift types in the volunteer scheduling tool.
  • Assess the feasibility and requirements for an OCR solution for intake paperwork (e.g., identifying specific forms, data fields).

First 180 days

  • Expand the knowledge base content to include more operational areas (e.g., foster care sheets, volunteer handbook sections).
  • Refine the internal chatbot based on feedback, potentially adding more sophisticated query handling or escalation paths.
  • Explore options for integrating the volunteer scheduling tool with a centralized system for better reporting, or migrating to a more flexible platform.
  • Develop a plan for implementing Intelligent Document Processing for intake paperwork, starting with a single, well-defined form.
  • Begin defining and standardizing key reporting metrics (e.g., 'animal served') across different funders and the county, in preparation for an enhanced reporting assistant.

Assumptions

  • The existing commercial shelter management system has an API or data export capabilities that can be leveraged for integration, even if limited.
  • Staff are open to adopting new tools and workflows, provided they see clear benefits and are adequately trained.
  • The organization has the internal capacity (even if limited to the half-time operations coordinator and external IT) to support the implementation and ongoing maintenance of new, relatively simple AI tools.
  • The 'shared drive' is accessible and can be migrated or integrated with SharePoint for better document management.
  • The organization is willing to invest in standardizing its internal processes and documentation as a prerequisite for effective AI implementation.
  • The 'unofficial group chat' for fosters can be transitioned to a more controlled and supported communication channel.
  • The organization has a clear understanding and agreement on the definitions of key metrics (e.g., 'animal served') for various reports, or is willing to establish them.

Questions to Validate with Shelter Staff

  • What are the specific pain points and time sinks for intake staff during peak cat season?
  • Can you walk through the exact steps and systems used when a new behavior observation is made, from kennel card to spreadsheet?
  • What are the most common 'out of date' pieces of information adopters receive, and what is the typical impact?
  • What is the current process for updating training materials, and what are the biggest challenges in keeping them consistent?
  • What is the current process for escalating urgent foster medical questions, and what are the biggest risks or delays?
  • Are there any existing efforts or preferences for standardizing reporting metrics across different funders and the county?
  • What are the current IT security protocols and data privacy concerns that need to be considered for any new software or data integration?
  • What is the current budget and resource availability for exploring and implementing new technology solutions?

Do Not Automate

  • Euthanasia decisions (veterinarian and Executive Director only).
  • Medical diagnosis and treatment plans, including determining if a foster animal needs immediate veterinary attention (clinician's call).
  • Behavior risk assessments, bite history evaluation, and restricted placement conditions (behavior team's judgment).
  • Cruelty and neglect investigations and determinations (municipal officers).
  • Final adoption approval and the match between an animal and a household (counsellor's conversation and judgment).
  • Declining adoption applications automatically.
  • Foster placement of medically fragile or behaviorally complex animals.
  • Volunteer eligibility decisions and actions following an incident.
  • Any decision with a safety consequence for a person or an animal.
  • Official communications to funders, the county, the press, or the board (requires named person's signature).

Guardrails

  • All AI-generated responses for public-facing interactions (adopters, fosters) must be clearly identified as AI-generated and offer an easy path to connect with a human.
  • Any AI system providing information must be regularly audited for accuracy and consistency with current policies.
  • Human oversight and approval are required for any AI-suggested action that impacts an animal's welfare, a person's safety, or a critical operational decision.
  • Data privacy and security protocols must be rigorously applied to all data handled by AI systems, especially sensitive categories like adopter/foster details and animal medical records.
  • AI tools should be designed to augment human capabilities, not replace critical human judgment or empathy.
  • Prioritize solutions that are simple, cost-effective, and maintainable with existing internal resources or readily available external support.
  • Ensure clear version control and a single source of truth for all policies and procedures that feed into AI-powered knowledge bases.
  • Implement a feedback loop for staff to report issues or suggest improvements for AI tools and their outputs.

How this result was produced

Safe run metadata from the saved generation. Digests, prompts, and provider payloads are not shown.

Provider
Gemini
Model
gemini-2.5-flash
Workflow
shelter-workflow-v1
Prompt
shelter-assessment-v3
Schema
shelter-assessment-v1
Retrieval
None (form-only workflow)
Total time
20.8 s
Generation
20.6 s
Input tokens
2639
Output tokens
3757
Total tokens
6396

Checks

  • Guardrails presentPassed

    count=8

  • Do-not-automate presentPassed

    count=10

Want to try the Workbench?

Access is invite-only while the product is in alpha. Request access if you want to run these workflows on your own material.