Saving lives with machine learning and artificial intelligence
When I planted a metaphorical stake in the ground at Best Friends’ 2016 national conference, declaring that our movement would lead the country to no-kill by 2025, I used John F. Kennedy’s 1962 “moonshot” speech as inspiration. No-kill 2025 would be our moonshot, and like JFK’s declaration that we would need new technologies, some of which had not yet been invented, I said that in order to achieve the bold goal of ending killing in shelters by 2025, our movement would have to work in new ways, develop new approaches to lifesaving and collaboration, and explore new technologies to get there.
I could never have imagined where the journey that started on that day would take us or how prescient Kennedy’s words from over 50 years earlier would turn out to be.
The first order of business on the road to 2025 was to quantify the problem we were setting out to solve. You can’t solve a problem if you can’t define it. So, I asked what I thought was a simple question: “How many brick-and-mortar shelters are there in this country?”
No one knew the answer to that question — and not just no one within Best Friends; no one in the entire country knew the answer to that question or the obvious follow-up: “How many dogs and cats are being killed in each of those shelters?”
No academic paper, no government record office, no shelter data bank. Nada! We were on our own starting from square one gathering the data needed to wrap our arms around this audacious goal, and we were already on the clock.
Looking back from today it sounds crazy that the best and most direct method that we had was to deploy our staff and volunteers to pick up the phone or email every county in every state in the country and ask for what details they had on the animal shelters in their jurisdiction. If that request was declined, we used the Freedom of Information Act to get the numbers we needed. It was time consuming and labor intensive, and in an analogy to the space flight timeline for JFK’s moonshot, we were somewhere between flying kites and hot air balloons. Relative to what we needed in terms of data collection to reach no-kill by 2025 we were just at counting fingers and toes!
Necessity being the mother of invention, we created an in-house data science team to facilitate shelter data access for everyone. Their first product was the pet lifesaving dashboard, a data mapping tool that garnered a Top 10 Innovator of the Year Award for Best Friends from Fast Company magazine. But that was only the beginning.
From fingers and toes to machine learning and AI
Data in the abstract is just numbers. We are immersed in data, and most of it we don’t think of as data. Our age, height, weight, blood pressure, and heart rate as well as the environment in which we live are all data points, but they don’t have much meaning unless they are compared with the same values from one, two, three years ago, and then they become part of a personal health profile. That’s useful, but not as useful as when that same data is compared to a larger sampling of individuals of the same age and environment. Then we can make more informed decisions about our health and lifestyle choices. Is my blood pressure way out of line for someone of my age? Should I change my diet, be more active, etc.? The more comparative data I have access to, the more accurate my health profile and the more relevant choices I can make.
Likewise, shelter data in the abstract can give a shelter director a snapshot in time of their lifesaving performance. That’s great information, but it doesn’t indicate trends, it doesn’t point to which programs are working to save lives and which are lagging, and it doesn’t point to what kind of programs might be needed tomorrow.
The holy grail of shelter data is to be able to provide actionable information and programing insights not simply based on past performance, but to somehow look around a corner into the future to predict what to expect and how best to deploy resources to save the lives of more pets in shelters.
From knowing nothing about the state of animal shelters in 2016 to knowing the future sounds like a crazy and boastful claim, but amazingly, we are on the cusp of just that.
Best Friends Animal Society has developed the most sophisticated predictive data estimation model in the country to determine the best way to reduce killing of dogs and cats in U.S. shelters. This predictive model is a critical piece of the Shelter Pet Data Alliance (SPDA), an interactive data platform for shelters and rescue groups designed to automate data reporting, provide timely data analysis, and soon connect them with support to help increase lifesaving. It contains the most comprehensive dataset from more than 7,760 organizations across the United States and is an incredibly valuable tool for shelters and rescue groups.
The model uses previously collected data (from more than 93% of all known brick-and-mortar shelters in the country) overlaid with the social vulnerability index, post-pandemic impact information, and county populations among other factors to predict shelter outcomes.
The model is dynamic — it is continuously updated with current and historical data as it is collected. Using machine learning, the model also provides real-time projections about shelters for which data isn’t available, allowing for more precise allocation of resources.
Best Friends’ 2023 data release is the most comprehensive and sophisticated dataset in animal welfare and will come in phases over the next few weeks. The sophistication lies in how the data is collected and analyzed, which will point to clear and direct actionable steps for city leaders and animal welfare professionals to help take their shelters to no-kill.
The broad-brush headline for 2023 is that despite very real challenges facing most shelters in a post-pandemic world with an unsettled economy, 67% of shelters across the U.S. maintained their save rate in 2023 while 15% made lifesaving gains and 18% regressed. The number of pets killed in shelters was well below pre-pandemic levels, and the national save rate stands at 83.4%.
With SPDA, shelters and rescue groups can easily navigate their intake and outcome data via a dashboard, see what’s happening in organizations in the local communities and beyond, create comparative reports, see local and national trends, and measure their positive impact over time. It enables a comprehensive shelter health profile as a basis to make the most relevant programing decisions. And soon, they’ll also be able to utilize actionable resources based on their 2023 data and our advanced data predictions.
Best Friends will continuously refine the model to include all known predictors of animal intake and outcomes to improve our ability to direct resources to the shelters and communities that need them most.
This work and the technology that our data science team developed is not only unique in animal welfare, it is unique technology, period. So much so that there is a patent pending on SPDA to ensure that this tech will not be pirated for profit to the exclusion of the free and beneficial use by animal welfare.
There is much more to come with in-depth analysis of 2023’s data followed in March with a rollout of state and local data and no-kill state and county additions, and in April we will release species-specific information (save rates, adoptions, etc.), percentage of shelters that are no-kill, and other detailed analysis.
Wow!
Together, we will Save Them All.
-Julie
