Nonprofit AI Case Study in Food Rescue with Joe Robbins

Carolyn Woodard talks with Joe Robbins, Chief Product Officer at Food Rescue US, about a real-world AI project built from the ground up: a predictive model that flags volunteer pickups at high risk of a last-minute cancellation.

Food Rescue US connects volunteers with food donors and receiving agencies through an app similar to a delivery service, coordinating about 150,000 pickups a year across 27 states. When a volunteer cancels within 24 hours of a pickup, site coordinators scramble to cover it or risk the relationship with the donor. 

Joe and his team partnered with a supply chain researcher at Michigan State University to build an algorithm that predicts cancellation risk and flags pickups at high-risk for cancellation. Rather than automating any decisions outright, the AI triggers an alert to the site coordinator, who checks with the volunteer about confirming the pickup or cancelling early. During the pilot and rollout the AI tool  flagged thousands of rescues that would have been last minute cancellations. 

Joe shares the practical lessons from a project that took a full year to scope, pilot, and improve: why good data capture has to come before any AI project, how a tightly scoped pilot funded through a grant reduced risk, and why keeping a human in the loop kept the tool trustworthy for staff and volunteers alike. 

He also offers a framework for nonprofits without an in-house AI researcher or technical staff to get started using AI for more than productivity tools.

Joe and Carolyn discuss:

• How a predictive model helps Food Rescue US flag volunteer pickups at high risk of last-minute cancellation, giving site coordinators three to eight days of notice instead of a same-day scramble.
• Why the project deliberately keeps a human in the loop: the algorithm surfaces a risk score, but a site coordinator decides whether to act on it.
• Why clean, well-understood data capture has to happen before any AI project, and how to audit what data your organization already has.
• How building a pilot into a grant proposal made it easier to test small before rolling out more broadly.
• A framework for nonprofits without an in-house AI researcher: start with free assistant tools, learn what context AI needs to be useful, then look for a tightly scoped, well-documented problem to solve where you already have the data to use.
• What's next for Food Rescue US: using AI to take logistical pressure off site coordinators so they can focus on building community relationships.

Resources Mentioned:

• Food Rescue US: https://foodrescue.us
• Tech Soup AI Impact Hour interview with Joe: https://engage.techsoup.org/c/upcoming-events/ai-impact-hour-for-nonprofits

_______________________________
Start a conversation :)

• Register to attend a webinar in real time, and find all past transcripts at https://communityit.com/webinars/
• email Carolyn at cwoodard@communityit.com
• on LinkedIn 
• on reddit/r/nonprofitITmanagement
• on the Community IT website

Thanks for listening. 

Transcript

Carolyn Woodard 0:22

In the webinar that I went to, like just that story about being able to predict how like what donations were likely to get canceled. And just that I thought was such an amazing little case study of Yeah.

Joe Robbins 0:37

But we had an opportunity and we figured, you know, we're gonna learn so much more this way. Um, and it's gonna equip the team to really approach kind of bigger projects and more, more integral projects in the future. So, so it did that. It did that. It was tough sledding, but but yeah, we have a much better kind of grasp on on what it is and what it can do and and kind of what the limitations are. So, so yeah, learning experience that was incredibly valuable.

Carolyn Woodard 1:03

Welcome everyone to the Community IT Innovators Technology Topics podcast. I'm Carolyn Woodard, your host, and today I'm really excited to be talking to Joe Robbins from Food Rescue about this really interesting example of using AI that I heard about. So, Joe, would you like to introduce yourself and Food Rescue? What do you do?

Joe Robbins 1:26

Yeah, yeah. Thank you, Carolyn, for having me here. I love, love the podcast. I'm really excited. this is a resource that folks have. I could have used this a couple of times, I think. Um, but yeah,

Joe Robbins 1:35

I'm Joe Robbins. I'm the chief product officer for Food Rescue US. Um, we're a nonprofit that's been around since about 2011. And what we do is we have a volunteer network that recovers surplus food from food businesses and then delivers it to what we call receiving agencies, but it could be a food bank, it could be a pantry, um, somebody who's doing distributions. Um

Joe Robbins 1:58

The way we do that is is we link these volunteers together with our um with our native app. And the way that it works is it's kind of like Uber, Uber Eats, but for food donations. So we have networks across the country. We're in 27 different states right now, um, and we do about 150,000 pickups a year. Um, last year we recovered close to 37 million pounds of food. So it's a huge network. The the concept is pretty simple, but the actual day-to-day gets gets pretty complex.

Carolyn Woodard 2:28

Can you just talk a little bit more about ... I have seen the app - full confession, so we have Food Rescue where I am.

Joe Robbins 2:35

Oh, cool.

Carolyn Woodard 2:36

And we I, you know, I have friends who have talked about it and do the volunteering for it. And so I just tell tell us a little bit more how the app works. So that if it's like a restaurant or a grocery store that has like they made 10 pizzas and it's almost time to close, can they get on the app and just say, hey, can someone come get these pizzas?

Joe Robbins 2:55

Yeah, yeah, that's exactly how it works.

Joe Robbins 2:57

So there's there's a portal for the food donors, the receiving agencies, and then also the volunteers.

Joe Robbins 3:03

So for the volunteers, the way that it works is they can log on and they can say, All right, I want to do a pickup this Saturday. They can go find pickups that are near them, that are in their area. We have like vehicle sizes and kind of what's going to be in the pickup so they can make a determination on what they want to do. And then they can claim that and add it to their schedule. Um, and then they have some reports and some achievement badges and things like that. But but that's how it works for the volunteer.

Joe Robbins 3:27

For the food donor, um, they can sign up. There's kind of a vetting process to make sure they know, you know, what to expect, what condition the food needs to be in in order for it to be donatable. Um, but once they're on, then they can log a donation in the app.

Joe Robbins 3:41

And then we have site coordinators who go in and assign those donations to volunteers.

Joe Robbins 3:46

And that's similar to how the agency uh portal works as well. They're able to look at their upcoming pickup schedule. They can put in food preferences. So some agencies are, you know, canned goods and dry goods. They kind of take stuff in, they store it and they hand it out. Some folks are completely repurposing food. So they're taking raw ingredients and actually cooking meals and serving them and things like that. So they can put all that context in.

Joe Robbins 4:11

And then our site coordinators use that when they're routing, routing the donations. So it gets pretty complicated pretty quickly, but but that's kind of how it works for each of the stakeholder groups.

Carolyn Woodard 4:21

And so, I mean, one thing that we've been talking about with you know, IT at nonprofits in general, but definitely with more forward-looking or closer to the cutting-edge IT is that nonprofits that start out with kind of a technology app or very comfortable. What they do is related to technology, often are pretty well positioned to take advantage of new technologies that come out.

Carolyn Woodard 4:48

And other nonprofits, you know, no shade. But if you're like a more traditional nonprofit doing something a lot more traditional, um, it can, you know, like sometimes your IT is a little bit of an older legacy of IT, and then, you know, kind of your cultural values are not, you know, like more conscious about IT.

Carolyn Woodard 5:09

And so one thing I thought was so interesting about your organization is that clearly it was started, you know, like this app is very central to how you do what you do, but it's such an interesting bridge between a lot of food banks are more in kind of the old school, you know, tech friendliness area of nonprofit um organizations. So I find that so interesting. And um

Carolyn Woodard 5:34

I'd like to hear a little bit more, I think, about you. So your background, you're a technologist, and you are working at food rescue. So can you tell me a little bit about your your career?

Joe Robbins 5:46

Yeah, yeah, absolutely. And I'll say I I think like many folks, I'm a I'm a sort of accidental technologist. Um, I double majored in philosophy and theology in school. So that was that's the the worst maybe starting point to to technology you can take. Um,

Joe Robbins 6:01

I will say though, my my primary role as a product officer is is value creation and understanding what are our value propositions and how do we get there uh with these functionalities and using that to prioritize the roadmaps. And I will shout out my philosophy degree there. I think I think it helped a little bit. Um but yeah, so I I worked in um,

Joe Robbins 6:21

I was an array designer for for a startup solar company um in Washington State, which was really exciting. It was kind of a booming industry, um, and then went from there to a small kind of startup nonprofit um uh technology company. And so worked with them. Um

Joe Robbins 6:38

The way that worked was, you know, we had clients and products who kind of build a product around a client. And then once we rolled it out, we'd spin that product off as its own company. So I really did probably almost nine years in kind of startup technology roles completely by accident. And

Joe Robbins 6:56

I'll say the startup world and the nonprofit world have some overlap, some similarities in terms of A, the the severe resource constraints, which are just kind of day-to-day, the table stakes. Um, and also because of those resource constraints, it really matters, you know, on the job training and talent. Like you're not gonna be able to afford people with, you know, decades of experience in the background. So there's a lot of if you can do it, we're gonna let you do it. And if you can take on more, we're gonna give you more.

Joe Robbins 7:25

And so I think that was my that was really my path in was working in startups. Um, you know, my my role at the solar company was also a more technical role. Um, so I kind of had that in my in my back pocket, anyways. Um, but yeah, my my first

Joe Robbins 7:40

The first tech startup I worked for, they were hiring a scrum master. And I had to, before the interview, look up what that was and try to figure that out. Uh so yeah, but that was a great company, and they they for a startup especially, they had really good professional development programs. Um,

Joe Robbins 7:58

I did get you know professional certifications from MIT and in some basic coding and programming. Um, but my role has always really been managing the teams. You know, we do software design, I do all the prototyping myself. We don't have a UI UX designer, so that's me. Um, and yeah, the database and analytics, that's also grouped into the role.

Joe Robbins 8:18

So you become a jack of all trades pretty quickly. Um, and really the skill set, I think even people with who are coming out of college with a computer science degree, they're finding, especially now, you know, what matters is are you up to date on the most recent, you know, material UI package, you know, kind of the most recent, and some of those things you learn, and then a year later it's outdated already.

Joe Robbins 8:39

So I think anyone in the technology community knows you're learning all the time. You're gonna be a lifetime student. Uh, and so yeah, got a late start, but uh I I've caught up pretty quick.

Carolyn Woodard 8:51

And you so you moved to food rescue and uh into this technology role. So, in that role, what do you do? You support the the app and and what else do you do? And how did you get interested - well, I mean, everyone's interested in AI right now, but how - what was your path into seeing this AI opportunity?

Joe Robbins 9:14

Yeah, yeah. So so I started at Food Rescue about three years ago. Um, and I'll I'll say something my my product manager fellows will will understand. Um, every company has kind of a different idea of what it is a product manager does. Um, and I will say you end up, I think no matter what, doing a little bit of everything. So,

Joe Robbins 9:34

So when I got hired on, the you know, previous role was really it was kind of a management role that was maintaining kind of the workflow. You accept you know, user feedback and then you kind of design that into features and then ship those features. Um, we changed that up pretty quickly. We built in discovery processes, and so we have you know feedback loops from all 50 of our sites.

Joe Robbins 9:56

You talked a little bit about the diversity we have in the field, super valuable when it comes to feature design because you get feedback from so many angles. When we build something, we have to design it for all of those groups, um, even if they have very different operations. So that's a fun challenge, but but yeah,

Joe Robbins 10:12

That's the the starting point of my role is making sure that we're taking the user feedback in, keeping them in the problem space, and then we're designing the solutions. I do build the prototypes myself. Um, and then we we scope those with the team, we get estimates, we build them into sprints, and then you know, we've got some QA and stuff on the back end too.

Joe Robbins 10:32

And I'll say outside of that, you know, we're doing all the reporting tools, all the analytics for the organization. Um, we're doing marketing right now for a feature launch that's coming out um at the end of this month. We also shipped our our native mobile app uh last last week, actually. So that was a big had to figure out new deployment processes, thank you. New deployment processes, um, new code infrastructure. So a little bit of everything. Um, but yeah, we

Joe Robbins 10:59

We had a a research opportunity um with um Dr. Stanley Lim, who's a professor of supply chain logistics at Michigan State University. Um, and that was really cool. You know, they wanted some of our data, they had a proposal in mind.

Joe Robbins 11:13

And we got to the end of that and we were kind of talking about additional projects. Um, and and he has background in in AI. He had asked us about, you know, if we thought about integrating AI into our app. Um and

Joe Robbins 11:27

AI is is, I think, at different points of development in different sectors. Supply chain logistics is is ahead of the curve for sure right now, that a huge amount of resources go into that field. And so we knew that there was some going to be some overlap for sure there um in what what was possible and what we could do. Um, and his skill set is really kind of designing these from the ground up.

Joe Robbins 11:50

So when we integrate AI, you can take existing tools um and then kind of put your own context of your own business in into those tools and then plug them in essentially. Um not easy, but but easier than kind of building from scratch. Um, and

Joe Robbins 12:06

He was proposing that we that we do build from scratch. So so we met with the team and we determined as long as we can scope this project correctly, the value of the experience for us, for myself and and for the developers is going to be going to be massive. And so so yeah, we sat down to do that. We started that last summer. Um,

Joe Robbins 12:25

And the project we came up with was uh kind of a tightly scoped research question. And it was a research question that we had really good data coverage on. I think that's key too for folks who are are thinking about dipping their toes into this. It's really hard to to try to solve a problem first and then build the data capture after the fact. Um, I said really hard, it's impossible and gets very inefficient very quickly.

Joe Robbins 12:51

So I think that's part of something I kind of learned on the fly was that you know your data is part of your product.

Joe Robbins 12:58

And so for previous clients, that was really important. We were working in the health insurance industry. So all the any feature we designed, that was one of the first questions is what's the data capture? What kind of analytics are we going to need once this gets out? Um, and so we've been doing that since I started here.

Joe Robbins 13:14

So the the problem we landed on was um volunteer cancellations. So right now we have volunteers will claim a pickup. Um, it's on a specific day at a specific time, and then if they cancel within maybe 24 hours of that pickup window, really creates a problem for our site managers, many of whom are also volunteers. So typically they either the site manager has to go and cover themselves, which is obviously disruptive, or we run the risk of not going and conducting a pickup. So, so it's a free service. We don't have service level agreements with our donors, but because we're recruiting them, we have to make sure we we protect that relationship. And so so it creates a real problem for us. Um,

Joe Robbins 13:59

So that was kind of what we set out to predict. Is can we predict when someone's at a higher risk of canceling within 24 hours of the pickup? So we have good scope of data, obviously, around rescue or pickup history, um things that might affect a pickup window.

Joe Robbins 14:16

Obviously, these are volunteers, they're not paid employees. So there's a lot a lot of factors that could go into a cancellation decision. You know, someone has soccer practice late or something like that. Um, but we knew enough about it where we felt confident we could put something out to test. Um, so

Joe Robbins 14:33

So we sat down, we kind of scoped that out, we built the data features, we built the algorithm, and then we launched a pilot with 17 of our sites in January of this year. Um, and we've been kind of looking at the results, kind of fine-tuning the model. Um, and

Joe Robbins 14:50

We have about 12,000 pickups a month typically. That's about the average. And anywhere from, depending on the site, maybe seven to twelve percent of those end up getting having a last-minute substitution. So, so not an insignificant number whatsoever. Um, and yeah, the

Joe Robbins 15:07

The first version of the model we put out had um about a 12% hit rate, which if you think about as a letter grade, that seems bad. But when you think about what we're asking it to do um and what we're asking it to predict, 12%'s maybe 10 times as where we thought it would be in the first version. Um, and we've continued to improve it uh with every update we've pushed out.

Joe Robbins 15:30

So we pushed out about three or four major updates. Um, right now it's predicting at about 15%, a little higher depending on the site. Um,

Joe Robbins 15:38

And it's it's based on pretty straightforward data points. So it's really the the rescuer history. Um, days of the week, some days get canceled more frequently than others. The weather factors into it. So we pull some weather data in from a third party. Um, interestingly enough, we learned that people are less likely to cancel when it's cold and more likely to cancel when it's very hot outside, which is interesting. You wouldn't think that. Uh, but yeah, so that's how it predicts it. Um,

Joe Robbins 16:06

And the way that we built it, I think this is also important for other folks who are looking to do something like this. The way that we built it is we're not automating any decision making with the tool. So the way that the tool works is it runs the algorithm, it assigns scores on cancellation likelihood scores to each rescue. And then if they're over a certain likelihood threshold, we create a to-do item for the site coordinator that says, hey, this is a higher risk rescue. Let us know if you want us to send a confirmation.

Joe Robbins 16:36

And then what we do is uh when that uh uh to-do item is reviewed, then we send a confirmation to the volunteer and say, Hey, just so you know, you have a rescue in a couple of days, uh, if you're still gonna make it, let us know. If not, please cancel now so that we can we can find some find a substitute for this. So,

Joe Robbins 16:55

So in the last you know, six months, this has been thousands of rescues that without this tool, it would have been a last second cancellation. We'd be finding out day of and scrambling. Uh, and instead we have you know three to eight days is the window, uh, three to eight days of notice where we can assign a substitute. So, so really fantastic results.

Joe Robbins 17:16

I talked a lot there, so I'll let you get a question in now. But uh yeah, it's a big project, a lot going into it.

Carolyn Woodard 17:21

Yeah, no, I just uh when you when I heard you talking about it previously, it's so fascinating because I think a lot of nonprofits, you know, like the the you know, the way in to AI is you start using those productivity tools, right? You it helps you draft an email or it helps you, you know, with your inbox.

Carolyn Woodard 17:41

I found this was so interesting as a way to, you know, kind of think about what AI is good at doing, like predicting patterns from data that you already have and utilizing that in a way that's very practicable for what your nonprofit does. And that's another thing that was so fascinating about this story.

Carolyn Woodard 18:03

I'm sure people have questions about, you know, like the security of how how do you manage, you said, you know, the AI doesn't make any decisions itself. It's giving the decisions to the people who are involved.

Carolyn Woodard 18:16

So can you talk a little bit about like your design process around those, that aspect of it?

Joe Robbins 18:22

Yeah, yeah, absolutely. And we we approach it kind of the same way we would approach, you know, more straightforward forms of data analytics.

Joe Robbins 18:30

So when we're building, we can we can obviously create an analytic and then we can trigger actions based on you know that number. So so if something is, you know, if something, the pickup window is coming up and there's no one assigned to it, we can trigger an action off of off of that data.

Joe Robbins 18:47

And so the way that we make the decisions for for standard data analytics is this something that we're extremely confident in in kind of this analytic and what it's telling us, or is it something that is going to be valuable context for for a person who then needs to be the one who makes the call? Um,

Joe Robbins 19:06

So we approached it the same way. And and obviously we we also picked a research problem that that can be answered in that same way. Um, but I think that's really important. I think that that software um in general, it's not unique to AI. Software kind of hides a lot of the mechanics of what's being done. And so there's always going to be a decision there of, you know, what do we surface to the user and then what is going to to detract from the user experience. Um

Joe Robbins 19:35

So I'll say that that that decision making process allowed us to think about it in a very similar way that we think about the rest of our software. Obviously, the the unique concerns come to come with the fact that you know, if you use a third party, then you're translating data to to a system that's outside of yours. We didn't have to do that, luckily. Um, but that's the frequent, that's a really big challenge for especially nonprofits where you're you're under-resourced anyways. A lot of the times you're not going to have a compliance officer or or someone on staff who's going to write those policies. Um,

Joe Robbins 20:12

So we did put together, you know, an AI governance policy. Um, and we we grabbed a template that we found, you know, it wasn't uh it wasn't uh anything new, but we did have meetings with the whole team, even people who weren't involved with the build, and just made sure we all understood kind of what this was going to do, how the data was going to be used. Um

Joe Robbins 20:33

Similar to standard data analytics, you know, we don't send any personally identifiable information um into the algorithm. It's all um ID codes and things that are they're gonna be useful for us answering the question. Um,

Joe Robbins 20:47

And that's something that is dictated a little bit again by you know the functionality we needed. There's projects where you can't avoid that, and then you really have to make sure you kind of lay the groundwork for your safety and and liability there. Um, but yeah,

Joe Robbins 21:02

I think for the most part, it's I think the best way to approach it is start out with your standard data capture processes and making sure those are the right data usage policies, everyone on the team understands those. I think that's the most critical piece is a lot of times folks are, especially if you're SOC2 compliant or something like that, you've got a policy for everything. And one person has maybe read those policies, and everyone else is kind of kind of shooting from the hip a little bit. So I think visibility is key there. Um, and then also making sure you understand what are we doing, what do we need to accomplish what we're doing, um, and then making sure you have have those bases covered.

Carolyn Woodard 21:40

I feel like the way that you - um correct me if I'm wrong, but it seems like this part of this tool is using AI in a way that's maybe similar to like your GPS telling you where how to get to the location. So it's uh maybe a lower risk of the user being like, what is this AI doing in my app? You know.

Joe Robbins 22:02

Yeah, yeah. And that's kind of it's kind of the same approach we take to, you know, data usage and data capture, anyways. You know, if it's if it's about how a user is using your tool, and if it's um information that is that is specific to those use cases and it's making the tool better, then it's it's fairly straightforward, and how you handle those things.

Joe Robbins 22:24

We also don't have, you know, we don't have any financial information. We don't fall under PCI compliance. We really have that the sensitive information that we have is we have names, we have addresses of businesses, which are pretty public, and we have contact information. And that stuff is pretty simple. And that stuff is never really going to be important for an AI tool.

Joe Robbins 22:46

So so I think that's the key thing is start out with your your data policies that you have in general. AI doesn't have to necessarily complicate those drastically.

Joe Robbins 22:56

And then same thing for implementation. I think in general you wouldn't want to to take a metric you're not super confident about what it means and then make decisions for users based on that. You want to implement those into the the the context window that the user's in when they're taking actions um and yeah I think that'll that'll get most folks pretty far in the process um

Carolyn Woodard 23:21

I think I have another question maybe another couple of questions for you um one is if you so your organization is very tech forward clearly um do you have advice for anyone listening whose organization is maybe not that edge of the cutting edge and um you know to think about once you've used AI for those productivity gains and you're getting more conversant and more comfortable with what AI can do and what it's good at doing -

Carolyn Woodard 23:56

How do you think about maybe a project that you could try with your organization with what your mission is if you don't have a researcher who's coming to you and has expertise in AI and can help you with the logistics of it and all that sort of thing. Do you have you know just like a framework or some prompts to think about where AI could fit in what your mission is

Joe Robbins 24:19

Yeah yeah and I I think you you framed this the right way too

Joe Robbins 24:23

I think the right entry point for for especially a organization that's maybe less tech forward is all of these you know assistant tools have free versions. And I think using those those heavily teach you kind of the the the basics of AI instrumentation.

Joe Robbins 24:42

Context is really important the that's the biggest limiter even the best AI tool it's not going to be able to help you unless it knows kind of what you know in terms of how to make the decision. So I think that's the the right entry point.

Joe Robbins 24:57

And then I think I think there's a lot of a lot of hype and excitement about AI tools. And I will say I at least in my experience not even in non nonprofits in in the for-profit sector as well data capture and database storage, data structure, a lot of those things are things that that are really easy to to lag behind on until someone asks you for something.

Joe Robbins 25:21

And then there's a lot of going back and cleaning up and so I I think that's making sure you have a a healthy data capture system um that's really step one because AI assumes all your data is correct. It's all clean it's all being stored and updated properly that has to be the first place to start.

Joe Robbins 25:40

And I I think I don't want to sound harsh but I I think you gotta you gotta start there. I think especially if you're a uh a more technically limited organization you have to do have the data first before before AI is going to be going to be useful for you.

Joe Robbins 25:55

And I think there are usages that are are less uh maybe less predictive and more more you know maybe you have a knowledge base and you want people to be able to interact with that knowledge base instead of searching for an article and then reading it. You know there's things it can do before before you're you're predicting things and you get heavy into the data analytics side of it.

Joe Robbins 26:16

But I think that would be the place to start is you know use the tool get a basic understanding of of how those tools work.

Joe Robbins 26:23

And then I think you got to start with an audit of what data do we have right now in the organization what needs to happen with that data before it's it's usable um before we can connect it to an algorithm or an AI tool and then have that tool look at that data.

Joe Robbins 26:41

And then sometimes you need to to build new features as well. You know if you have a a journey that you want to make a prediction about and you you have holes in your visibility on that journey for your users or your stakeholders those holes are going to to limit the the accuracy of any predictive work you're doing on that journey. And so I think that's the place to start is do an audit of your your data capture, your data storage mechanisms.

Joe Robbins 27:06

And then I think from there you'll you'll come out of that process with the to-do list of you know we need to figure this out we build a data lake maybe um we need to make sure that we we're cleaning this data and we have fact tables in addition to to functionality tables. And then there might be new features or new new data capture mechanisms you need to bring in.

Joe Robbins 27:27

And I keep saying features because I'm assuming you know you have a software that's in-house if you're using you know third party tools most tools have you know export options and ways to to get data out of them. They have webhooks and integrations. So that's another good place to start is you know if you're using Salesforce if you're using HubSpot there's really a a rich data set already living in there it's going to be pretty clean in terms of how it's stored and how you can pull it.

Joe Robbins 27:55

So starting with that see what we have in the inventory seeing where the gaps are um and then you can kind of start start planning your AI project those are really those have to happen first.

Carolyn Woodard 28:08

That makes sense. I love something else you said though too about piloting. So you were very careful to have the algorithm and the data that it was using only in certain sites to begin with and then see what happens, is it doing what you think it's gonna do, what are the impacts, how are people reacting to it - and then you can with what you learn you can make it better and expand it.

Carolyn Woodard 28:29

That's something that I think unfortunately a lot of nonprofits don't have a lot of time to play around with pilots for this sort for technology like they might do pilots for a program that they're gonna start in like one school and then they want to do it in 50 schools - but they often don't use that same logic toward um IT systems or or tools. A nd often they don't get - they don't have the luxury of being able to do them, right, because they got the grant to do the product that does the thing. They can't like play around with, well let's just use the CRM on like two of our sites instead of the whole shebang, so um but

Carolyn Woodard 29:09

I think AI AI tools themselves also lend themselves to doing pilots and seeing how it goes and asking the AI how did it go? You know, like give me the results and the analysis and then taking it to a larger audience. So I like you know that you said that too.

Joe Robbins 29:27

Yeah a pilot can be can be tricky because you're already investing in a new tool you're you're changing operations for people and so that's a lot of a lot of investment already and then to delay the the return on that investment to do a pilot can be tricky.

Joe Robbins 29:43

I will say two things is is I learned you know we're we're grant funded and so I learned that if you want to do a pilot build a pilot into the grant a lot of funders are open to that if they you know if they know upfront that that's going to be a part of the process. So I think that's a that's a good tip for folks.

Joe Robbins 30:01

The other thing I'll say too and this is not this is not best practices um but it's something that you know in the startup world and I think in nonprofits sometimes it's it's the best option you have is, if your users understand kind of where you're at, sometimes you can get away with you put something out, you know, you're not pilot testing it but you put something out maybe label it and say this is a new feature or this is a new tool and make sure your users understand that their feedback is what's going to make that better.

Joe Robbins 30:31

Because that can allow you to you know I've I've worked on teams that have you know four or five QA engineers and so everything before I see anything it's been really well kind of went through and and and prepped and researched.

Joe Robbins 30:45

And then I've worked on teams where I'm doing all the QA so I'm having to go and the results are going to be different and that's just sometimes that's just the the environment you have to work in. So I'll say that's a that's an option you know if as long as you have the user feedback loop set up that's the critical part.

Joe Robbins 31:01

And then if you can't do a full pilot as long as your users understand this is a new thing, you know, we're relying on you to help us kind of get it dialed in and get it to to that final version if they are can be stakeholders in that process and not not just end users, right? Then that can be really helpful and and you're gonna make a better product that way even if even if you have a a huge team and all the resources in the world you're still kind of assuming you know what the users are going to need and and you could build something that's perfect.

Joe Robbins 31:32

And if the users get confused on page one and they can't they can't interact with it then you've still missed the mark right so so I think that's something for for smaller groups if you can't do a full pilot see if you can let users know that it's new let it know that it's in a maybe a beta phase is what you can call it um and then they're motivated to to give you that feedback they understand that that's the ask um that it's not a final version that can be really helpful too kind of an on the fly on the fly pilot.

Carolyn Woodard 32:01

Yeah I think that's true. I think a lot of nonprofit um you know users and constituents and volunteers will give you a lot of grace if you're transparent, like we're in this together we're trying to make this better can you help us?

Joe Robbins 32:14

Yeah yeah and it seems to you know if you ask someone for feedback they give it to you and you fix the problem that's a really positive experience. If you send something to a user they find a problem and they tell you about it and then you still fix the problem then it feels a little different for them which is it's such a funny psychology thing that you you kind of learn on the job and product. But yeah frame the release

Joe Robbins 32:37

Frame the release with where you're at and be transparent about that and most of the time you're gonna get really good results.

Carolyn Woodard 32:43

Yeah. Okay I have a last question for you because I do appreciate your time and I know we're running out of time. What's next? Like do you have an AI project that you're thinking about now that you've done this one you're like oh I've got an idea...

Joe Robbins 32:56

Yeah yeah so I I think um not giving too much away um we've got some things coming out this year that are going to really add pretty significant data capture to to what we have now um data capture on a couple of really important parts of our workflows. So we're gonna be in a position to kind of like I said earlier to tackle a couple of problems that are maybe out of our reach right now.

Joe Robbins 33:22

But I think our focus is on that site coordinator role. Right now there's a huge amount of pressure on that role. Those people are managing pretty complicated logistics but they're also recruiting you know all three of those stakeholder groups kind of keeping everyone engaged. They're really there

Joe Robbins 33:38

We're asking them to be community leaders and you know supply chain logistics experts. So I think there's a a huge opportunity for us in helping to not automate their role but automate the some of the trickier decision making components of that role.

Joe Robbins 33:56

And if we ended up in a situation where where our site coordinators are fully focused on maintaining those relationships and and building those communities and they were not thinking about how many rescues they have on the schedule and and who's going where and what's going where, I think that'd be a really good position. So yeah, so I think that's the the area focus for us um and we do have had some new things coming out so stay tuned for that.

Carolyn Woodard 34:23

I love that I love that because you know we've been saying for a while now that like the humans need to do the human stuff and that I think the potential of AI in the nonprofit sphere and in philanthropy is we're so understaffed and so under resourced and or you know like the resources are just too expensive.

Carolyn Woodard 34:43

Like we can't hire a fancy consultant to tell us about our supply chain problems, you know, it's just out of reach.

Carolyn Woodard 34:51

So having the AI be able to handle the like logistical stuff that AI is pretty good at like scheduling or you know predicting those sorts of things like the busy work that you're like as a volunteer you don't want to be spending all of your time filling out forms for example, you know, and then letting the volunteers focus on that human side ...I really, there's a lot of potential there.

Carolyn Woodard 35:15

So thank you so much for sharing with us all these ideas and your experience I really appreciate it.

Joe Robbins 35:22

Yeah thank you for having me. It's been a pleasure