WEBVTT

1
00:00:03.560 --> 00:00:16.999
Neon One Webinars: Hey everybody, it is good to see you all logging in. I always love my early birds. Could y'all, if you are logging in, please do me a favor and let me know that you can see and hear us. Ideally, you can see my screen.

2
00:00:17.420 --> 00:00:25.189
Neon One Webinars: And then do a little sound check for us. I'm looking at the chat, just drop your answer there. Maybe tell us where you're tuning in from.

3
00:00:25.320 --> 00:00:30.190
Neon One Webinars: I'm tuning in from Lakeland, Florida. It is incredibly hot.

4
00:00:30.630 --> 00:00:40.180
Neon One Webinars: It is… let's see… 105 right now, as it feels like. Thank you, Cathy, I really appreciate it.

5
00:00:40.390 --> 00:00:46.039
Neon One Webinars: Kennesaw, Georgia. Very cool, Lorraine. Samuel, where are you joining us from?

6
00:00:46.040 --> 00:00:54.189
Samuel Chen: I am in Toronto, and the weather's amazing. It's sunny, it's… I'm converting it in my head. It's 73 Fahrenheit, I think?

7
00:00:54.190 --> 00:00:56.040
Neon One Webinars: Oh my god.

8
00:00:56.740 --> 00:01:06.890
Neon One Webinars: That is gorgeous. Erick in Spokane, lovely. Let's see, can you guys see my screen? Let's see.

9
00:01:07.340 --> 00:01:08.390
Neon One Webinars: share there.

10
00:01:08.910 --> 00:01:12.329
Neon One Webinars: Awesome. Participants can now see your screen. Cool.

11
00:01:13.360 --> 00:01:23.820
Neon One Webinars: I'm very envious of everybody who is experiencing lovely weather. We will not get lovely weather in Central Florida until at least late September.

12
00:01:24.670 --> 00:01:46.859
Neon One Webinars: Alright, thank you everyone so much for being here. I am really excited to talk about this topic today. Before we get into it, I want to introduce myself really quickly. My name is Abby. I've been in the nonprofit tech sector for a long time, and I'm really, really passionate about understanding nonprofit technology, how nonprofits can use technology to connect with their donors.

13
00:01:46.860 --> 00:01:52.279
Neon One Webinars: And then my job is to understand what donors want from the relationships with the nonprofits they have.

14
00:01:52.280 --> 00:01:59.179
Neon One Webinars: And then share that with you all. I do have a speaker with me. Samuel, do you want to introduce yourself?

15
00:01:59.820 --> 00:02:11.299
Samuel Chen: Hey, I'm Samuel. I'm the product manager, of AI Intelligence at Neon One. I actually own reporting, all of the data, and any automations that's on the platform, like workflows.

16
00:02:11.300 --> 00:02:27.040
Samuel Chen: I've previously led product teams to build AI features across different industries, like finance, healthcare, oil and gas, and advertising, and Nonprofit is one of those that I'm super interested in. And I think some of you in the chat have actually spoken with me before, so it's nice to see you all here again.

17
00:02:27.410 --> 00:02:28.830
Neon One Webinars: Good.

18
00:02:29.090 --> 00:02:33.759
Neon One Webinars: It's always so nice to have familiar names in the chat.

19
00:02:33.850 --> 00:02:45.459
Neon One Webinars: Speaking of that, since Samuel's alluded to the fact that he's talked with some of you, I want to just, like, set some expectations really quick. This session is not going to go deep into AI tools in Neon One itself.

20
00:02:45.460 --> 00:03:05.669
Neon One Webinars: We certainly have some cool stuff coming up, but this session is more of a high-level educational, situation. So, if you use Neon One, if you don't use Neon One, we're all going to learn together. And there are going to be opportunities coming up in the next few weeks where you can learn more about in-app

21
00:03:05.670 --> 00:03:10.389
Neon One Webinars: like, AI-powered tools, but this session is going to be much more high-level than that, so…

22
00:03:10.530 --> 00:03:35.169
Neon One Webinars: That said, please do talk to us. It is the pits to get on a webinar and feel like you were just shouting into the ether, so if you want to talk to us, do so in the chat. I'm going to ask questions, you can ask questions, talk to us in the chat. That said, the chat can move pretty fast, so if you have really explicit questions that you want to make sure that we get into, drop those in the Q&A box. We are much more likely to see them.

23
00:03:35.550 --> 00:03:47.110
Neon One Webinars: And with that on there, we do have someone who is manning that chat also, to keep an eye on that. I saw Jean is asking if anyone else is hearing a severe echo. Am I echoing?

24
00:03:47.690 --> 00:03:49.170
Neon One Webinars: Let me know.

25
00:03:49.430 --> 00:04:00.970
Neon One Webinars: while we do that, I want to just, like, kind of figure out what… where we all are. How do you feel about using AI tools in your day-to-day?

26
00:04:01.110 --> 00:04:13.850
Neon One Webinars: Are you kind of with Moss here? You need some help? Are you BFF with your various robots, like Fry from Futurama? Let me know where you are. I'm probably somewhere in the middle.

27
00:04:14.040 --> 00:04:38.790
Neon One Webinars: I do use AI sometimes, but it is not, yet a huge part of my workload. It's very much in the background. Lorraine needs help. Lorraine, you and I are kind of in a similar boat. I have my little comfort zone where I operate, and I am learning to stretch outside that comfort zone. We got the basics, lots to learn all day, just not for image generation. Erin, good call. I love that. We, we looked at some survey data and

28
00:04:38.790 --> 00:04:42.439
Neon One Webinars: People do not love AI-generated images, so I like that you're not using it for that.

29
00:04:42.600 --> 00:04:55.430
Neon One Webinars: Okay, cool. It sounds like we're all kind of on the same page, we have a lot to learn, so with that said, we have an expert here who is very good at explaining how things work, so let's ask some common questions.

30
00:04:55.670 --> 00:04:57.490
Neon One Webinars: So, Samuel, what…

31
00:04:57.760 --> 00:05:04.200
Neon One Webinars: what are the different kinds of AI that are out there, and how do those… how do you build those tools?

32
00:05:04.420 --> 00:05:15.639
Samuel Chen: Yeah, I think this is a really good question to start with, because the word AI gets used for a lot of different things now, and I'm gonna simplify it down to two types that really matter for you, and here's a secret up front.

33
00:05:15.640 --> 00:05:39.679
Samuel Chen: most of them are just prediction machines. The only difference is what they predict. So the first type is what you might hear, which is a subset of AI, it's called machine learning, and this has… this has been one that's around fundraising in all different industries for decades. Think about a major gifts officer with 20 years of experience. They can look at a donor record and just know intuitively, if this person is ready for a bigger ask, or this donor is about to drift away.

34
00:05:39.680 --> 00:05:48.930
Samuel Chen: And they just can't explain why. It's intuition built from thousands of donors they've seen over their career, and there are machine learning models, that do this.

35
00:05:49.270 --> 00:06:13.650
Samuel Chen: So generally, a machine learning model is that intuition, just bottled into math, right? It takes historical data, thousands and thousands of records, giving histories, and then the model learns that pattern. That's what you've probably heard when people say they're training a model. And it then makes that prediction, which donors are at a risk of lapsing, who has a higher propensity to give, what's a good ask amount for this person.

36
00:06:13.650 --> 00:06:16.140
Samuel Chen: It's that gift officer's gut feeling

37
00:06:16.140 --> 00:06:20.139
Samuel Chen: That's now a mathematical, equation.

38
00:06:20.140 --> 00:06:42.999
Samuel Chen: Now, the second type is what most people mean when they say AI today. Those are the chatbots, ChatGPT, Claude, Gemini, I'm sure all of you have used these tools and gotten a lot of benefit if you've already been experimenting, too, and see how powerful they are. And these are powered by something called a Large Language Model. And they're essentially trained on everything humans have ever written.

39
00:06:43.000 --> 00:06:53.769
Samuel Chen: And here's the surprising part. It's actually just a prediction machine. It just predicts something different. It predicts the next word. And I'll talk about how that works in a second, because

40
00:06:54.100 --> 00:07:18.769
Samuel Chen: I think there's gonna be a question that brings this up. But for now, the thing you need to know is how these are too differently built, right? Machine learning models are trained on your data. It would be on, let's say, your donors giving history, and nobody else has that. So it only makes sense for us to build those ourselves if we ever do. Large language models are the opposite. They're trained on the

41
00:07:18.770 --> 00:07:42.619
Samuel Chen: data, and training one costs hundreds of millions of dollars and enormous amounts of computing power. A handful of companies have done that, Anthropic, OpenAI, Google, these companies you might hear in the news cycle these days, and it would make no sense for us to build those from scratch, just the same way you don't build your own email server. You use the ones that's best available, and you put into energy into what you can do.

42
00:07:42.620 --> 00:07:56.840
Samuel Chen: So the job for the Neon Ones data intelligence team isn't really building the brain with these AI tools. Our job is taking a world-class brain, teaching it to be an expert in nonprofit fundraising, connecting it safely and accurately to your data.

43
00:07:57.560 --> 00:08:11.539
Neon One Webinars: I really appreciate that, because it… I like the distinction between training something on your data and then training something on the world's data, and I never really thought about these tools, especially predictive models, being

44
00:08:11.590 --> 00:08:29.999
Neon One Webinars: mathematical expressions of human intuition. I think that's really interesting. And you're right, I did have a question about LLMs specifically. So you've kind of told us that LLMs are predictive, and it seems like they're kind of guessing what word is going to come next in a series of words. Is that how they work?

45
00:08:30.500 --> 00:08:54.649
Samuel Chen: Yeah, so if we open it under the hood, and we run it, like, one tick at a time, you know how when you use ChatGPT, it's streaming the text out? What's actually happened is much simpler than you think. It just does one thing over and over again. It reads everything it's written so far, and the prompt that you give it, and then it just predicts the next word, and that's it. It writes one word at a time, reads everything, and then predicts the next word, and it does this incredibly fast.

46
00:08:54.650 --> 00:09:19.240
Samuel Chen: Which is why it seems like such a… such a conversational experience. And here's the part that also surprises a lot of people. It's not looking up a pre-written answer in a database somewhere. There's no… it's not like there's a filing cabinet of responses, or it's programmed. Every time it generates copy, it's composing it word for word on the spot. And even if you ask the exact same question twice, it's writing the answer fresh

47
00:09:19.240 --> 00:09:37.319
Samuel Chen: both times, and that's why you get these little quirks with these LLM solutions. And for you to understand this intuitively, most of you actually have a tiny version of this in your pocket if you have a smartphone, and it's been around for around 10 years. I don't know if you can all guess it, but it is

48
00:09:37.320 --> 00:09:56.249
Samuel Chen: the autocomplete on your phone. It's not the exact same model, but it's very similar in concept. Your phone reads what you typed, and then it suggests the next word. And just imagine if that autocomplete has read essentially all human text, every book, every article, most of the internet, and where your phone can finish the sentence.

49
00:09:56.250 --> 00:09:57.849
Samuel Chen: an AI tool can

50
00:09:57.850 --> 00:10:05.190
Samuel Chen: finish your appeal letter with your voice, if it's prompted the right way. So where does this knowledge come from?

51
00:10:05.340 --> 00:10:18.930
Samuel Chen: it's… it's like all that reading. It absorbs the patterns, the grammars, the facts, it learns how a thank you letter sounds different from a grant report, and no one programmed the rules in. They didn't memorize those documents, because

52
00:10:18.930 --> 00:10:42.950
Samuel Chen: And because of this, it also leads to things like hallucination, which I'm sure we'll dive way deeper into. But because it has also been trained on everything, some people say that there's a staleness to it, right? It then gets rid of all the creativity, and then it all sounds the same. We call it AI slop, and that's why it requires such intentional prompting.

53
00:10:42.950 --> 00:10:51.390
Samuel Chen: Or a team like ours, where when we're building something, we want to imbue the CRM's knowledge into it, and to make it one-of-a-kind and personalized.

54
00:10:52.010 --> 00:10:58.490
Neon One Webinars: Okay, so you… to use some of your imagery already, so we are using…

55
00:10:58.570 --> 00:11:15.879
Neon One Webinars: a brain that's trained on, like, all of human data, and then we, when we're writing prompts, or even building tools in Neon One, we are kind of telling the brain, like, we need you to think this way when you're completing this task, and then it's kind of making predictions about how the words should flow that way, is that right?

56
00:11:15.880 --> 00:11:27.399
Samuel Chen: Essentially. It's just predicting the next word, but because it does it so well and cohesively, it looks like it's writing, it's thinking, it's reasoning.

57
00:11:27.670 --> 00:11:46.530
Neon One Webinars: It's so funny to… to think about it as being, like, essentially a very advanced autocomplete. Did you guys ever do that thing, like, on Instagram, or even, like, I would see it on Facebook, it's like, oh, answer this, or type this phrase, and then hit autocomplete until it tells you, like, what your personality is, or whatever. That's…

58
00:11:46.950 --> 00:11:50.829
Neon One Webinars: One, it's old tech, so it's funny to think about the fact that, like.

59
00:11:50.890 --> 00:11:59.779
Neon One Webinars: LLMs are an extension of that, and two, it's just… it makes… I don't know, it makes me feel quite old, because no one uses Facebook anymore. So…

60
00:11:59.790 --> 00:12:14.480
Neon One Webinars: Okay, so that's how LLMs work, and we've kind of acknowledged that they're quite different than some of the more, like, specific data analytical part of things. So, when we're building AI tools.

61
00:12:14.810 --> 00:12:20.330
Neon One Webinars: And we are using those tools to…

62
00:12:20.410 --> 00:12:37.020
Neon One Webinars: analyze data or predict what's going to happen next. How do tech companies like us keep individual donors' information safe while also understanding how individuals fit into large

63
00:12:37.250 --> 00:12:39.060
Neon One Webinars: Trends and behaviors.

64
00:12:39.330 --> 00:12:45.450
Samuel Chen: Yeah, this might be one of the most important questions of today's session, so I'm gonna be really concrete.

65
00:12:45.450 --> 00:13:08.420
Samuel Chen: Remember how I said these models are trained by reading enormous amounts of text? The number one fear people have that I hear typically is a very reasonable one. It would be, is my donor data becoming part of that, right? Is the AI learning from my donors, and then using that knowledge for someone else? Is there that leakage? And the answer is definitively no. And I want to explain exactly how we guarantee that.

66
00:13:08.420 --> 00:13:11.879
Samuel Chen: Because it comes down to, like, one architectural choice.

67
00:13:11.880 --> 00:13:35.570
Samuel Chen: There are two ways to give an AI knowledge. One is you train it, right? You feed the data into the model, and then it becomes part of that model permanently, baked into its memory, and that lesson can't be unlearned. And the other is what we do. And the best analogy is, like, an open book exam versus memorization, right? And we never train the model on your data. Instead, what

68
00:13:36.110 --> 00:13:59.969
Samuel Chen: the AI tool we're currently building is called Gen, soon to be launched. If you ask them a question, our system would go into your secure database, it'll pull the relevant and specific records to answer that question, it shows it to the model at that moment, the model reads it, answers it, and then the exam is over. The book is closed, nothing was ever memorized, and it does that over and over again. The next question will always start fresh.

69
00:13:59.970 --> 00:14:06.260
Samuel Chen: So the data stays and lives in your database the entire time, and it never becomes part of the model itself.

70
00:14:06.260 --> 00:14:29.920
Samuel Chen: And the second one is just contractual and legal. We only work with AI providers through enterprise agreements, and that includes something called a zero retention guarantee. That means the provider is contractually prohibited from logging any prompts, AI answers, any of that from training the models. It's a legal binding contract.

71
00:14:30.260 --> 00:14:53.710
Samuel Chen: And that's really the key difference from using something like a free version of ChatGPT, or even a paid one, where you paste donor information into a consumer AI tool. And you… if you're all doing this, I really implore all of you to go into settings and make sure you turn off the consent to them using your data to train their models, to improve the product there.

72
00:14:53.710 --> 00:15:02.670
Samuel Chen: Because that text can be stored and used to improve their models. And if you're using a free version, you're essentially trading your data for a free service.

73
00:15:02.670 --> 00:15:05.970
Samuel Chen: A few more commitments that we've made,

74
00:15:06.140 --> 00:15:28.310
Samuel Chen: in our AI governance policy that I think is worth knowing is your data is yours, full stop. We build off of that principle. Every AI feature can be turned off by your admin, and if you turn it off, your data is not processed by AI at all, if that is a concern. And anything AI-generated is always clearly labeled as AI-generated, and I think…

75
00:15:28.320 --> 00:15:34.939
Samuel Chen: Abby can talk to a little bit about this, but we also build our AI in accordance with Fundraising AI's ethical framework.

76
00:15:35.850 --> 00:15:49.679
Neon One Webinars: That is really important, and if any of you… so, I know that a lot of you mentioned that you either use AI tools a lot, but not necessarily daily, or you use it very basically. If you are looking to understand

77
00:15:49.690 --> 00:16:08.190
Neon One Webinars: like, beyond… like, we're talking about the technical aspect of things here today, but if you want to consider, like, things like how to use it ethically, how to use it to support your organization's work, if you are looking for information, training information, any kind of information, I would really encourage you to look at fundraising.ai.

78
00:16:08.190 --> 00:16:10.839
Neon One Webinars: I'm actually going to drop it into the chat.

79
00:16:10.950 --> 00:16:27.860
Neon One Webinars: check them out. They have put together a very extensive framework for how people like you can use AI tools ethically and safely, but they also have put together frameworks that tech companies like Neon One have built into our own systems.

80
00:16:27.860 --> 00:16:34.280
Neon One Webinars: Because it is incumbent upon you and us to keep everybody safe as we are using these tools. So…

81
00:16:34.310 --> 00:16:39.000
Neon One Webinars: Check that out. It is a lot of really great work.

82
00:16:39.070 --> 00:16:42.900
Neon One Webinars: So, I like that you called out,

83
00:16:42.970 --> 00:16:53.510
Neon One Webinars: that pasting data into free instances of ChatGPT, or free instances of Cloud, or whatever it is, you are going to be… that data is going to end up training those models.

84
00:16:53.510 --> 00:17:11.749
Neon One Webinars: So, I want to, like, lay out there… I actually… Aaron said it perfectly in the chat. Rule of thumb, AI built into your CRM platform is safe. Don't take donor info out and put it into a free chat GPT window. And if you're putting it into another system, like Chat or Cloud.

85
00:17:11.810 --> 00:17:25.619
Neon One Webinars: and you have a paid version, if your organization is paying for that, you can go into your settings and tell it that it may not use that data to train its models further. That is a really important way to keep your people safe.

86
00:17:26.410 --> 00:17:44.730
Neon One Webinars: even with the best data, though, we all know that AI tools can sometimes hallucinate, and so Samuel, I want to understand a little bit about, like, why do these tools hallucinate? How can we prevent it? And then, like, if it inevitably does happen, how do we catch those mistakes before they

87
00:17:45.080 --> 00:17:47.790
Neon One Webinars: Kind of cause any kind of confusion.

88
00:17:47.790 --> 00:18:12.300
Samuel Chen: Yeah, so we couldn't really build on the knowledge that we just built together from, I think, two questions ago. Remember when I said, that these models work by predicting the most plausible next word? So here's the catch with that. The plausible usually means true, but it's not always true, right? The model doesn't actually know facts. It was built to produce text that sounds right, and most of the time, it's correct.

89
00:18:12.680 --> 00:18:21.499
Samuel Chen: some of the times, it's wrong, because there are texts with wrong information. But when it doesn't know something, it won't stop and say, I don't know.

90
00:18:21.750 --> 00:18:26.040
Samuel Chen: Doesn't always do that. It can make up statistics,

91
00:18:26.110 --> 00:18:50.799
Samuel Chen: in the industry, that could mean a fake donor name, a campaign total that isn't real, and it can make it sound plausible. That's what a hallucination is. The mistake is more likely the more specific the question can get, and you ask an AI, what makes a good year-end appeal? It can do great, because it reads thousands of them. It asks, if you ask it, how much did our fall gala raise in 2024,

92
00:18:50.800 --> 00:19:15.319
Samuel Chen: and it doesn't have access to your data. If you do this with ChatGPT, let's say, it has no choice but to invent something plausible, right? Specific numbers, names, and dates about your organization is usually where raw AI is just the weakest. So that sounds horrible. It might sound like I'm scaring you not to use any AI tools. But how do we catch it? And there's two ways. There's what we do on our side, and what you can do.

93
00:19:15.570 --> 00:19:24.030
Samuel Chen: What we do is… is very simple. We never let the model answer questions about your data from memory.

94
00:19:24.030 --> 00:19:39.370
Samuel Chen: for instance, in the future, we're gonna talk about this feature in another session, but if you ask Jen how much a campaign raised, it doesn't predict a plausible number. It runs an actual query against your actual database and reports what it found.

95
00:19:39.370 --> 00:19:48.899
Samuel Chen: The AI's job is to understand your question and explain the results, not actually be the source of the number. The database is the source, right?

96
00:19:48.900 --> 00:20:13.899
Samuel Chen: And this is the part where I want you to remember the most. When we build AI features, we prioritize sources for you, so you could double-check and verify the answer itself. And that's… that's what's really important here, that human in the loop, where you go in and you double-check these things. We're not removing, like, entire processes and delegating it to AI. We want that human in the touch for you, so your donors and constituents, volunteers.

97
00:20:13.900 --> 00:20:20.240
Samuel Chen: members can all feel that genuine human part of it. And… and that is, like, just such an important part.

98
00:20:22.230 --> 00:20:39.590
Neon One Webinars: That's perfect. And I was gonna say, like, ask. So, one thing that I hear joked about a lot, like, treat it like an intern. Like, hey, intern, can you go tell me how much we raised at this gala? And then they bring you… they do the work, but then you check it. Is that kind of accurate?

99
00:20:39.870 --> 00:20:44.710
Samuel Chen: Yeah, that is… that is exactly the analogy that I… that I always give.

100
00:20:45.370 --> 00:20:48.260
Neon One Webinars: Perfect. Okay.

101
00:20:48.600 --> 00:21:07.700
Neon One Webinars: So, let's see… here's a question. So, you talk to a lot of nonprofits, and you also have a deep understanding of these tools and how they work. What is something, or maybe even a couple things, that you wish every nonprofit professional knew about these tools before they started using them?

102
00:21:07.880 --> 00:21:32.669
Samuel Chen: Yeah, I think… I think it's so funny how you brought up the intern thing. This is… this is the main thing. I would say the number one thing isn't the technical portion of it, it's so intuitive, right? You're just talking to a chatbot. The main thing is the mindset shift, and it's treating generative AI features or tools like it's an intern. And what I mean is, sometimes when people try these tools and it makes a mistake, they just walk away and say, I tried it

103
00:21:32.670 --> 00:21:36.230
Samuel Chen: got something wrong, it's not ready. But if you…

104
00:21:36.240 --> 00:21:59.479
Samuel Chen: re-shift that mind, and you say, okay, if this was a talented new intern, when it makes its mistake for the first week, you don't fire them and swear off interns forever, right? You put a little effort into learning how to work with them, what they're good at, what needs review, and how to give them direction. And the people getting real value out of AI aren't the ones with the best tools, they're the ones who learn how to use the tool, experiment.

105
00:21:59.480 --> 00:22:11.909
Samuel Chen: And give it a chance. And, like, for example, I've gotten a lot of value out of them since 2023, and oh my goodness, they were some stupid tools when you compared them to the ones today.

106
00:22:11.910 --> 00:22:32.590
Samuel Chen: So, I think it's like the same two skills as working with an intern. You gotta know what to delegate, right? You wouldn't let an intern send something to your biggest donor without reviewing it, but you'd absolutely let them write the first draft. That's the same with AI. There's some tasks you want to delegate, and some you want to review, and some you want to take on, like, full responsibility of.

107
00:22:32.590 --> 00:22:55.759
Samuel Chen: And then second, you want to give it direction, just like giving an intern direction. If you tell an intern, write something for the newsletter. You will probably get something very generic, and it won't be good. But if you tell them something more specific, write 200 words for a monthly donor newsletter, it's about the food pantry expansion, use a warm tone, we use terms like X, Y, and Z, we call our donors.

108
00:22:55.760 --> 00:23:20.660
Samuel Chen: I don't know, pirates because something is pirate-themed, I don't know. But then, it's going to be much more specific. If you get… if you give a vague ask, it's gonna give you a vague output, and the quality is really determined by how much direction you put in. And something that's very encouraging is whatever frustrations you hit today, this is the worst the technology is probably ever gonna be. It gets better

109
00:23:20.660 --> 00:23:24.559
Samuel Chen: every few months, so the… I tried it once, and it wasn't great.

110
00:23:24.640 --> 00:23:27.019
Samuel Chen: That opinion expires every few months, and…

111
00:23:27.650 --> 00:23:31.089
Samuel Chen: To use that analogy, the intern is just learning really, really quickly.

112
00:23:32.120 --> 00:23:45.270
Neon One Webinars: I like that you called that out, because that is absolutely something that I am predisposed to doing. I remember, like you said, the very first time I used a generative tool to repurpose a piece of content, the output I got was bad.

113
00:23:45.270 --> 00:23:54.689
Neon One Webinars: looking back, I did not know how to write a prompt, or how to guide it through creating what I wanted, so of course I didn't get the output. The intern didn't know what I was talking about.

114
00:23:54.690 --> 00:23:59.270
Neon One Webinars: And then the tools have just advanced so much more. So.

115
00:23:59.700 --> 00:24:10.279
Neon One Webinars: I… I don't know. I have internalized that, and will take that from this session, like, just because it was bad ones, this is the worst the tech is gonna be, it's only going to improve from here.

116
00:24:10.400 --> 00:24:15.390
Neon One Webinars: And I appreciate that, especially now that we've talked about these tools as

117
00:24:15.560 --> 00:24:29.630
Neon One Webinars: being… guessing, basically. We can train our interns as much as we can, but we are still the experts. We will always be the experts. We are the ones that need to have the final say, kind of in everything. So I appreciate that.

118
00:24:29.770 --> 00:24:48.829
Neon One Webinars: Samantha, just… I'm going to ask that question at the end, I promise. I don't want you to feel like I'm neglecting you, but I have one more question for Samuel before I ask that one. Okay, so you've been in a lot of different AI spaces. You've worked with banking and all kinds of different industries.

119
00:24:48.890 --> 00:24:59.099
Neon One Webinars: In the AIC… er, in the nonprofit sector specifically, what are some of the biggest opportunities you see for nonprofits to use these tools to make their lives easier?

120
00:24:59.980 --> 00:25:14.370
Samuel Chen: Yeah, I think the biggest opportunity is the very optimistic one, which is it's going to allow teams to do more human-to-human tasks with the same amount of people, allowing the nonprofit to do way more impact.

121
00:25:14.370 --> 00:25:38.890
Samuel Chen: If we think about the gaps that most… most of them live with today, it is sitting on years of rich data, but don't… you don't have a data analyst to mine it. I don't know if this is hitting home for some of you, but it's like having that list of donor touches you know would matter, those thank you calls, those personal notes, those check-ins, but you can't get to them all, because you still have to do those reports, those drafts.

122
00:25:38.890 --> 00:26:03.880
Samuel Chen: admin works that's eating up time. The opportunity of AI is really closing that gap. It's the same people, the same team, but the tedious work shrinks, and the meaningful work grows. Where that shows up concretely, I think, is some of you can start tomorrow with general-purpose tools, exactly what that intern we just talked about and those use cases we talked about. And I'll say this, our goal at Neon One is that soon

123
00:26:03.880 --> 00:26:15.910
Samuel Chen: more and more of that work can happen right inside Neon, where the AI is secure by design, knows your data, learns your organization's voice, so your drafts sound like you, and your answers come from your actual records.

124
00:26:15.910 --> 00:26:18.169
Samuel Chen: I can tease out some…

125
00:26:18.170 --> 00:26:40.539
Samuel Chen: near-term, soon-to-be opportunities that maps to the two types of AI we talked about. On the generative side, the opportunity here, like, right off the get-go, is answers without an analyst. So today, when someone on your team has a question, how did our monthly donor trend this year, that would be a report. And some of these reports are complicated, and we hear about how some of you need to use Excel.

126
00:26:40.540 --> 00:26:49.579
Samuel Chen: And those questions pile up. So, hint, hint, there is something that we're working on, testing in early stages, and you'll hear more about in the future.

127
00:26:49.800 --> 00:27:09.479
Samuel Chen: On the machine learning side, the opportunity is that intuition at scale that we talked about. Some of you probably have that gift officer, right, that knows exactly the right ask. Well, let's have that as a model, and then give everyone that strong intuition, so we can prioritize the right constituent to work out with, and build that human connection.

128
00:27:09.480 --> 00:27:29.240
Samuel Chen: I guess just to wrap it all up, everything you heard today, our testing regimen, how we build the features, we even have subject matter experts in… from the industry try to break these products, and that's a level of investment we're making, because we really believe this is what can give you all an outsized opportunity to drive your mission.

129
00:27:29.930 --> 00:27:48.590
Neon One Webinars: And I think that's incredibly valuable, and I like that you called out specifically that we have industry experts working on this with us. The work that you all do, the appeals that you write, the impact updates that you send, those are so special, and so human, and so important, that we don't want to just give

130
00:27:48.630 --> 00:27:55.380
Neon One Webinars: a generic free ChatGPT version of any of these tools to use. So,

131
00:27:55.460 --> 00:28:03.420
Neon One Webinars: we are… while we work on that internally, that is something that I would tell you all to, like, encourage you. I know there's always the, like, trepidation around

132
00:28:03.520 --> 00:28:20.870
Neon One Webinars: using some of these tools to connect with donors, because you don't want to lose the personal touch. And your personal human touch is absolutely what makes you special. It's what's going to make you efficient, and hopefully these can just kind of speed up the process. It should never, ever, ever replace what you're doing.

133
00:28:20.870 --> 00:28:31.200
Neon One Webinars: I… I know that we have a couple of questions, so I'm gonna come in here. We have one from… where did it go? One from Lorraine.

134
00:28:31.920 --> 00:28:45.540
Neon One Webinars: Boop! So here's a good question. Lorraine said, I struggle a little with writing prompts, particularly asking for data manipulation in Gen. So, I have a two-part question for you, Samuel. One, what

135
00:28:45.850 --> 00:28:49.980
Neon One Webinars: what would you tell someone who is using Neon One?

136
00:28:50.230 --> 00:29:02.989
Neon One Webinars: to do some of these things? What would you tell someone to do with Jen? And then, what would you recommend for anyone, regardless of the tools that they're using, to pursue as they look to get better at writing prompts?

137
00:29:02.990 --> 00:29:13.709
Samuel Chen: Yeah, I guess, just for some context, Lorraine is one of those lucky customers that we have to test out an early alpha of one of the tools.

138
00:29:13.710 --> 00:29:38.059
Samuel Chen: And the first thing I would always say is be as specific as possible, and this could be any AI tool. But first, always start off extremely as specific, and then you'll be able to tease out when it has enough context, and then you can lighten up on the context and prompt that you're providing it. So, I guess with Jen reporting specifically, it would be naming the exact columns that you want to see.

139
00:29:38.060 --> 00:29:48.469
Samuel Chen: what those relationships are between the data, and giving it that context so it provides you the most accurate data as possible. But yeah, Lorraine will be chatting soon.

140
00:29:48.870 --> 00:30:05.950
Neon One Webinars: So I know that writing prompts was something that did not come naturally to me, largely because I didn't know the tools particularly well. What would you… what piece of advice would you give someone like me, who needs to develop the skill of writing prompts?

141
00:30:07.170 --> 00:30:31.699
Samuel Chen: The biggest one really is just practice. The more you do it, the know what you get, but it's being very, very specific. Don't think of it like it is a human being. Still think of it like it is a computer that needs more specific instructions. But it's not like code, right? It's this interesting middle ground where you want to give enough context so it does the thing you want.

142
00:30:31.700 --> 00:30:49.710
Samuel Chen: to do. The good news is, in the future, we will be providing, what some golden standard prompts look like and what you shouldn't do, best practices. This is something that the team is going to work on to help you build that intuition so you don't have to spend, weeks building it up yourself.

143
00:30:49.960 --> 00:31:11.119
Neon One Webinars: Totally. And I'm gonna say two things. One, y'all, if you haven't checked out Erick's message in the chat, I highly recommend that you do. Erick dropped a really wonderful prompting framework into the chat. I have already copied it to my computer's clipboard, and I will be using that in the future. The other thing I would tell you.

144
00:31:11.120 --> 00:31:20.230
Neon One Webinars: One thing that I encounter frequently when I'm talking with nonprofits about any of these tools, whether we're talking about predictive tools or generative tools or whatever, is…

145
00:31:20.390 --> 00:31:24.509
Neon One Webinars: There is an expectation that people feel like they have to be good at this immediately.

146
00:31:24.520 --> 00:31:41.880
Neon One Webinars: Using AI is a skill, and it's a skill that you will develop over time. If you are like me, you don't enjoy developing skills. I just want to know how to do the thing, and I want to do the thing, and I want to do it well. So be patient with yourself. You are building a new muscle, you are developing a new skill as you're using these tools.

147
00:31:41.880 --> 00:31:46.280
Neon One Webinars: And it is okay not to be perfect at it. Immediately, I am talking to myself.

148
00:31:47.370 --> 00:32:02.710
Neon One Webinars: The last question I want to ask, I think, comes from Samantha, and this is a question about tools in Neon One specifically. So, Samantha asked, will the built-in AI help with figuring out operational boundaries? I asked for some clarification.

149
00:32:02.710 --> 00:32:09.840
Neon One Webinars: And they said, if they can't figure out how to make Neon do something I need, troubleshooting a member conflict, building out a workflow, etc.

150
00:32:09.940 --> 00:32:13.460
Neon One Webinars: What do you think, Samuel?

151
00:32:13.460 --> 00:32:21.860
Samuel Chen: Yeah, we actually have a tool built into Neon that helps with that, and it is an AI tool as well. It's our support tool.

152
00:32:21.860 --> 00:32:41.009
Samuel Chen: If you go to the question mark and go to chat support, you can ask questions about how to do things in Neon One, and it will reference our wealth of documents that you no longer have to look through, and then provides you the next steps. For example, like you said, building out workflows and troubleshooting a few things.

153
00:32:41.010 --> 00:32:49.029
Samuel Chen: And hint, hint, we're working towards making this experience all-in-one and a little bit smoother, but that's something in the future.

154
00:32:49.680 --> 00:32:58.109
Neon One Webinars: Perfect, thank you so much. I think those are all of the questions we have. I… oh, I noticed one here.

155
00:32:58.150 --> 00:33:06.170
Neon One Webinars: I am not going to do a good job with your first name, so I'm just going to call out the question about maintaining the database with donor information.

156
00:33:06.170 --> 00:33:31.080
Neon One Webinars: That is something that I don't know will be directly handled by AI tools. However, if you are currently using Neon One, you're going to get an email from me… you're all going to get an email from me. You're all going to get an email from me tomorrow, but for those of you who are currently using Neon One, there is going to be an opportunity at the bottom of that email for you to check out, our events calendar, and we have open offices

157
00:33:31.080 --> 00:33:47.369
Neon One Webinars: hours. So, if you have questions about maintaining your database, collecting donor information, making sure that information is accurate, you don't have duplicates, etc, that is something that you don't have to wait for a tool to come out. We can help you with that now.

158
00:33:47.690 --> 00:33:56.259
Neon One Webinars: So, do keep an eye out on your inbox. You're gonna get an email from me tomorrow with a link to this recording, and some other resources.

159
00:33:56.260 --> 00:34:15.219
Neon One Webinars: If you are a Neon One customer, that email will include a link you can use to find open office hours, product-specific webinars, support documentation, anything you need. So you don't have to wait for gen reporting, you don't have to wait for some of the other tools that we've been teasing. We can help you with that right now, so do check that out.

160
00:34:15.679 --> 00:34:31.040
Neon One Webinars: That said, it has been a delight talking to all of you. Thank you, Samuel, for explaining these things. I know understanding how tools work is the first step to using them well, and I so appreciate you sharing this with us.

161
00:34:31.040 --> 00:34:37.120
Neon One Webinars: Everyone, do keep an eye out on your inbox for an email from me tomorrow.

162
00:34:37.190 --> 00:34:45.809
Neon One Webinars: Thank you for spending some of your Wednesday with us. I know how valuable your time is, and I am so happy that we got to spend some time together.

163
00:34:46.020 --> 00:34:48.680
Neon One Webinars: Have a great Wednesday! Goodbye, everybody!