Accelerating Prospect Development: Crafting a Custom AI Chatbot
Written by: Chris Copsey

Are you tired of hearing about AI this and Machine Learning that, and want to get your hands dirty with a possible solution that can help your fundraising shop AND move your office to the cutting edge of the AI bandwagon? Then look no further than the world of Large Language Models, or LLMs.

Surely, most of us have either used ChatGPT or heard enough about it from your one friend who goes on and on about Crypto currency to be skeptical. In short, Large Language Models take large swaths of data and can act at your own personal help desk for any questions you might have. It has a range of uses in the world, but especially for Prospect Research and fundraising.

Are Development Officers complaining about a lengthy research document? Have an LLM summarize it in a few paragraphs! Having trouble writing an annual letter appeal? Have an LLM write a letter in the “voice” of your Dean! Does the thought of receiving the same Prospect Management question over and over again make you break out into hives? Create an LLM Chatbot that can help answer these questions instead.

This last example will be the focus of this article. I will be giving you some instructions on how to make your own AI Chatbot that can help ease the burden of email fatigue and be a great line on a resume, which you will be needing since, you know, AI is going to be taking all of our jobs, right? (Just kidding).

Background

The first thing to know is that this year, Microsoft launched their Copilot software, which is essentially a chatbot integration that currently lives in Microsoft browsers. You might be familiar with this logo:

Earlier this year, they launched the ability for users to build their own “Copilots”, as they call them, which are just bespoke AI Chatbots.

To get started, cruise on over to www.copilotstudio.microsoft.com.

**Note: This is NOT an endorsement or paid plug for Microsoft. It is merely the software that I chose to use. Be vigilant about any software that you use. If you have an IT department, contact them about any software that is secure for your information.

Instructions

Once on the site, you are given the option to either buy a license or have a free trial. I recommend starting the 60-day trial, which can still offer you some usage in that time. If your work or institution has a subscription to Microsoft 365, you might be able to just log in with your username and password. My institution, UNC Chapel Hill, has a contract with Microsoft for an enterprise version of Copilot, which allows internal usage of data that is not recorded on the back end by Copilot (as opposed to ChatGPT which does save your data). Again, check with your institution to see if you have a similar setup.

Once inside the Copilot site, click “New copilot”. Next, a box will appear asking to name your Copilot and to enter a website that will serve as a reference for the Copilot. Since I am working with a Microsoft 365 subscription, I was actually able to paste a link to my Development Office’s Devnet, where many of our Prospect Management policies are kept. Once you have linked up to a site, hit create at the bottom, and you’re ready to start Copiloting!

Once you have created your Copilot, you will be able to begin feeding it documents to give it a “brain”. On the left-hand side, there is a list of options, including a “Settings” dropdown. Click that, and then navigate down to “Generative AI”. Here, under the “Upload a document” area, you will be able to upload PDF documents that the Copilot will be able to reference.

In my instance, since I wanted to create a Prospect Management chatbot, I uploaded 20 documents relating to Prospect Management.

Once you have uploaded the documents, and are ready to start testing your Copilot, utilize the internal chatbot box that is to the left of where you uploaded the documents. Ask it questions! This is the great part of Generative AI and Chatbots- quizzing it and seeing if the answers line up with your internal policies.

If you are happy with the responses that your new chatbot buddy has given you, it is time to publish. Again, navigate to the left side and select “Publish”. There will be a button in the middle of the screen to press and then publish.

Finally, as the icing on the chatbot cake, you want people to be able to access and use your chatbot, right? Under the Publish button you will notice an area saying “Optimize your copilot”:

Click the “Configure channels” button to take you to a list of places you can push your chatbot into:

Now, since this is a trial account, you will not be able to publish to all of them. But the first option, as a Copilot, will give your chatbot its own tiny website and link that you will be able to share with others. Myself, I used the second option, which was to push it out to Microsoft Teams, which allows myself and others to chat with it like we would any other person on Teams.

After the initial build, you can easily add more documents to the chatbot that you have created. There is a maximum number to add, but I have added some PDFs that are over 50 pages long. Almost all of these documents were either how-to’s, FAQs, or policy documents that were spread out over an entire website. Collecting them all and loading them into the chatbot has allowed me to synthesize them all and access their collective information all at once, rather than going through dozens of individual documents.

Internal reaction, after the initial “Wow, what is this?” has been stellar. There were a few attempts to “break” the bot by asking it confusing questions, but it held up very well! Most of the rollout and testing among staff has been around calibration and making sure that the answers are actually correct and not hallucinations! Formal launch of the product writ-large is expected at the beginning of the new fiscal year.

Again, this is just one of the interesting things that AI can be used to do as it related to Prospect Development. Once you create your chatbot, poke around the Copilot site and try to improve it! Happy Building!

Chris Copsey is the UNC Health Foundation’s Assistant Vice President of Prospect Development. Since 2019, Chris has provided the Health Foundation with data and analytical support, mainly supporting grateful patient identification, cultivation, and solicitation activities. He assists with the strategic coordination of critical data and information sharing between the donor and alumni database, EPIC, and other resources as necessary, and serves as a liaison between the Health Foundation and University of North Carolina Development office as it relates to data management and information sharing. He is also very active in Apra, serving currently as a board member of Apra Carolinas and previously helping to organize their one-day conference for philanthropic data professionals called Data Science Now.

Chris graduated from Ball State University in Muncie, Indiana, in 2008 with a dual degree in Political Science and History. He enjoys reading books on American History, playing golf (poorly), and plotting out the best ways to grow vegetables in his yard- a work in progress! Chris lives in Snow Camp, North Carolina, with his wife Trinh, and son, Theodore.