The Solopreneur’s Toolkit: Custom GPT Solutions with Nancy Bain at AIE 2024

Hi, I am Nancy, and I'm the CEO and founder of Supernova Media. I have 20 years experience as a bootstrapped entrepreneur and digital marketer, and today I'm excited to share with you one of the tools in my digital marketing toolkit. We'll be looking at, uh, custom GPTs, their benefits, um, how to manage their builds in notion and, uh, the chat GPT builder interface.

So we've got a lot to cover. So let's get at it. What is a custom GPT?

Well, a custom GPT is essentially a customized version of Chat g pt. They are fine tuned with a singular, uh, purpose in mind, and they're designed to help streamline our workflow and inform our decision making. So what, what are the use cases of custom GPTs?

Well, there are several. Um, there is the GPT store, and I encourage you to visit it on, uh, chat GPT, but for, uh, digital marketing purposes, um, we can use them for data analysis, content creation, SEO, personalization, and our marketing campaigns and so much more. Um, what are the pros of GPT in digital marketing?

Well, they help with operational efficiency, enhance decision making, um, and improve customer experience. Uh, custom GPTs boost performance and, and in and ROI in our campaigns, and they help us save time for key tasks, uh, via AI actions In Zapier, you can connect to over 6,000 AppSec and, uh, perform automations that can help with lead generation, customer purchase to email, marketing list, post purchase follow up, all kinds of, uh, opportunities for operate, uh, automations there. Um, managing expectations is key with the custom GPTs because they're not silver bullets and each is a specialist and not a generalist.

Um, in fact, I took the opportunity on LinkedIn to poll my colleagues and ask their, um, experience, uh, with the custom GPT and how well they follow instructions. And you can see here that, um, um, there's room for improvement and, uh, so they don't always follow instructions perfectly. Um, this can often be managed, um, with a simple prompt to, uh, get them, um, back on track.

But if you know this heading into it, it can, uh, save a lot of head scratching, um, later. So what are the cons then of, uh, GPTs? Um, well, they're designed to solely exist in the chat GPT interface and hosted by, uh, ai.

So you're not going to be able to host them on your own servers, um, because they are single use case. You're going to need more than one and likely a lot more than one, which is why we, uh, manage our builds in notion. And, uh, we'll take a look at that, um, very shortly.

Um, like any ai, they're prone to, uh, hallucinations and, uh, inaccuracies, and they're going to need, uh, human oversight. So these things don't happen. Um, while certain metrics will be easy to track, um, others, um, will be harder to track.

Um, so definitely some non-quantifiable KPIs there. So this is the custom journey and, uh, custom GPT journey, and it, uh, represents my framework and how I manage my builds, um, over in Notion. And, um, it'll make more sense once we get over to, um, notion.

And so let's talk about that. You know, why I, I begin all of my builds for my custom GPTs in Notion. And, um, and I encourage you to do the same.

We consider Notion to be our command center. Um, we do that because we have more than one custom GPT. In fact, we have 21 custom GPTs, and so you're going to need a one stop organizational powerhouse.

And for us, that's notion. So we love it for, uh, centralized documentation. Um, this is where we manage, uh, performance tracking and Notion is a great, um, tool for, um, collaboration.

So let's go take a look at that notion build. So here we are in Notion and, um, it looks kind complicated, but it isn't. Um, first off, we've got our strategy and our use case, and for each of our, um, custom GPT where there are knowledge base files associated with it, um, sometimes when you build a custom GPT, you just let it think by itself, and you don't upload any knowledge based files.

And those custom gpt work well, um, too. Um, but this particular GPT, um, manages my, uh, Gumroad account, and it helps me to, uh, write my description SEO optimized descriptions and summaries based on the, um, my tutorials that I upload, and it does it using my writing style. So, um, for each of our custom GPTs, um, where there are knowledge to face files associated with, we define a problem statement, um, the project overview and the GPT requirements.

And as you can see here in this command center, you can assign these two different members, um, within your team, assign due dates. And the best part for me is that they are all linked, um, to, uh, the Google Drive and the corresponding folder. Now, this is crucial because there's a lot of iteration when it comes to, um, building your GPTs, and, um, I like to have my documents, um, organized because there is going to be some version control going on.

So, um, with your custom for the problem statement, this is a simple, what is the issue, what is the solution, um, what is the impact? And so these are the files that represent the planning stages, um, of our custom, uh, GPT. Now, these files are the specifications and instructions.

These are actually uploaded to the knowledge base of our custom GPT gum road guide. And so these are structured differently. These files are structured differently.

They, uh, con uh, contained markdown, um, and, uh, header text. And, uh, this we have found in our experience, uh, works far better, uh, for your knowledge based files. We've experimented, um, with PDFs, um, we've experienced with different lengths of our knowledge based files, and we've found, uh, that, uh, docs using markdown, um, um, help the model perform more consistently.

Um, certainly keeping all of your files, um, a con consistent type, uh, if you can, a consistent, uh, uh, type of document or file type. Um, so, and this is our training data, um, for the, uh, custom GPT. And, um, so the, these represent the different bundles, um, that I have available over on, uh, gum road, um, for, uh, sale.

And this, uh, you'll note this file here. This is my writing style. So this writing style was developed by, uh, training, uh, an assistant on 600 pages, um, of my voice.

And I've narrowed my writing style down to a single document because I don't recommend, um, uploading 600 pages of your voice as a knowledge based file, uh, due to context, uh, uh, concerns. Um, so here's my gum road call to action. Each of my, you know, the description includes a call to action.

And so every key stroke that I can save, um, I've saved that, uh, uh, document to the, uh, knowledge base, um, as well. So, um, for model training and, and updates, I alluded to this earlier when I referenced version control. Um, I've often, um, in, in iterating my custom GPTs reverted back, uh, to an earlier model, um, and a model, you know, and, and specifications.

So, um, this is where we keep track of, um, uh, iterating, uh, the model so that we can easily re revert back. Um, these files are, um, for, to track, uh, to help me track the performance of Gumroad Guide, uh, or, uh, to help me track the performance of my products rather on, uh, Gumroad. And, um, and then we have our automations.

And so we've got, um, automations, uh, every sub subscriber that, um, that, uh, to Gumroad, um, will automatically be added to my mailing list. And we manage those, um, with Zapier integrations. So this is a pretty comprehensive, um, build.

Um, but the beauty of this build is that because all of our files are centralized, it means they're portable. And, uh, we've been recently testing and using, um, Gemini, and, uh, I'm excited to take these files into their AI studio and, uh, and, and compare, uh, solutions. So, um, narrow down to the, um, custom GPT builder interface.

Now, this is my, uh, gum road guide, and let's go in and have a look at the backend. So, um, if you're going to build a custom GPT that does not have knowledge base files, this is where you're going to do your work. I have, I never work in here.

Um, I, I don't use the create tab. I haven't had good results with the create tab. Um, I go immediately to the configure tab.

And, um, so let's just have a look at what's available back here. Here's, uh, some instructions. I've also up upload, uploaded a document of instructions.

Um, and, uh, we also have, uh, our conversation starters. These are the knowledge base files, um, that were linked to, um, over in the Notion document. And I've given, um, full capabilities, uh, to the guide.

You'll note that in my conversation starters, I call them leader prompts. I've, I've used things like refer to writing style, um, in the knowledge based files because one of the, uh, biggest, um, frustrations for me is, um, that they don't consistently refer to the knowledge based files. So I've, I've found that including in the conversation starter to refer, uh, to the knowledge-based files can oftentimes, uh, save, uh, a prompt in correcting it later.

So, um, that is, um, custom GPT in a nutshell. Um, we talked about the benefits. We've talked about their limitations and how to manage them in, uh, notion and the, uh, custom GPT builder interface.

Um, I like to invite you to subscribe to Tech Tactics. Uh, we meet every other Sunday at 11:00 AM for tea. Um, I host, uh, live, uh, support, uh, sessions via Slack.

And, uh, every other week you'll receive a, uh, an email newsletter, which is in fact a lesson. And, uh, we go through it together. So, uh, thank you for your time today.

I hope you enjoyed this overview on custom and happy innovating.