OpenAI and Amazon Web Services just released no-code tools for creating custom AI bots that can automate and optimize many e-commerce and retail tasks.
OpenAI announcement on November 6, 2023, it would roll out custom versions of ChatGPT and soon open a GPT store where developers could release pre-trained generative transformers (i.e. “GPT”) intended for specific tasks.
Then, on November 16, AWS released PartyRock, its experimental builder of low-code AI applications running on Amazon Bedock.
![Screenshot of the PartyRock webpage. Screenshot of the PartyRock webpage.](https://www.practicalecommerce.com/wp-content/uploads/2023/11/111723-partyrock-570x380.jpg)
PartyRock is Amazon’s low-code AI app maker, currently in experimental mode.
Almost no code
For marketers, the main takeaway from these announcements is that generative AI can be customized, at least to some extent, without a developer. The tools are about as easy to use as Zapier or your favorite spreadsheet, as long as you’re comfortable with formulas and pivot tables.
These announcements represent a shift in AI applications since customization has until now required advanced programming capabilities and access to APIs (still necessary for high-volume, niche applications). Already, no-code builders are enabling amazing AI robots using “instructions” and “knowledge.” Only “actions”, connecting custom GPTs to data, are low-code
Using OpenAI’s generator, let’s create a custom GPT and discuss some possibilities.
Ad copy
Here is our scenario. Imagine you are a marketing manager for a direct-to-consumer nutritional supplement brand. Your business purchased host-read ads on 50 podcasts. The ads were successful but required a lot of copy.
Let’s create a GPT for this.
Paid OpenAI accounts now include an “Explore” link at the top left of the ChatGPT page. This is where custom GPTs will reside.
![Screenshot of the ChatGPT menu showing the Explore tab. Screenshot of the ChatGPT menu showing the Explore tab.](https://www.practicalecommerce.com/wp-content/uploads/2023/11/111723-cpt-explore-570x380.png)
Paid ChatGPT users can access the GPT generator via the Explore link.
Opening the Explore link will display “Create GPT”, the option to create a custom (semi-custom, actually) version.
Unsurprisingly, OpenAI provided a chat-based tool to start with.
The robot starts:
Hi! I will help you create a new GPT. You can say something like “create a creative that helps generate visuals for new products” or “create a software engineer that helps me format my code.” What would you like to do?
![Screenshot of GPT generator via chat. Screenshot of GPT generator via chat.](https://www.practicalecommerce.com/wp-content/uploads/2023/11/111723-initial-builder-questions-570x380.jpg)
Enter GPT instructions manually under the Configure tab or by chatting with the GPT builder. Click on the image to enlarge.
Answering the initial queries kicks off the first round of questions. Meanwhile, a preview of the current custom GPT appears on the right side of the page. The preview shows how the new chatbot works.
![Screenshot of a ChatGPT preview. Screenshot of a ChatGPT preview.](https://www.practicalecommerce.com/wp-content/uploads/2023/11/111723-cpt-out-put-570x488.jpg)
Test the output via GPT preview while you configure the custom version. Click on the image to enlarge.
This back-and-forth and a subsequent field in the configured GPT constitute what OpenAI calls the “instructions” part of the process. Spend time with these instructions, and don’t be afraid to ask him to change. You can exit your GPT and update it later.
Under the “Configure” tab at the top of the generator are the GPT-specific instructions. You can modify them directly.
![Screenshot of "Master of advertising script" in the Configure tab. Screenshot of "Master of advertising script" in the Configure tab.](https://www.practicalecommerce.com/wp-content/uploads/2023/11/117123-configure-screen-570x580.jpg)
Assemble or modify the GPT under the “Configure” tab. Click on the image to enlarge.
One of the best features is the ability to add “knowledge” to GPT. For example, we can download the brand’s writing style guide, although the knowledge can be more or less anything.
![Screenshot of the Knowledge page. Screenshot of the Knowledge page.](https://www.practicalecommerce.com/wp-content/uploads/2023/11/111723-knowledge-570x303.jpg)
GPT can include brand-specific knowledge. Click on the image to enlarge.
Finally, the GPT can have coded “actions”. These are similar to a ChatGPT plugin connecting to live data.
![Screenshot of "Edit actions" page. Screenshot of "Edit actions" page.](https://www.practicalecommerce.com/wp-content/uploads/2023/11/111823-action-form-570x533.jpg)
Actions allow a GPT to connect to external APIs such as Zapier. VSlick the image to enlarge it.
Actions make custom generators low-code rather than no-code, because you work with data and APIs to extend the capabilities of GPT. For example, we could connect our podcast editor to an upcoming sales endpoint and a calendar displaying a schedule of podcast ad placements. This information could inform the (human) writer which promotions to mention in a given host-read ad.
Many possibilities
Podcast ads may or may not help your business. Perhaps instead you need a GPT to analyze post-Christmas promotions to know how many units to sell to make a discount worthwhile.
Or what about a GPT that helped identify seasonal peaks in product demand? Would a retailer want to know how the weather in Michigan impacted their online sales?
We will learn over the next few years how AI and machine learning are happening. At the moment, the possibilities seem numerous.