Product Information and ChatGPT - Part 3
Product Information and ChatGPT - Part 3
Things that ChatGPT will NOT do for you
This is Part 3 of a series of articles in which I will cover everything you need to know about applying ChatGPT to manage product information in online retail.
And to remind you: I’m using “ChatGPT” in this series as an accessible term for Large Language Models (LLMs), the underlying technology powering the famous chat app, as well as most of the AI buzz in the last few years. 😉
Product Information and ChatGPT Series:
Part 1: What can ChatGPT do for my business? (read it here)
Part 2: What’s the best way to adopt and integrate it? (read it here)
Part 3: Are there any drawbacks or limitations to this technology? (this article 🙌)
Client Disappointments
I mentioned in previous articles that my conversations with retail clients the past year had many recurring themes.
Clients were interested in capabilities and the best ways to apply them, which I covered in Part 1 and Part 2, respectively.
A third recurrent theme is that clients had similar disappointments when they learned of a few things that ChatGPT cannot do. 😒
So, what exactly were these things? Let’s have a look …
Things ChatGPT Cannot Do
1. Read Human Minds
Many people expect ChatGPT to pick up on some “obvious” things without being told directly. 🤖💡
Take for example a women's fashion boutique. Say they offer a waterproof set that includes a jacket and a matching bag, which are also sold individually.
We ask ChatGPT to describe the features of the bag. Since the bag is part of the set, it should know that it’s waterproof too, right?
To the surprise of most clients, the answer is no. 😮
The reason is simple — it’s very common for ChatGPT to only consider the information of the specific product it’s working with (so the bag, not the set).
Software developers find such quirks very intuitive and will often nod their heads in agreement, since behavior like this often corresponds to the most straight-forward code implementation. 👨💻
But the average user finds it confusing and disappointing. 😒
More generally, the average user has high expectations of ChatGPT. They expect it to have a holistic understanding of the entire product line, the business's nuances and even their personal preferences and thoughts.
Many software developers will refer to this as “reading minds” because it lacks clarity.
But “reading minds” is actually not completely impossible, and I’m going to suggest a way to pull it off at the end of this article. 😂
2. Learn Over Time
Another common expectation is that ChatGPT has the ability to learn from its interactions over time. Users are often surprised when they need to repeat instructions they’ve already provided.
For example, if a user corrects ChatGPT after it generates a poor product description, they’d expect it to apply that feedback going forward.
But ChatGPT would repeat the exact same mistake on the next product. 😅
The reason is again rooted in how the software built on top of ChatGPT is designed. Each interaction is typically treated as an isolated event, so ChatGPT would not retain or build upon past interactions.
This limitation is very frustrating to users expecting a more adaptive experience.
Is there a solution? You bet — just keep reading. 😉
3. Provide Esoteric Knowledge
Finally, while ChatGPT contains a wealth of information within it, it doesn’t contain databases of niche or specialized information, so called "esoteric knowledge”.
For example, it cannot provide information about products from identifiers such as ASIN (Amazon Standard Identification Number).
It also cannot access real-time data, because everything it knows is limited to its training data, which is an extremely large — but nevertheless static — snapshot of Internet data taken at a certain point in time.
This too confuses the average user. 😵
For example, an online retail shop owner I spoke to received a list of scanned product bar codes from one of his suppliers and was disappointed when he learned that ChatGPT cannot convert this to product information. 😥
Overcoming these Limitations
Believe it or not but there’s actually a way to overcome all these limitations! 💡
The trick is to realize that they don't stem from ChatGPT itself, but rather from the way it's typically integrated into tools and platforms.
So, when an AI tool is designed the right way, it can go beyond these limitations.
Now what sort of tool might that be? 🤔
It’s a category of tools I touched on in Part 2, which I described as “New Generation PIMs for AI”.
They’re tools built from the grounds up for AI processing, with the necessary user experiences and foundational design work to leverage the full capabilities of ChatGPT.
New Generation PIMs
A New Gen PIM is an AI tool that’s designed specifically for PIM applications.
Its specialization allows it to collect and factor in additional information compared to classic AI tools, offering unique features and solutions. ✨
For example, a New Gen PIM can analyze your catalog in more depth, to better understand your product range. It can also collect information that is specific about your business, and apply these during information processing and content generation.
Imagine a PIM asking you to outline your target audience and then take this into consideration when creating product categories, taxonomies and attribute sets.
When you think about it, specialization and contextual awareness address the three limitations mentioned earlier:
👉 While it’s impossible to literally read human minds, a New Gen PIM comes close. It can prompt you for information that is relevant to your business and take a holistic view of your product data.
👉 For the same reason, it’s also capable of learning over time. It knows what user inputs and feedback are relevant to subsequent tasks, allowing it to avoid past mistakes and build upon its learnings.
👉 Lastly, it can integrate with specialized external databases, allowing it to generate product information or content based on esoteric knowledge.
Examples
Here are some examples of New Gen PIM tools currently available in the market (copied over from Part 2):
I recommend exploring these tools and what they offer. Even if you’re not looking to use them right away, it’s useful to stay informed about developments in the PIM tooling market. 🤖
Conclusion
This is the last article in this series, which I hope will serve as a useful practical guide for anyone who’s interested in using ChatGPT to manage product information! 😃
ChatGPT is a very exciting technology that’s causing wide disruption to many industries, including retail product data management.
As models become more powerful and more tools emerge, keeping up with the latest developments will be increasingly important.
I’ll be covering any interesting updates in future articles. Don’t forget to subscribe or follow me to stay in the loop! 🙌
If you like this content, please subscribe to receive new articles straight into your inbox. 📩
Follow me on LinkedIn for more on #retail, #pim, #digitaltransformation and #software. 😉