Product Information and ChatGPT - Part 1
A practical guide on using AI for product information management
This is part 1 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.
It’s distilled from more than 100 hours of consultancy with online retailers, including in-depth discussions of realistic catalog management use cases and building various software tools and integrations.
Product Information and ChatGPT Series:
Part 1: What can ChatGPT do for my business? (this article 🙌)
Part 2: What’s the best way to adopt and integrate it? (read here)
Part 3: Are there any drawbacks or limitations to this technology? (read here)
Why I’m Writing This
I’m a software architect and I consult with clients on integrating and building software solutions for their businesses. 👨💻
During the past year (2023), there was one thing that dominated nearly all conversations I had with clients in the retail industry …
ChatGPT.
As many people in retail know, most Product Information Management (PIM) is about the manipulation of text data such as product titles, descriptions, attributes and tags, and that requires tedious and error-prone manual work. 😢
So naturally, the release of a powerful text comprehension and generation technology like ChatGPT made everyone excited about new possibilities. ✨
Clients taking an interest were:
Online retail shops
Software startups creating solutions for these retailers
Content providers and consultancies that offer them services
In my conversations with these clients, there were three main questions:
What can ChatGPT do for my business?
What’s the best way to adopt and integrate it?
Are there any drawbacks or limitations to this technology?
For clarity, when I say “ChatGPT” here, I’m actually referring to its underlying text processing technology which is Large Language Models (LLMs). Although LLMs were made popular as chat applications, they can take various other forms including extensions to existing software.
I’ll continue to use the term “ChatGPT” throughout this article for simplicity, since many of the people I spoke to were unfamiliar with the distinction. 😬
With that out of the way, let’s now turn to the first question on the list, which is the topic of this article.
What Can ChatGPT Do for My Business?
Let’s get one thing straight — ChatGPT is a breakthrough for PIM. 🧙
And by this I don’t mean that I feel excited about it or have bought into the hype and media frenzy.
But rather that early reports from the sensible application of ChatGPT to selected areas of PIM have demonstrated cost, quality and productivity improvements in the order of 10x, surpassing the impact of any single technological development since the PIM acronym first came into use. 🎯
The software market has been saturated with mindless attempts to apply ChatGPT in every domain, with mixed results, and the PIM market is no exception. To avoid contributing to this pandemic, I’d like to stress that real impact is achieved only when ChatGPT is applied to the right problems in product information management.
So what problems are these?
After a year of extensive discussions and hands-on prototyping with real catalogs, here are my top five picks.
1. Parsing Unstructured Text
ChatGPT can parse text data from any source exceptionally well. This includes supplier data in various formats (including spreadsheets and PDFs, which can all be dumped into text) and data scraped from online sources.
This means that a big pile of text and files go in, and structured information come out, so no more manual data entry. 🙌
While this may not be the most exciting application of ChatGPT, it’s definitely the most impactful, cutting hours of work down to minutes for many businesses. ⚡
2. Checking for Consistency
ChatGPT can apply sophisticated consistency checks that are difficult or impossible to express as Excel formulas or equivalent classic checks in most PIM tools.
An example is to cross-reference two fields to make sure they don’t contain contradictions, say:
Product Description: "This lightweight, portable device features a stunning 10-inch display and 256GB of internal storage, perfect for all your entertainment and productivity needs."and
Product Specifications:
- Screen Size: 10 inches
- Storage Capacity: 128GB
- Product Category: Tablet3. Content Creation
ChatGPT can create product descriptions, social media posts and other types of marketing copy from just a bit of product information.
This is perhaps the most straightforward application of ChatGPT.
But it’s also the one thing that ends up disappointing clients in practice. 😕
While it’s true that ChatGPT can generate content with the push of a button, this usually suffers from several issues, including:
Brand voice guidelines are difficult to enforce.
Language is dull, repetitive and full of clichés.
Output includes false information (the infamous “hallucination” problem).
Luckily, a new generation of product content software tools have made great progress on solving these issues. I’ll discuss this in depth in Parts 2 and 3 of this series. 🙌
4. Catalog Syndication
One of the main difficulties with multi-channel commerce is setting up the tools and flows to push catalog changes to channels (syndication).
Configuring a single channel can involve mapping hundreds of fields from one schema to another. To make it worse, many of these mappings are not 1-to-1, meaning that some fields need to be combined or broken down, usually in non-trivial ways.
For example, a merchant may have a wireless router with:
Connectivity: 802.11ac 750Mbps 2.4GHz/5GHzwhich a channel expects to be exported as:
Speed: 750 Mbps
Frequency: Dual-Band
Wi-Fi Standard: Wi-Fi 5Setting up such a transformation can be a real pain. It requires understanding the product category’s technical specs, translating them into text processing rules then spending hours to ensure every pesky edge case is covered. 😖
ChatGPT makes all of this go away! 🙌
The above example is actually from a real product which was processed by ChatGPT. It was given only the input data and the output channel field names!
5. Extracting Information from Product Images
ChatGPT can understand images, and this is huge! 🤯
Some suppliers do not provide any details on their products besides their images.
Retailers can either invest a lot of effort to research or figure out these details, or just post sketchy product pages (title, image and price) on their stores and suffer terrible conversion rates.
ChatGPT offers a huge opportunity for retailers facing this issue, as it can extract detailed product information from images, and it’s very reliable.
Here’s my favorite example …
Can you guess what this product is?
What about its material, features, applications, audience and safety instructions?
ChatGPT can guess all these with 100% accuracy. See its answer here.
Up Next
I hope this article provided some clarity for those looking for concrete examples on what ChatGPT can do for product information management, away from all the hype and cryptic AI marketing that’s plaguing the web these days.
Stay tuned for Part 2, coming up next Tuesday, where I’ll discuss how to add the above capabilities to your product management system.
Whether your store runs on Shopify or any other e-commerce platform, and regardless of how you store your product information (a PIM or just some spreadsheets), Part 2 will have a practical solution that fits you. 🎯
Update: Part 2 is now published, read it here.
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. 😉