Photo: Getty Images/Alen-D
Will ChatGPT supercharge chatbot conversations?
Through a special arrangement, presented here for discussion is an excerpt of a current article from the blog of Aptos.
Despite a few limitations, ChatGPT may massively impact how people search for, evaluate and purchase products in the future.
Here are a few thoughts on how OpenAI’s new chatbot could be used in retail:
- Customer service: ChatGPT can be integrated into a company’s website or mobile app to provide customers with instant and personalized assistance. The model can answer customer inquiries, provide product recommendations and assist with ordering and returns.
- Product search: ChatGPT can be used to create a natural language processing (NLP) based search function for a company’s website or mobile app. This allows customers to search for products using natural language, such as “I’m looking for a blue dress in size medium” rather than using specific keywords.
- Personalized product recommendations: ChatGPT can be used to analyze data and make personalized product recommendations. The model can analyze purchase history, browsing behavior and customer preferences to suggest products that are most likely to to be of interest to the customer.
- Up-selling and cross-selling: ChatGPT can be used to suggest complementary products or services to customers during their shopping experience, increasing the chances of up-selling and cross-selling.
- Chatbots: ChatGPT can be used to create a chatbot for customer service, which can be integrated into social media platforms, websites and mobile app to provide assistance to customers 24/7.
Most impressive, this succinct list was gathered, collated and generated, word for word, by ChatGPT — a machine.
The potential for equally “thoughtful” responses to traditional consumer queries — i.e., “What is the best multi-purpose tool to use for building a wood deck?” or “What are the best shoes to buy for a person running a marathon?” — would mark a huge leap over the stilted conversation flow of today’s typical service bot.
Think of the conversion-driving power of a custom-trained bot serving up highly informed, thoughtful search results, delivered equally well in both written and spoken form. (I’m talking to you, Alexa!)
Think of the conversion-driving power of delivering customized, intuitive and informative answers to shoppers’ questions, consolidated into human-language responses that don’t require you to click through to five different pages to find the complete answer.
If able to overcome some limitations, including access to real-time web-based information and bias risks, ChatGPT could revolutionize product search results.
Discussion Questions
DISCUSSION QUESTIONS: What do you think of the potential for ChatGPT to elevate product search and chatbot conversations? What risks or limitations do you see?
Poll
Loading …BrainTrust
The one mere fact that ChatGPT can offer NLP so consumers can ask questions on their own terms, in their own style will be a game changer. I had the opportunity about four years ago to work with a startup that had developed some NLP software for B2B internal document search and it was simply amazing. Speed, clarity, and effectiveness are just a few of the powerful benefits. I’d say we’re at the tip of positive disruptiveness with ChatGPT.
I have been impressed with ChatGPT — especially the quality of responses it gives which also seem very natural. If this can be applied to retailer chatbots and service functions then I can see the experience being elevated. Search would also be transformed as you could simply ask it to find things within certain parameters (price ranges, sizes, etc.) and it would come up with a curated list of options: almost like having your own personal shopper. There are a lot of very exciting applications here!
I agree with your thoughts on how to expand the use cases within site search, Neal. For my original article, I tried somewhat generic searches to see how the demo version would respond. When I asked for recommendations of the best running shoes to use for running a marathon (without regard for color or size), here is an excerpt from the response:
“When it comes to buying shoes for running a marathon, it’s important to consider factors such as fit, support, and cushioning. Some popular shoe brands for marathon runners include Brooks, ASICS, Nike, and New Balance.
It’s also recommended to try on several different styles and brands to find the one that feels the most comfortable for you. Additionally, it’s important to make sure the shoes you choose are appropriate for your foot type and running style.”
Imagine if the same question were asked of a licensed and trained version of ChatGPT, with access to size/color availability, per your comment.
Editor’s note: One of my duties each morning is running through the stories for a final copy edit. When I got to the bullet points above, I found the writing flat and repetitive. I thought, “What’s with Dave? He writes better than this.” I swapped out some language to put a little life into the copy, fixed one glaring typo and I removed the fifth bullet point because it seemed redundant. Then I saw Dave’s admission that it was written by ChatGPT, so I put all the original copy back in.
So yes, sure, ChatGPT, recommend a new flannel shirt for me, but don’t apply for our editorial staff any time soon.
OK, Rick, I had no idea of this back story – that is interesting…and, my editor here at Aptos said the exact same thing when I submitted the original post to her for review…perhaps telling about the state of the machine’s copywriting skills, thus far!
New technologies or new twists on existing technologies are exciting but we must separate the hype from the underlying tech. Crypto anyone? AI is the flavor of the month but in most cases, that’s all it is. Even the mighty ChatGPT has flaws — mainly, it still can’t think. However it has supercharged the ability to collect and present data and when used properly, it can deliver trustworthy results. All of us know how frustrating chatbots can be, because they’re not us (though many of us have frustrations with humans too, so don’t blame the bots!). The limitations remain language, understanding and actual intelligence but ChatGPT, buzzword or not, has supercharged the ability to analyze data quickly and mostly correctly. The potential to elevate product search and chatbot conversations is there and maybe ChatGPT and its underlying tech will generate enough interest to attract more investment and to get us closer to true artificial intelligence (as scary as that thought might be, right HAL?).
ChatGPT is probably the most enthusiastically watched tech in the news today, and the interest and expectations are well deserved. I cannot remember a new product release that has created this level of activity, prognostication, and experimentation. That said, we are still probably one or two generations of development away from the pervasive application of augmented intelligence technology. The general engine, APIs, use case libraries, and marketplace familiarity are still in the formative stage, with a lot of experimentation going on. But at the rate of development, those two generations of development could realistically occur within the next 18 months.
Using ChatGPT will make product search and chatbots smarter, faster and more personalized for a superior customer experience. ChatGPT frees up retail workers from drudgery, so they can focus on serving shoppers with creativity, personality and emotional connection. Humans get to be more human.
Potential risks include frustrating customers if ChatGPT uses outdated or inaccurate information. Also, bias affects data collection and analysis.
Agree 100% on the risks, Lisa – we will still need humans to ensure the bot stays well “trained” (their word, not mine) and protected against bias, but IMHO as a site search tool, those risks seem entirely manageable.
ChatGPT may need an additional $10 billion to help it scale, get up to date and stay that way, and get even smarter, but holy cow — it’s pretty impressive right now. Much more interesting than the AI-generated selfies.
Microsoft recently invested exactly that, Paula – $10B USD – to help it scale and, theoretically, add real-time web crawling. Apparently, Google has issued a “Code Red” (whatever that means) in response…
I know. I knew the number. Code Red means “Damn — we’re in danger of becoming irrelevant.” I’m a really good Googler but I could never have gotten a consolidated answer from the engine itself. I have to go to another page to do that.
I assumed you knew the number, and yes agree re: fears of irrelevance in Mountain View…
Have we all forgotten the infuriating results we’ve gotten from other chatbots?
I walked into this whole ChatGPT thing with the same skepticism, Cathy – 100%. Then I saw Ryan Reynolds’ Mint Mobile commercial, including his request to include a joke, a swear word, and to “sound like him.” I have huge admiration for Reynolds, and when I saw this short clip from him, it motivated me to dig in:
https://www.youtube.com/watch?v=_eHjifELI-k
Disclosure – I work for Microsoft – comments here are my views and not the views of my employer.
ChatGPT offers huge potential for so many organizations who interact with consumers. The ability to take into account context and rapidly provide a response that is eloquent is incredible. The ability for this to act as an advisor to help with responses offers the ability to improve overall customer service. The challenge comes in terms of total trust — ChatGPT recommends review of some of the output by a subject matter expert. This needs to be built into the process.
Keep us posted on how it’s going with ChatGPT, Oliver – I am fascinated by the race to ad real-time web crawling…and my guess is your company is as well!
I think this is a great example of potentially blending the in-store and online experiences; making the online experience more friendly. In the article’s example, today one might simply type in “medium blue dresses,” and several suggested searches pop up — for women, for girls, for weddings, etc. to choose from.
If this were a chatbot conversation, it could carry out similarly to walking into a store and saying you’re looking for a blue dress, and a conversation ensues — is this for a special occasion, any particular length of hem you prefer, do you like solids or prints?
See what I just did there? I combined these two use cases — someone in tech jump on this!
I wonder if this would replace the never-ending product names in Amazon’s marketplace where vendors try to capture every possible keyword search leading to multiple lines of product names? Here’s one for “medium blue dress”: “PRETTYGARDEN Women’s Summer Wrap V Neck Ruffle Sleeve Tiered Midi Dress Tie Waist A Line Long Beach Dress.” — WHEW!
As far as I can tell, Karen, Open AI and their ChatGPT tool are not the way there with multiple, complex use cases, but they are getting close. If you have not seen the way ChatGPT negotiated cable television prices with a human service agent, this might be worth a quick scan. It will reveal the state (and limitations) of their complex conversation skills: https://www.pcmag.com/news/watch-donotpays-ai-chatbot-renegotiate-a-comcast-bill-to-be-120-lower.
It might indeed. The answers from ChatGPT are remarkable from the playing around I’ve done but are thorough as opposed to succinct. I wonder a bit about unintended consequences with an honest, unbiased chatbot. Coming soon to a chat for you: “It’s also worth noting that short people like you don’t generally look good in stripes, but if you choose to ignore this advice, be aware that you can get this same shirt from a different retailer for 25 percent less.”
Very interesting point re: unintended consequences, Peter. Definitely worth watching, especially for “less tall” guys like me!
I am using an app that is like ChatGPT, called chatAI. It is truly incredible. I asked it about calculus problems, linear algebra, how OPS+ is defined for baseball Sabermetrics, what planet the Coneheads are from, how to do a certain analysis in R, recipes for meatloaf, tell me about Chris Stapleton, how to integrate the Bass diffusion model from differential equations. Not once did I get a bad or nonsensical answer! I continue to be utterly blown away.
As someone who naturally experiments with new stuff, I’ve been in the weeds on ChatGPT, as well as other AI writing tools and AI photo generation. A few things I can report:
Couldn’t have said it better myself — trust me, I tried! Well said, Brian!
Taking into account the general feelings about ChatGPT, adoption or otherwise, we all need to take stock and recognize how new advancements in technology evolve and increase the bandwidth of business.
Consumers want a quick and effective purchase journey, I see the technology ironing out a lot of pains within chatbot interactions. As the technology gets better the majority of demographics will be ok with crossing over bumps in the road.
Chat capabilities get better year after year. ChatGPT is making this available to the masses at a price everyone can afford (currently free). Once it’s integrated into customer support software, it will be a powerful tool. Still, there will have to be an easy way to get to a human agent if the customer needs to talk to someone.
It’s hard not to get excited about a breakthrough technology, especially when it gets a $10 billion validation from a technology behemoth like Microsoft. Our imagination can run freely on the range of use cases and the possibilities for improving customer service or personalization for retailers.
My issue with ChatGPT is more fundamental. It is a general content-creation machine that lacks human intelligence, novelty, and insight. ChatGPT’s responses mimic a high school essay in structure and argument. Ask it to regenerate a reply, and you get reprioritized points with a slight variation on the same approach. The responses are always logical, but as a whole, the answer can be impractical, repetitious, or even nonsensical.
As a society, there’s a real risk of over reliance on machines for rigorous thinking and argumentation. Additionally, the many forms of biases will find their way into ChatGPT’s training data, thus promoting false information.