Imagine that you are about to enter a physical home appliances shop. The precise moment your foot goes through the entrance an assistant appears in front of you saying,
“Hello and welcome to Pliannces, I’d be delighted to help you! How could I assist you today?”
You’ve only stepped in to browse, but already the salesperson is harassing you,
“Hi, thanks but I don’t need assistance. I just want to have a look around.”
The assistant continues undeterred, “Certainly! Remember that I can help you solve your questions and find what you are looking for.”
You try again, “Thanks, but I’m just looking.”
The assistant isn’t going to stop, “Sure! I’m a super-powered assistant. You can ask me things like…”
“I don’t want to. Thanks again.”
Now they are not even listening, “… and I’ve been trained with loads of data from…”
”No really, I’m fine.”
You walk off, exploring the different aisles, but the shop assistant follows your every step,
“I see that you are interested in TVs, let me tell you the main pros and cons of this Oled TV…”
“No need, I’m just looking around.”
The assistant is now desperate to show you what it is capable of, “Maybe you’d like to pair your TV with a new soundbar, what type of video content do you often watch? How big is your living room?”
Eventually, you lose patience, “PLEASE STOP!”
This interaction could go on forever until you eventually leave the shop. The assistant is never going to give up trying to be “helpful.”
Is this the mistake we’re making with AI?
I feel that something similar is happening in the way we are presenting AI assistants to our users. Using tricks like flashing sparks, banners with vibrant gradients or AI-generated results, the tech industry wants to hook users into the greatness of genAI assistants and their potential to meet users’ needs.
These attempts at raising awareness are sometimes made in a clumsy, distracting way that negatively affects the value of these AI features (is that another bore chatbot!?). It’s particularly annoying when users only want to continue using your product in the same way they have always done. Why is that?
Just like the story of the physical store, sometimes you only want to have a look around. Freely navigating the different store sections and browsing the available products. Maybe you don’t have a clear idea of what you want, but you may buy something if you like it.
But sometimes you don’t know exactly how to find the right product or what it is called. That’s when expert shop assistants add value to the shopping experience. They can point you to the right aisle, discuss the specific product features you need or help you compare brands and models. As customers, we know what shop assistants are capable of because we have already interacted with these people and we’ve learnt the value they provide.
Looking for an item in an online shop is a standardised experience: you use a search field for keywords and toggle on and off filters and sorting values to narrow the options displayed. In general, this works quite well. Sometimes you still have to go through hundreds of results but you know what you want and have control over what is shown. We rely on standard search (keywords and filters) and, whatever guidance we need, we seek it outside of the online shop. We ask friends or family, search on different platforms, watch videos or visit brick-and-mortar stores where we can talk to experienced shop assistants (even when we later make the purchase online!).
Watching customer interactions on Adevinta’s Leboncoin platform, it is clear they invest a lot of time and effort in finding the perfect product. We see them trying to search using very vague criteria, or clicking endlessly through product categories until they finally find what they want. If asked, these users would surely appreciate having someone to answer their questions while searching in an online shop.
Unfortunately, having real people online to help users is costly and not a feasible option when serving millions of users. The recent arrival of genAI and LLMs allows us to include AI search assistants in Adevinta’s online marketplaces, helping these users answer questions while searching. For example, through conversation, these assistants can automatically translate the user’s needs into specific search queries (keywords, categories and filters) and guide users on different aspects to consider for their purchase.
This is our vision for Leboncoin’s AI search assistant. We know that standard search will always work better for users who are confident about what they want. However, an AI search assistant can provide instant value to users unsure of what they need. We use artificial intelligence to accelerate their pre-purchase research process and tailor results to their needs.
Sounds like a perfect match, right?
Well, yes, but we should not be misled by this shiny new technology. Why?
Through user research, we’ve found out that Leboncoin’s users:
- Are not aware of the benefits an AI search assistant offers
- Do not expect this functionality in their online shopping experience
Users’ mental models are evolving as technology advances, but not at the same exponential speed. Tech-savvy users may already be asking chatGPT or their preferred LLMs for guidance and some are already trying our platform’s assistants for testing purposes.
However, the majority of users aren’t aware of what this new AI technology offers or how to interact with it because they are used to standard search functionality. Additionally, some might have had frustrating experiences with automated chatbots in the past, which they are now projecting on this new era of conversational assistants.
This highlights the importance of user education and awareness to help users navigate the mental model shift. It’s not just about grasping the user’s attention with flashing lights to increase adoption. At this early stage, it’s key to teach the basics of what these new assistants can do and how to interact with them. This includes what they can’t do and what they truly are: smarter machines (as Ben Goertzel mentions in this episode of Invisible Machines podcast).
An AI search assistant is a smart search tool that allows you to express what you want and the machine will find it for you. It’s a new way of interacting and shopping. It’s almost like chatting with a human store assistant online… but it’s not human at all. As users, we also need to be aware of its limitations and practise critical thinking. LLMs don’t reason, their answers are not always the same (or correct) and they will always generate a response — they are overconfident, there’s no humble generative AI yet. Nevertheless, in the context of online shopping, it’s a breakthrough experience: they allow you to express your needs in a conversational way and return valuable results and information instantly at your preferred online shop.
The talented team behind Leboncoin’s AI search assistant is fully convinced of its potential benefit to our users. Implementing a feature like this is a fascinating journey, full of challenges and discoveries. Similarly to what a shop assistant would do, we are adapting to our users needs and behaviours. We are improving our assistant’s knowledge about the products in our marketplace, category by category, iteration after iteration, to ensure that when customers need assistance in their search journey, we can provide them with practical guidance to teach them the value this new technology offers them.
Stay tuned for future articles where I’ll dig deeper into the project and its specifics.
