Product Search is Broken: Here’s How
By: Team Inscripta
Let’s say you are looking to gift your brother a camera on his birthday. You only have a couple of days to decide on the most suitable product at a price you like, and you don’t know too much about cameras (if you do, then imagine a category you don’t know much about — laptops, mobile phones, home theater systems, wi-fi routers, …). What do you do?
Search and Filter
The easiest option would be to type in “camera” on Amazon (or any other e-commerce website). You are faced with something like this:
You wonder how you could work these filters to get what you need. You know your brother is interested in landscape photography, and since he often goes trekking, the camera would have to be reasonably sturdy. And it would be great if it could work with certain accessories he already has.
So, what kind of camera do you need to look at? What display size would be ideal? CMOS or MOS? (BTW, what are they?) And gluten-free — what’s that got to do with cameras?
Explore More: Browse, Study, Call, Consult, Compare … Phew!
You realize you need to understand cameras a lot better before you can make this decision. Are digital viewfinders better than optical ones or the other way around? Is it worth paying a hundred dollars more for those extra megapixels? Is there some sensor type particularly suited to nature photography?
You could read a few product-review blogs, or go through some camera buying guides. Maybe you could consult a friend or family member. You could possibly visit a physical store and talk to the salesperson there. All this is time-consuming, tedious and sometimes a touch embarrassing for someone who doesn’t even know what questions to ask. After all this, you are still not sure whether the information you have is up-to-date and accurate with respect to the products currently available in the market.
Filters you will love? Not.
One popular way to supposedly improve the experience is to have better filters — filters that most users can understand, and that cover the entire range of attributes customers care about. For example, many online portals add numerous tags to products to make it easier to narrow down to the right product. The problem with this strategy is information overload: there are just so many filters that the user ends up being overwhelmed.
And this turns out to be the story for many product categories: there is a bunch of technical specifications with no clear indication as to how each specification affects your product experience, and if there is an attempt to add meaningful filters, there is an intimidating number of choices. For example, let’s look at the results for “laptops” on another website. There are over 700 results and the following filters — 26 filters with a total of 250+ check-boxes!
Problems with filters
While there is no denying that filters are quite useful for informed users who have a specific kind of product in mind, in most other scenarios they fail due to some combination of the following reasons:
- Choices are not meaningful to the customer. Example: a laptop for deep learning or gaming is what the user wants, but the options are “GTX-1080i”, “Intel Graphics 620”, VRAM sizes, and so on)
- Not in the relevant order (that matches the users’ prior knowledge). Example: the user knows she needs a Windows laptop for desktop publishing, but she still has to fill out filters for processor, memory, OS, software … all of which can be inferred fairly easily
- Information overload: for most categories, especially with many derived tags/filters, there are too many choices, and the user does not know where to start (the laptop example above)
- Hard boundaries instead of soft preferences. Example: the user is okay with spending a few dollars above the range (around 200 dollars rather than strictly 100–200) if her other preferences are met. This is especially important when there are no products with the exact set of filters the user has in mind
Next generation product search
Overall, the current product search and filter mechanisms are too rigid. What is needed is a flexible interface that can elicit the users’ prior knowledge effectively, and that is cognizant of the users’ ability to articulate.
As discussed above, adding more filters to the search interface is not the solution. The user already knows a lot about her needs and is willing to convey them — second-guessing them with her clicks is sub-optimal. Also, users shouldn’t have to understand arcane technical specifications of products to make the right decision.
In a nutshell, we envision an interface that:
- Presents aspects that are relevant based on the user’s current understanding
- Speaks a language that is based on the user’s product experience, and not technical specifications
- Understands nuances of the user’s preferences
- Is driven primarily by the user, but proactive enough to bring the user’s attention to aspects that she is unaware of
- Allows the user to interact in a conversational manner
At Inscripta, we are working towards a next-generation conversational search platform that aims to provide precisely this. We will describe this platform in greater detail in part 2 of this post. Meanwhile, for an early demo of our product (Revlo), or to learn more about how Inscripta can help you incorporate conversational AI within your own product, drop us a line at firstname.lastname@example.org.