nicolas leroy

Hunch as a shopping recommendation tool

February 08, 2010

What is Hunch?

Hunch is the new startup of Flickr co-founder Caterina Fake; as such, it received quite a lot of attention when it was launched in March 2009 as a tool to help users make decisions. A kind of Yahoo! Answers-like, powered by the community, with an assistant approach: a series of questions that lead to a recommendation at the end. In Hunch’s own words:

In 10 questions or less, Hunch will offer you a great solution to your problem, concern or dilemma, on hundreds of topics. Hunch’s answers are based on the collective knowledge of the entire Hunch community, narrowed down to people like you, or just enough like you that you might be mistaken for each other in a dark room. Hunch is designed so that every time it’s used, it learns something new. That means Hunch’s hunches are always getting better.

Among all the questions available on Hunch (“What’s a good way to go green?”, “Which NY Times bestseller should I read?”…), a good amount are shopping-related… What camera / camcorders / Mac / credit cards / women’s shoes… to buy? A way for Hunch to monetize its audience, as recommended products are enhanced with sponsored links powered by BizRate or Amazon.

In early December, Hunch launched a new interface; as Chris Dixon, Hunch’s co-founder, explains:

The main idea behind the new interface is that you now see Hunch’s top recommendations updated in real time as you answer questions.

Besides giving instant feedback, this also lets you see how each answer you give affects the recommendations.  We think it does a better job exposing Hunch’s intelligence, particularly the statistical intelligence the system has acquired from the tens of millions of user feedback clicks collected over the past few months.

It is time to do a quick review of Hunch as a shopping recommendation tool.

Choosing a camera with Hunch

The “Cameras” topic is pretty popular on Hunch; how convenient for me :) It is organized around a list of questions that enable to filter, sort and rank results among 114 camera models. The topic homepage lets you start right away – first question on the left-hand side column, result set in the main column.

Cameras - Hunch_1265617815527

Three types of questions are asked:

As soon as a question is answered, the list of relevant products is refreshed; and in the end, if enough questions have been answered, you get a list of products that you can manually review and keep/remove from the result set.

The result set can then be viewed in a specific “product page”-like view, which contains some extra modules and offers access to a “why did Hunch pick this product” page.

Result page - Hunch

Result explained - Hunch

When logged in, a set of tools that offer write access to Hunch content becomes available. I encourage you to register on Hunch, and test those tools to create new topic, enrich existing topics, enhance your profile and the statistical engine of Hunch by answering various questions: all this stuff is really clever.

Overall, Hunch provides an original experience to choose products. As all the content is produced by the community, it gives a particular flavor to the result set, even if the relevancy is not always optimal: users may be more keen to understand the results than in the case of pure algorithmic-produced result set.

A drawback of the Hunch experience for the shopping-related topics is that the community must create and maintain the product database. I think it would be an improvement to let users start new topics from an existing product database (i.e: licensing BizRate database, or using Freebase or ProductWiki databases). This would avoid users to spend time on managing product definition; and allow them to focus on questions and mapping products with answers.

What does it mean for CSEs?

CSEs are really good when it comes to find the best deal for a specific product. But, when doing product research (i.e: I want to buy a digital camera), the experience is not always satisfying. Sure, faceted navigation on tech features and user/expert reviews are convenient ways to narrow down results; buying guides can also guide users to some extent. But CSEs lack an integrated search/browse experience where tech features and product usage are mixed, and which can be appealing for expert and non-expert shoppers. Quite a challenge indeed :)

Hunch demonstrates the social dimension can improve the buying process, in a far more subtle way than what “social shopping” sites are doing. It could be wise for CSEs to get inspiration from Hunch:

4 commentaires

Tom Pinckney
on Feb 08, 2010 / 4pm
Hi, it's Tom from Hunch. Great blog post!

I would like to follow up on the issue you raise about surprising questions being asked. One of the interesting things about Hunch is that anyone can create questions and the system learns whether those questions are correlated with what people like.

In the case you mention about being asked whether you are pro-life or pro-choice, the system has learned that this question is highly correlated with what kind of camera people like. Here's one example that shows there's a statistically significant increase in probability between people who say they are pro-life and people who like the Nikon D40:

The system has learned that this question is not just correlated with the Nikon D40, but many of the cameras in Hunch.

That said, I agree this is a jarring question for many people. One thing we're still working on is the right mix of editorial review, explanation about what's going on and algorithmic learning for these questions.

on Feb 08, 2010 / 5pm
Hunch is definitely one of the most interesting new sites. Will probably be a "Top 100" site before long. There are so many things they will be able to do with their universe of smart objects.

I think your "cameras" example is good at highlights the strengths and weaknesses of the system. On the plus side, it is unique. So even if it is not your primary tool, it makes sense to check in and see what they say. Also, the community feel and game mechanics are very conducive to leaving product reviews and adding content.

The downside, IMO... it is disconnected from the rest of the comparison process, at least for the more intensive topics. You can't set must-have attributes and you can't easily compare the underlying data... sometimes you at the mercy of the "black box" and plenty of "black box sites" have come and gone.

Although I'd guess they will iterate quickly. And for topics that do not have any real decision support tools, Hunch is a revelation. After you find enough of these topics that click for you, Hunch starts to become top-of-mind as a portal of decision tools.

on Feb 08, 2010 / 7pm
@Tom Pinckney - thanks for the explanation ; I understand the logic and the need to find the right mix. It also shows me how exciting it should be to analyze the correlated data produced by your engine. Keep up the good work !

@Sean - Totally agree with your comment and the uniqueness of Hunch. I wonder: by using Hunch APIs (I have just scratched the surface), there may be ways to mash up data with other data sources (i.e: coming from a CSE)

on Feb 09, 2010 / 12am
Great post. Your conclusion is spot on. We'll try out some of those ideas. The future of shopping online needs to up its game. However, shoppers are only temporally interested in any given thing they are buying so getting so much info from them can be challenging.