Starting Up by Rahul Pathak

Alex Iskold has a great post on Read/Write Web about Google’s Custom Search Engine and how it shifts the value in vertical search from infrastructure such as crawling and search relevance to UI and site selection.

I completely agree that Google CSE is a net positive for the vertical search space. However, it’s important to keep in mind that it only allows you to focus on finding web pages restricted to a specific slice. Having the crawling and search relevance components freely available is a huge boon to new entrants. That stuff is really hard. However, Google CSE doesn’t help you when you’re search for domain specific items that aren’t web pages e.g. products, flight information, resumes etc.

For example, take Couponlooker, which is focused on helping people find online coupon codes. If you search for “amazon coupon codes” on couponlooker, you’ll get back actual discount codes with expiry dates.

If you do the same search on a Google CSE, you’ll get back pages with coupons on them. A subtle difference to be sure, but it’s extra distance between the user and their goal.

Google’s Custom Search is a pretty incredible product, but you’re limited by the underlying infrastructure and search semantics. This is not a bad thing - the guts of the system are incredible and for web search haven’t been beaten yet. If you’re looking to do something different though, you’ll probably need another approach.

Ok, sorry, couldn’t resist the A-Team quote. I just got an email from Casey, one of our developers, who said:

Couponlooker saved me 10% at Godaddy.com <eom>

That’s the kind of email you love to get.

Last week, we rolled out some small tweaks to Couponlooker that should make it even better for finding online coupon codes.

• Fewer sale announcements, more coupon codes:
  We now work harder to make sure that it’s easier to find actual codes that you can use at checkout rather than just links that you have to click on to obtain the discount. Our data suggested that the latter were often just indicators of sales and general savings rather than specific discounts that you could obtain when checking out.
  
  
• Merchant Normalization:
  A lot of coupon sites list merchants under different names e.g. Amazon, Amazon.com, Amazon com, Amazon Coupons, etc. This can lead to clutter in search results and is also detrimental to the user. From the user’s perspective, the 4 variations above should all be returned when they search for Amazon coupons. This is an ongoing process, but we’ve been working on this and results should start to improve.

Improving search results is a gradual process and the Tweak-Measure-Iterate cycle is in full effect.

When you’re evaluating whether or not your site is meeting its user engagement goals, it’s important to look at the right metrics. What’s right for a social network is not right for a search engine. By the way, if you’re not measuring things like visits, page views/visit, time on site, etc you should start immediately. (We use Google Analytics at Judy’s Book.)

In my notes from the Facebook Seattle Garage, I mentioned that the FB team was talking a lot about page views/user, time on site etc. These metrics make a ton of sense for them. Their product is all about user’s spending most of their online time adding to their data on Facebook, acquiring new friends, downloading new apps, etc.

If you take Couponlooker, these metrics don’t make a ton of sense. Our goal with couponlooker is to have users come to the site, find the coupon code they are looking for in 1-2 pages and then leave. If we see page views/user and time on site start to spike, it could mean that users just love searching for coupon codes, or more likely, it means that there’s something wrong with our search relevance. In Couponlooker’s case, we’re looking for 2-3 page views/user (1-2 pages of search results and a click) and we’re looking for a relatively short time on site. Our goal is to satisfy a user’s need quickly and have them return in the future. Naturally, repeat visits and direct traffic (users who typed in or bookmarked your site) are the life blood of any site.

At the end of the day, you’ve got to measure your performance to see how you’re doing and to improve. Just make sure you measure the right things. Figure out the optimal user behavior and the metrics should be easy to figure out.

We released some significant new functionality on couponlooker today. (We also put out some new features on Judy’s Book which I’m really excited about but more on that tomorrow.) Kurt and Dave (Ops and Test) deserve a special thank you for great work in helping us iron out the kinks in this release and making sure the site is up and running.

The biggest user-facing improvements on couponlooker are the inclusion of ‘% off’ filters and related merchant filters to help users navigate search results.

The reason I’m excited about the filters is that they add significantly to user value. If you do a search for ‘digital camera’ on couponlooker, you are presented with a set of results for digital camera coupons (naturally enough) and at the bottom of the page, we present a set of links that allow you to refine your search:

If you click on say the 50% link, you are shown digital camera coupons for 40-50% off.

This is a simple change, but goes a long way to helping users navigate the search results. The merchant filters serve a similar purpose. They also help normalize data across coupon sites so “Dell Home” and “Dell Home, Inc.” aren’t treated as separate sites.

We also released an early version of a sponsored listing system that allows sites to promote specific coupons. These are clearly marked as sponsored listings and clearly distinguishable from the main search results.

As before, we continue to de-dupe coupons found on different sites in order to help users find the unique offers out there that are available to them.

Alex has a good post that goes into some detail about how we implemented the above features and it’s well worth a read if you’re interested.

With couponlooker, our goal is to drive traffic to the best coupon sites on the web. We’re also continually adding sites to our index. If you have a favorite coupon site, let me know and we’ll get it into the system.

These aren’t radical changes, but we’ve got to keep getting better at the basics in order to deliver value. There just aren’t any shortcuts.

Last week was a big one at Judy’s Book from a release perspective. We launched a new home page (existing members will have to sign out and clear cookies to see it), put more local offers in the system, and pushed out a host of bug fixes. There were a couple of cool improvements that didn’t make our release deadline and they should be coming out in the next couple of weeks.

We’re very focused on helping consumers find out what’s on sale near them and the new home page reflects that.

Oh, we also released some improvements to Couponlooker - we added some more sites to our index and improved the performance of the widget. Bloggers with the widget installed will now see it load much quicker and perform much faster on searches.

I came across a cool thread on LinkedIn Answers via Powerset’s Blog. In it, the COO of LinkedIn asked the following question:

“If you could build the perfect search engine, what would it do?”

The answers are pretty interesting and the themes that crop up are around context, intent and some interesting questions are raised about the role of social search and commenting on search results.

There are some ideas here worth exploring in the context of local deals search and coupon search on couponlooker. In both these cases, restricting the domain should make it possible to do a better job of inferring intent and delivering relevant results. Definitely a case of easier said than done, but a problem worth attacking nonetheless.

It’s also clear that spending more time on LinkedIn might not be a bad idea.

On Thursday we released a number of couponlooker search improvements. You can now use quotes around phrases to get an exact match and use the ‘+’ and ‘-’ operators to require or exclude specific terms. In addition, multi-word query performance has been improved as well. Response overall has been very encouraging so far and we’re getting requests for index-inclusion from coupon sites that were not in the initial set.

I love the stress associated with the day of a big release. You’re running around trying to sort out last minute issues, getting bugs ironed out, making sure your data is in the right place and your marketing team is anxiously waiting for the site to be live so they can start their work. It’s hectic, people are crabby, but it’s really fun.

I went through this last week with the release of couponlooker. It felt great to get it out and to take the 15 minutes to appreciate what we had been working on. Even though I knew there were issues and things we had to fix, I think it’s important to take the time to step back and celebrate the accomplishment.