How we test fake sites on live traffic | Hacker News Reader

How we test fake sites on live traffic

blog.42floors.com

165 pointsjaf12duke39 comments

A theory on why the "ugly banking site" did so well. Out of all of the designs, that was the most traditional, and probably most "comfortable" to the user. The site to me feels like it has depth, and it's friendly to the user, encouraging them to explore. The others while more effective (conversion rate) felt like they were engineered to spoon feed me just the right data, and didn't feel like "complete" sites open to browsing.

This of course is all anecdotal based on my years of being a web geek but sometimes the "feel" of the design speaks volumes where data and raw number crunching cannot.

Perhaps the "Ugly banking site" makes them think you're not a bunch of kids with some VC money ;)

Seriously though, it could be that a less trendy design gives the users some reassurances that you've been around since that design style was in fashion.

I love that Ugly Banking site is clearly based on US Bank's website, which was actually redesigned in the last 12 months.

https://www.usbank.com/

It looks no more like that one than most other bank websites I've seen :)

Agree. The 'ugly banking site' is a traditional design who is still a reference for a major part of visitors.

> The winning variation was Google Hover Clone

I was enjoying the article, but feel like it just abruptly ended. Why was the clone the winner? I would have loved to hear more about your process for judging the successful mockup.

Bump.... if Ugly Banking Site had lowest bounce rate why wasn't it the "winner"? If you're just looking at a static page what else is judged other than bounce rate? If you're measuring conversions, isn't most of that influenced by the existing site? Or do you attribute it to some lasting impression based on how the site was entered? Glad Ugly Banking Site didn't win, but I am curious about the critical metrics.

Fewer bounces doesn't mean more conversions. The winning site might have generated the most tours of office space but the Ugly Banking Site may have generated more overall page views.

That's what I got from it.

I understand & agree, but my point was that conversions would still be accomplished from the existing site, just entered via a static landing page. The conversions weren't coming from the static page, it was just an entryway into the existing site.

Enter = no bounce, Exit = bounce

Correlating stats collected once they leave (click anything) with these static pages is a bit less scientific, so I was interested in hearing the logic.

Good point. I just added this note to the end of the article:

Our criteria for picking the winner came down to one number: tour request rate. In other words, what percentage of users found a listing they liked and contacted it using a particular design. We use other data points to augment this core metric (bounce rate, time on site, number of listings viewed, search criteria revisions), but ultimately a variation only wins if it's better than the control at getting users to an office space that they like.

Nice experiment. Do you have any insight whether this is due to the design being inherently superior, or if it is just the fact that users are familiar with the design due to it being a 1:1 clone of Google?

At our company, we've found that a cohesive design throughout a funnel will always* drive better results. If you think about google as a step in your funnel, then you'll be improving conversion through that same principle.

Also, you've guaranteed familiarity by using Google to buy users. It might be interested to buy users via Bing or Yahoo or something, to see if that same design wins.

But, that's purely gravy. You've probably hit a pretty good balance of testing and results with your experiment already.

* "always" -- not really, but usually.

Fascinating.

Next question: Would it improve your success rate to have multiple versions of your site that echo the designs of the sites that drive most of your traffic, and automatically direct users to the version that matches their referrer?

There's probably more overhead involved than could ever be worthwhile, but now I'm curious.

When we rolled out hover2 in production it looked significantly different from Google, so we were concerned that we might lose the lift. But we didn't. So we chalked the lift up to the hover UX paradigm rather than the visual similarity to Google.

Cool, thanks for the info and the reply. I think there may be something to that type of design for comparing locations vs listings, it is similar to what sites like airbnb use. I generally like to think that those types of sites have done the testing already and came to that design (or stuck with it) for a reason.

I really enjoyed this article - the one question I have is if the "winner" has been rolled out to the masses (not only the SEM channel.) I'm a little scared for you guys because the test results show that the design works well for visitors from SEM, not necessarily from other channels. Have you tested this on traffic from other channels (organic, referral, etc)?

How long did you have the fake sites up? I would have thought you couldn't leave them up for very long before frustrating users, but it seems like you would need a good chunk of data before you would know it was statistically significant. Or did you also have some sort of disclaimer for those users so they knew it was a beta test?

We left the fake sites up for 3 weeks. The parts of the design that required a DB connection, like sorting, pagination, contacting a listing, would just redirect to the matching operation on the live site on click.

The keyword is "Fake". The static HTML5 websites take much less time to produce thus reduce programming costs and shortens precious time-to-lauch.

In a nutshell: Moqups > PSD > PSD2HTML + Static data > Fake sites

Getting user feedback on fake stuff is great. Tim Ferris didn't know what to call the book that turned out to be The 4 Hour Workweek, so he spent money on AdWords to see which title more people clicked on. [1]

[1] http://futureperfectpublishing.com/2009/01/23/the-sticky-goo...

Hover clone was the only one that put the locations on a map. This test seems to, more than anything else, have proven that when searching for office space _exact_ location is one of the primary criteria.

I wouldn't be at all surprised if "just show me where it is on a map" is a feature most commercial real-estate search sites are missing.

This explanation does not address the fact that all three of the original designs had a heavy map emphasis including the ability in Unified View to make the map full screen. Keep in mind that the split test was 9-way in that each variation had to outperform the control. I think hover1 outperformed all versions because it gives the ability to quickly process massive amounts of data without clicking or changing pages.

The ability to make the map full screen != a large easy to use map and definitely != ability to see exact location.

What percent of users clicked to make the map full screen?

"The ability to quickly process massive amounts of data without clicking or changing pages" sounds like it may be a good design mantra for your business.

You should really consider changing the title to something a bit more descriptive. Its a very good article about strategies for A/B testing design ideas, but I didn't get that from the title, and would it skipped it if not for the fact that I really enjoy 42floors blog posts.

Looking at a listing[1], I notice that you provide email addresses to users and presumably act like a craiglists for anonymous messaging...

I'm just curious though, don't you find that you get spammed out if you do this?

I think providing email addresses to users on a marketplace is a great way of keeping discussions/messages boxed up in your platform, but showing the emails on the site must result in way too many spam emails?

[1] : http://42floors.com/ny/new-york/25-w-39th-st/3171

We used to expose the raw email address of the underlying user. Now it's a two-way email proxy, which is why the domain is "Floors.me" e.g. rich.farmey.31719@42floors.me

We do get spammed but we've been able to keep a handle on it. Mailgun's filters capture most of the spam and we have backup handlers in place to address the rest.

Do users know who they're emailing with the anonymous addresses? The username@42floors.me address may look like they have to communicate via 42floors (meaning more friction) instead of directly with the manager. Craigslist emails are more obviously throw aways.

The hovering map seems to be behaving oddly (Chrome/Mac OS) for me and blocks off a large section of the top left with a beige colored background when I click on "More Map".

Otherwise, very surprising to me that this was the winner. Did you see similar results across all browsers and form factors (especially interested in desktop vs. tablet).

Very interesting, thanks for sharing! What specific actions did you measure to determine a winner?

Noticed you answered the question above. *Was just clawing my eyes out browsing craigslist for NYC spaces, remember hearing about you guys a little while back, thanks for refreshing my memory ;)

I kinda liked the map more. I thought moving to this new hover layout was a business decision (e.g: get some clients to pay for being featured on top of the list as the map makes all listings have equal look/chance to be discovered). apparently you were just A/B testing.

I'm not affiliated with the company, but Visual Website Optimizer offers the ability to do A/B tests between different URLs - rather than just variations within a page (which it also does).

So didn't users get pissed when they landed on the fake site?

Real data, just pre-populated. Not live. Cached, for weeks.

You'd never know unless you tried to book and were told the space was already taken.

The hover version exposed the most information in a usable and visually-understandable format. I am not surprised.

Did you conclude this before reading the answer?

What is this gem called? I've seen it on boatbound and MyTime, now on 42floors. Same Rails based layout.

what did it mean 'fake'? Weren't all 8 fully working versions of the website?

Isn't there a worry that "fake" sights would damage your reputation?

This reminds me of the worry of the potential A/B tester: won't my users get confused if they see site A from one web browser/computer, but site B from another?

I think this is just a self-defeating worry, though. Most people won't know or care about your site or brand, or remember you from one session to another. If they don't like your page, they'll just click the back button. If they do get upset, well, the internet is effectively a pool of infinitely more people who can click your adwords.

So exercise caution when changing big features used by paying customers, but don't be afraid to experiment and show things to random visitors.

Reply on news.ycombinator.com