Episode 3: April 4th to April 11

(Side note - late tonight 5/31/2006 - we are going to release a new version of the site and homepage that begins to focus us in a whole new direction... but I'm getting ahead of myself).

April 4th, 2006:

While the rate of new uploads continued to decline the number of photos in total was significant. By the third week over 4.58 million photos have been uploaded in total. Flickr by comparison uploaded 1M in 6months and 10M in two years (btw - I later realized a comparison to Flickr or any other photo site just doesn't matter), but on April 4th I didn't understand this. We did a comparison of how Riya stacked up in Alexa ranking and found the following. (see image). We used all of the Web 2.0 photo companies on Baris Karagodan's excellent list . Here is how Riya stacked up. In fact, of the vintage of Web 2.0 photo companies that had launched in the last 1 year Riya had quickly surpassed all. All those ahead of us had a year or more lead. (btw ... If you think this was just the Techcrunch effect, our Alexa rank today is even higher than these lists show).

Flickr - 73
Photobucket - 88
Fotolog - 243
Frappr (Location) - 2207
Slide - 6129
Riya - 9267
Bubbleshare - 17042
Zooomr - 24320
Pixagogo - 26050
Zoto - 29687
23hq - 37776
Smilebox - 65331
Donmai.us - 96163
Phlog.net - 97478
Sharpcast - 100244
Flyinside - 157216
Fotoflix - 215005
Jellybarn - 329018
Theblackstripe - 441657
Ondergrond.org - 3,047,454

When we compared Riya to Image search engines (excluding Google and Yahoo Images since Alexa didn't show us Google images seperate from Google text search, etc) we were already number two. I will elaborate more in a later episode on why this image search analysis was so important.

Photo Search / Image
Picsearch.com - 5001
Riya.com - 9267
Ditto - 21,887
Pixsy - 60048
Cydral - 393,281

Okay so maybe things weren't so bad... things were slowing but we had achieved a lot in a little time... I went to bed on April 4th late at night feeling good ...

I roll out of bed at 8:30am on April 5th. Pick up the Treo to read email and see the following from Dhiraj our new QA guy.
>>Subject: Red Alert :- Manual Training not running on PROD. No message in the queue either waiting >>for face rec.

I nearly dropped my morning tea. Red alerts are reserved for when a critical system is down (either our search engine or our auto-recognition system). This meant that anyone who used the product was not getting any faces recognized. As yes we watch Star Trek in case you missed the reference...;-)

Again our engineering team is on it. Later that day Azhar sends our the following email:

>>Guys:
>>The following are the issues that we need to fix on PROD:
>>Nikhil (Recognition):
>>1) Manual Training not working via inline flow
>>2) Manual Training not going to QA page
>>3) Per Diem: Manual training (ROI and Bulk) seems to run on the same set of faces every time. >>According to Diem from the behavior of the manual training, it looks like less than 1K of vis sig was >>used.
>>4) Per Dime: Auto-rec recall is very low. It seems half of my unrecognized faces wasn't passed to >>the auto-rec. There are hundreds of photos in the 2005_01_01 folder, which I pretty sure the auto->>rec would return at least 80% of them. In reality, none of the faces in this album was auto->>recognized.
>>5) Per Diem: Inline-flow doesn't show any face. It seems that this happen only with, again, the faces >>in the second half of my face list (just try with a few faces in the top most albums).
>>6) Per Diem: Same faces appeared again and again on QA page. I've looked at the nolist files, the >>file get longer and longer with the same entry. Have no idea what's the cause of this.

>>Neelesh (Search):
>>1) Clicking on Public people's page on some photos show no result
>>2) Public photos turned private are slow in getting removed
>>3) Text page shows that the number of photos with text is around 9k or so since the start
>>4) Regenerate all search indexes

>>Tanvir (Uploader)
>>1) Please change the message "JRE 1.5 required.... " popup window to say "Jave Runtime >>Environment (JRE) 1.5 is required..... If you like to install is yourself please go to the following link." >>I cannot recall exactly what the logic is at this point i.e. do we install it automatically or not. I know >>at one point we were doing it.
>>--
>>Azhar

For the first month the experience of training and recall did not match the full ability of Riya. So if people were happy enough with Riya (given the 71% who were) then if they had actually seen the product in with no auto-recognition bugs they would of had an even better experience. Burak and his team always hated these bugs btw, they worked so hard to make the best algorithm and because of bugs people weren't seeing it.

At times Burak and Azhar would fight about this. Given their styles it was mostly Burak reporting bugs to Azhar, Azhar pointing out algorithm failures to Burak, and me talking about men in glass houses throwing rocks...;-). Like many things at Riya these meetings had the normal 5 minutes of fighting (clearing the air) and then 2 hours of problem solving. I termed this the Riya lover's quarrel meeting format. The two of them actually have gotten quite close and these sessions brought them even closer. I'm not sure they know this, but all three of us are like brothers. We've developed a very close respect and admiration for each other.

One of my key beliefs is the worst thing for a startup is passive aggressive behavior. Many people who join startups from big companies bring this culture with them. The political nature of those companies requires them to do this to survive. At a startup, you just can't let things simmer in the background, you have to pull issues to the forefront and deal with them as fast as you can with the least amount of denial. The Riya lover's quarrel meeting format is designed to get issues out as fast as possible.

Once the dust cleared we focused the rest of the week on all of the ideas to improve the site. What we came up with was interesting:

a) We had realized that there was one major thing different about Web 1.0 vs. Web 2.0. In 1.0 no one had anything on the web so if you had a new service people would just use it. In 2.0 most people have something they already use on the web to do the task. So if you have something better (even if it is a lot better) you also have to have the key features they used from their 1.0 site before they would shift to you. Web 2.0 has a lower cost to awareness (due mostly to blogging), but switching costs are considerably higher (at least for photos). This was a big insight for us.

b) We also realized that having a team in India had allowed us to have a big engineering team (we were now 50 employees up from 25 in January) without the costs associated with having it all in the US. This in turn allowed us to build a lot of different features in parallel.

Our first strategy was to address the issues of how to make the site more sticky. For the entire week, we compiled lists of features, did mockups and specs, and organized the team. In a company meeting I sent out the following organizational team chart (see photo).

We were going to build everything in an effort to make things more sticky.

By the end of the week, we were ready to execute this new strategy.

Then if dawned on me... I was about to make a huge mistake. You never never never have a parallel strategy. You focus on the one thing put all your eggs in that basket and do it super well. You maybe wrong but at least you will execute it well. As a startup you just can't afford to hedge. I had learned this the hard way at Andale, but the desire to hedge is just so much a part of human nature (or at least my nature) it had come back in.

By the end of the week we scrapped this approach and started anew...

Next episode: What to build next rev 2...

riya
web2.0
riyarocks

Comments

[John Walker]

I've been reading these episodes and they are absolutely riveting to me. Please keep them coming. I'm a software dev and although I'm in a more traditional product-based part of the business, this is really fascinating.

I've been beta testing Riya since your initial release and I'm very impressed. However, and here's key info for you, I have posted photos and pointed lots of non-technical family members to Riya. Here's the number 1 request...How do I get prints of your photos? I know that is anathema to web 2.0, but I've got to satisfy their needs since they are family and most of my photos are intended for them. If you made that possible, you've got stickiness. Right now, I've got to go shutterfly.

If you're thinking big, then go for the real market which is not the 56,000 users subscribed to TechCrunch. Thanks for Riya, as I really do love it and want to use it.

[Harry]

a team of 50 developers to be managed across the Pacific? You have to expand on that experience!
Just a thought: Riya has one strong feature: its face recognition capability and from that better search results. Have you ever thought building on that and leaving out the hosting/sharing part (where others have the lead and critical mass)? Do you have to host the photos or could you just pick them from the web??
That would mean less compexity, costs and developers.

Thanks for the great insight.

[Balaji Sowmyanarayanan]

Go on Munjal,
The blog is more like a thriller novel.

[Scott]

I'm amazed. This is perhaps the best play-by-play of the startup experience I've ever read.

I'd love to hear more about how you realized your parallel strategy was off, and what you did to realign.

[Ro]

Munjal,

Why wait for people to upload pics? I have a hell of time finding related images for associated people/products on the web. Why not take Riya cross-web vs user-centric?

Ro

[anamik]

Hey Munjal,

Given your stress on using alexa to find out where riya stands i thought this might be of interest to you.

On June 1, 2006

http://munjal.typepad.com today's rank : 7815
http://riya.com today's rank : 7026

[John Koetsier]

Munjal, I cannot thank you enough for telling it like it is ... it's like having a ringside seat on your start-up.

I can at least partially relate: I've done "intrapreneur" start-ups within a large company, and had release dates that can't be met and death marches to meet them and killer bugs that just about throw everything in the trash. It's great to see you tell so honestly how you dealt with it.

I am on the edge of my ringside seat, btw, impatiently awaiting the next installment. Hurry up and post it!

:-)

[Prakash]

Alexa numbers doen’t represent the real world.Alexa stats represent more technical,marketing oriented or someone who have alexa tool and not the public. better to do some research when you post about alexa stats or just google "alexa stats" ;):).
Keep up the good work, good luck.

[Dominic]

Great post keep up the good work.Feel free to visit my blog and possibly comment or give me some suggestions or tips. Thank You

http://www.10makemoney.blogspot.com