Put down that axe, Mickey!

In the German ballad Der Zauberlehrling, made famous by Disney under the name The Sorcerer’s Apprentice, a young apprentice learning the ways of spell-casting is tasked with the boringly mundane and repetitive job of fetching water.  Lazy and overwhelmed by work, the apprentice decides to use magic as a shortcut for achieving his goal.  Due to a lack of experience casting spells at this level, the situation gets out of control when he realizes he’s unable to make his “cheap labor” stop.

Several months ago, I hired twenty workers using the website vWorker.com in an attempt to populate my database with thousands of recipes.  At the time, I believed getting recipes into the database was the single most important task to attract users, and figured in this case, quantity was more important than quality as the good recipes would just bubble to the top of search results.  However, the workers I could afford to hire were, in general, not the types that really take pride in their work.  What I have now is thousands of mostly bad recipes, with no real way to review them or weed out the bad ones.  The few users who do check out my site are instantly given a bad impression, as anyone who cooks will surely notice glaring mistakes, such as “flour tortillas” in an ambrosia cake recipe or “onions” in a chocolate chip cookie recipe.  Even worse are mistakes that are tough to spot, such as 5 cups of flour when the original recipe says 2 1/2.  A novice chef who prides them self in their ability to follow directions will face kitchen defeat in a floury mess.

These workers generally make themselves unavailable, do not respond to emails, do not report their status, but continue to enter bad recipes by the dozens.  Out of the twenty I hired, only three have finished the task in a more or less satisfactory manner.  The majority have either stopped doing any work, quit, or I’ve unsuccessfully attempted to fire.  vWorker seems to have caught on as I’ve been hogging so much of their arbitration resources that I think I’m now a legend at that company.  They actually stopped allowing me to fire workers since I didn’t pre-negotiate a fixed deadline for the projects, so technically the workers have not failed yet.  vWorker had the nerve to tell me, “Mike – you don’t need to check each recipe your workers enter; it’s their job to do the work accurately within a reasonable timeframe.”  They said this about a worker who had already entered almost 200 recipes with “1” for the amount of every single ingredient.  I had to construct a special SQL query to delete all of these!  When I attempted to alert this worker, he would not respond.  I then disabled his account, and he simply created another account and continued working.  Yes, one account split into two; now you’re getting my opening analogy.  vWorker still would not allow me to cancel this contract, which is probably just as fine because I’m certainly convinced the cancellation of the contract would in no way be motivation for these worker to stop working.  I was finally able to reach this worker and bribe him with $10 to go away.  He still had the nerve to complain about the fact he was forced to pay the cancellation fee even though I had to spend hours cleaning up a mess he took no responsibility for creating.

One thing that I’ve learned is paying these workers more money in no way guarantees better results.  In fact, it appears to have the opposite effect.  People who enter recipes for the money are simply in it for the money.  Two of the three workers who have completed the contract took the job for $20 and $25 for 1,000 recipes respectively.  One worker who I hired pre-beta entered around 500 recipes and refused to take any payment at all, saying she was not doing the work for the money; she simply liked the site.  Another worker just wanted to do great work to get a good review on vWorker.  She moves slow, but the recipes she’s entered have all been accurate and complete.

The workers who charge more, such as over $100 per 1,000 recipes entered, mostly rush through the work, do not proof read, or quit by the time they get up to around 100 or so entered.  A lot of these workers simply “sub-contract” out the work to people who will do the work for even cheaper, then don’t supervise the work simply hoping to collect free overhead money.  One such “sub-contractor” even emailed me directly, saying his boss had basically disappeared and would not respond to any form of communication.  He stopped entering recipes (after over 500!) for fear that he would never see a dime.  Of course, I have no way to correct this situation as his boss is not responding to my emails either.

From this experience, I’ve learned a valuable lesson.  Think twice before you build a website based around content.  If this is the route you choose to go, make sure at least one of the following conditions is true:

1. Your site offers value without much content. Mine does not, since without recipes no one can “plan” anything, and the meal planner will yield “no results” for most any query.
2. You have a plan for automatically importing or aggregating content. Make sure this actually works by prototyping such a process first; make sure you can actually aggregate data correctly.  My ingredient normalization architecture basically prevents me from any automated recipe importing without human supervision.
3. You have the budget to hire skilled workers to generate content. Sure, you can hire workers in third world countries for pennies a day, but the content they generate will look like it came from third world workers who work for pennies a day.  I’m sure if I paid amateur chefs $20/hr to craft each recipe with love and care, I could build the greatest recipe database the world has seen, but I’d need a few hundred thousand dollars in capital to have even the slightest hopes of a respectable database.

The mistakes of Mickey were out of laziness, however my mistakes were simply caused by inexperience in business.  So far, I have not had to “axe” anyone (in the literal sense anyway), but I’ve definitely burned through an unnecessary amount of time and money trying to solve a problem that quite possibly can’t be solved.  I have, however, made some changes to the site that will at least help impede the onslaught of awfulness.  First, authors can now link to a “credit URL” to allow users to see where the recipe came from.  I’ve instructed all my workers to make use of this feature if they want their recipes to count.  This way, even if there are some issues with the recipe, a user can click on a link to see the original source which will hopefully be more accurate.  This also gives me a way to proof read and correct recipes as time allows.  Second, I’ve decided to restructure my best team in a way that permits 2 workers to enter recipes, and one worker will be a full time proof reader.  This worker has been doing a fantastic job going through existing recipes and making sure everything is accurate, and letting me know when ingredients are missing from the database.  They’re already modifying dozens a day!  This means the quality of existing content is actually improving on a daily basis.

Hopefully, these tactics will lead to a more viable solution.  Otherwise, it may be time to move on to a business plan that doesn’t require a massive amount of content to get started; however, that’s a topic of a future post.