Make your data better – use the Turk

Need to improve your CRM data but too busy to do it yourself? Here's a trick that is cheap and easy. Amazon has a service called the Amazon Mechanical Turk marketplace. I'll refer to it as AMT for short. It is defined as a marketplace for work that requires human intelligence. More specifically, it is a simple way to deal with small tasks that are tedious and time consuming. When you have simple tasks that require a human involvement you upload your tasks/HITs (Human Intelligence Tasks) to the Turk and someone, somewhere will do them for an amount that you agree to pay per hit. In my real world example that I am doing as I write this I am using the AMT to classify my CRM account records by industry. It took me about a half hour to set this up and is costing me $.10 per record. Pretty cheap, right? Here are the steps I had to take to make this happen:

• Create an account with Amazon Mechanical Turk (this can use your existing Amazon.com account)
  
• 
  
  Design the web form that the person doing your HITs will use to do them (see below)
  
  
  
  
  • Note that my form gives very detailed instructions on what the person needs to do. In this example, I am showing the person a company name and a web URL and am asking the person to go to the URL, read about the company, write a description (in their own words) of what the company does and then select an industry from a list of industries which I coded into the form using basic HTML. Note that I asked them to write the description in their own words because this ensures that they actually spend time on each task and make sure they get it right. Plus, it makes spot checking the results easier for me.
  •  

• Export the CRM data from the CRM system. In my example I had to export company name, URL and account ID. I need account ID so that I will be able to update the records back into my CRM when the work is done
  
• 
  
  Import that CRM data into the Amazon Mechanical Turk
  
  • Note on the form screen shot that the ${AccountName} is a variable that will be replaced by my actual data. The worker will see the same form for each HIT but with a different Company Name and URL for each record from my CRM system
    
  
  
• Set a bid amount for each HIT
  
• 
  
  Publish the HITs and watch live on screen while people do them. Below is a screen shot of the job progress. It updates real time if I want to watch it but, of course, I don't need to. I can simply come back later and see how much has been done. From here I can review and export the completed HIT for re-import back into my CRM
  
  
  
  
  
  

As of right now my job of 1,614 records is 23% complete and I am paying the equivalent of $1.46 per hour based on how long it has taken so far. I set my job up so that multiple people can work on it at the same time and I bid $.10 per record so my total was 1,614 x .10 plus a ~$16 fee for Amazon. In total I paid $178.80 to get industries assigned to these 1,614 accounts. In reviewing the results including the descriptions I asked the worker(s) to write and the classifications they have chosen I can tell that I am getting very good work. Below you can see an example of a record where the worker correctly identified that the Dobla BV is a producer of chocolate and, thus, belonged in the Manufacturing – Food and Beverage industry category

What is fascinating is that I do not know, or care, if my worker(s) is/are housewives in Houston or college students in Calcutta. And, at the end of my job, I can spot check the results and accept or reject individual results.

Try it sometime and see what you think. I've done it three times now for various tasks and it has worked well each time. I have learned to include very detailed instructions (with an example) on my forms and I believe that is important.