How to use Mechanical Turk to generate ideas focused around a specific niche or speciality.
artificial intelligence.
Mechanical Turk is a micro worker platform created by Amazon in 2005, originally the idea of Jeff Bezos. He predicted there was a business to be built around connecting micro workers with those looking to get small research tasks completed. Bezos called the phenomenon “artificial artificial intelligence.”
By 2011 over 500,000 people had registered to be "Turkers" and perform micro tasks.
The Mechanical Turk name comes from an 18th century chess-playing device. Challengers competed against it, believing they were playing against an impressive automated machine. In fact, the machine was an elaborate hoax. It included a space to hide a human player, who could control the movements of "the Turk," using human skill to beat the challengers instead of an automated process.
The beauty of Mechanical Turk is we can pay a little bit of money, for example $20, and in return startup ideas will simply pop into existence!
I know this, because that's where all of Nugget's 3,700+ ideas came from.
For Nugget, I've commissioned generalist ideas. Which is to say, anyone can submit an idea about anything.
But, for you, we can be much smarter than that. We can commission ideas that are tightly focused around a market or niche that you really like. There's no point commissioning generalist ideas, because you already have access to all of Nugget's!
I like this quite about Mturk from Nugget member @eden:
The MTurk research was by far my favorite -- it's so easy and quick to take a broad pulse, and the results are always surprising and informative!
At first I thought Mechanical Turk would only consist of workers in call centers and such. But nothing could be further from the truth.
I've been highly impressed at the depth and breadth of people who have submitted ideas, both by their financial status and also their expertise and business domain.
Turns out, many people participate in Mechanical Turk for intrinsic rewards along with financial rewards. They just find it interesting to do.
I've had submissions from folks that own law firms, farmers, CPAs, chefs, developers, hedge fund managers, you name it.
- Step #1: Create a Mechanical Turk requester account
- Step #2: Create a HIT
- Step #3: Synthesize HIT results
1) Go to https://requester.mturk.com/
2) Click "Create an Account"
3) Follow instructions
1) Take some time to choose a niche or market that you really like and would like to explore. After you have selected your market, take note and continue. Example niches or markets might be:
- Chess Players
- Photographers
- Designers
- HR Professionals
- Ruby Developers
- Etc.
It helps on many levels if you can find a market to serve that you really care about. If you have a hard time coming up with one, here are a few ideas to help you out:
- Are you subscribed to any subreddits of interest? What are the niches and markets? Would you like to serve any of those?
- What other services or products do you use on a regular basis? Would it be fun to build tools for creators of any of those?
- Look through your Amazon order history. Is there a specific type of product or software you buy? Who makes it? For example, perhaps you regularly buy games created by Indi Game Developers. In which case, you could consider building tools for Indi Game Developers.
3) Click "Create > New Project"
4) Click "Survey"
5) Click "Create Project"
6) Enter a subject that speaks to your target niche or market. For example, something like "Calling all Chess Players" or "Calling all HR professionals".
Keep in mind, this is only going to work with relatively large niches. So, something like "one-armed chess players" will not get any results!
7) Enter a description like "What software solution could make your life easier as an HR Professional?" Note, switch out HR Professions for your niche or market.
8) Start off by running a batch of 10 assignments for $0.75. Here is how to do that.
9) Specify worker requirements. I usually require that, in the past, workers have been approved 80%+ of the time. I often limit the HIT to only show to American workers. I always hide the HIT from anyone who is not qualified to see it.
10) Click design layout
11) Click source
12) Replace the source with the below html
<div style="font-family: Helvetica, Arial, Sans-Serif;">
<div style="background: #F7F2DA; padding: 5px 20px 10px 20px; margin-bottom: 15px;">
<p><strong>What is your biggest pain point as an [YOUR NICHE]?</strong></p>
<ul>
<li><strong><strong>If you could wave a magic wand to solve your [YOUR NICHE] problems, what software would pop into existence? How would it make things better?</strong></strong></li>
</ul>
</div>
<div style="padding: 5px 20px 10px 20px; border-radius: 10px; ">
<p><strong><strong>PLEASE READ THESE <u>IMPORTANT</u> GUIDELINES</strong><br />
<em>Unfortunately, <u>we will reject submissions</u> that do not follow them.</em></strong></p>
<ul>
<li>Only do this HIT if you are an [YOUR NICHE]!</li>
<li>Only mention problems related to [YOUR NICHE]!</li>
<li>Please be as detailed and specific as possible!</li>
</ul>
</div>
<p><textarea cols="80" name="answer" rows="7"></textarea></p>
</div>
13) Replace [YOUR-NICHE] in the above html with your market or niche. In this case, I switched out [YOUR-NICHE] with "HR Professional"
14) Click "Source" again and double check that it took.
15) Click "Preview and Finish"
16) Click "Finish"
17) Click "Publish Batch"
18) Click "Next" and verify that all the details look as expected
19) Click "Purchase & Publish"
20) Click the "Manage" tab
21) Watch the results come in. Depending on how popular your niche is, your results can take anywhere from 1-8 hours to complete.
22) You can also click the "Results" button and keep refreshing the page.
After you get back your results, analyze them and break them into common problems with counts. For example, if we were running a HIT for HR professionals, our final synthesized list of problems might look something like this.
(12) retaining staff is difficult
(5) motivating staff is difficult
(2) staff scheduling is difficult
In this way, we can survey 50+ people in a single niche and get a general sense of the core problems that folks in that niche are facing.
In the example above, we specified results from 10 people. Ultimately we will need more data than that. If you're up for it, I recommend you publish again and specify results from another 40 people, giving you 50 people's responses to synthesize.
In some cases, I have pushed to 75 or 100 if I really want to be sure I have spotted a common trend.
Here are some pro tips about this whole Mechanical Turk thing.
There will be HITs that you don't like or that are not filled out very well. In general, air on the side of NOT rejecting unless:
- It is obvious that the HIT is spam and not related to your question
- The HIT is blank
- The HIT only contains a few words something like "I want a better phone"
If people have made a genuine effort to answer as best they can, it is bad practice to reject the HIT, and will quickly get you black listed amongst workers on the worker discussion sites where they rate requesters.
- I find the best starter price is $0.75, but you may be able to go up/down to test if you get better results.
- Don't expect 100% of responses to be good! I generally find about 50%-75% of responses end up being useful in some way.
- If you specify higher approval ratings (i.e. 95%+), then you will have to pay more money per HIT in order to get those better people's attention.
- In your dashboard if it says the worker has a 0% approval rate. It means that you have never approved them before and is not related to the quality factor you specified on the main entry form (confusing, I know).
- The time of day you publish makes a difference. If you specify US Only and publish at 3am US time, you won't get any results until the next US day.
- If you commission a batch of 10 HITs and then commission a batch of another 40 HITs, then folks from the first 10 will get a chance to answer in the second 40. If you want to be absolutely sure you're getting unique answers, then commission a batch of 50 from the get-go. In other words, as soon as you have commissioned one batch, there will always be a (very small) chance that the same person will answer a question in a future batch.