Leveraging the crowd to accomplish large-scale projects
Andrew: My guest today is co-founder of CrowdSource.com, and he also owns a domain name business. Ryan Noble, welcome to the program.
Ryan: Thank you. I’m happy to be here.
Andrew: Ryan, let’s start by talking about CrowdSource.com. I know that’s turned into a big business for you. What exactly is it?
Ryan: Sure. Around 2007, Stephanie Leffler, my co-founder of CrowdSource, and myself were working in the publishing space, and were in need of big data and many reference points and research based on topics. We had no real way of doing it at scale, so we came across Amazon’s Mechanical Turk and realized the power of the crowd. At that point, upon using it, it was very helpful, but it was also very much in its infancy. We found that it was hard to get data in and hard to get data out. Hard to establish the right workers, to get them the right work at the right time. We decided to build the platform from the ground up that could help to basically fulfill tasks, and even more complex jobs and workflows at a tremendous scale. From there, we have moved on ever since.
Andrew: I’ve used Amazon Mechanical Turk, and what you said there is exactly right. I almost feel like you have to be a programmer to use it. You have to create these templates and CSV files with certain spots in them, and that sort of thing, and it’s very complicated. CrowdSource.com doesn’t run on Turk, right? It’s just very similar to it.
Ryan: Right. We are actually partnered with Amazon and oDesk and several other of the large employment aggregators or employee aggregators, and they can on-board onto our platform and use our workflows to perform work.
Andrew: Got it. If I wanted to hire people to do work, but I don’t want to deal with the challenges of figuring out how to use, say, Amazon Mechanical Turk, then you guys have a better way for me to interface, essentially.
Ryan: Right. We basically become experts at recruiting, testing, qualifying, onboarding workforces. Then based on worker qualifications, enable them to have access to specific types of work, so that it’s basically quality assurance from a worker’s side that’s preemptive, versus something that’s reactive on something like Mechanical Turk.
Andrew: Can you give me some examples of the types of jobs I can run through CrowdSource.com?
Ryan: Sure. I mean, we really range from a variety of tasks. Some are what you’d consider microtasks, which are community moderation, which can be viewing images and making sure that images or videos are appropriate for posting to some of the larger social media sites and video sites. Then all the way up to article writing, product descriptions, and then as sophisticated as white papers and things like that. The majority of our work is done at a tremendous amount of scale, so thousands of articles or buying guides for retailers.
Large retailers, for instance, have to onboard 250 thousand new product SKUs per quarter, and the data they receive from manufacturers is often incomplete, often lacks attribute data, often doesn’t reconcile appropriately with the big box retailers or e-tailers, structure for their e-commerce platforms. Then all the data is duplicated as well, so there are many different components to it.
We’ve developed, and this is using the products as an example, but from taking data from a manufacturer and 250 thousand product SKUs for instance, and then running it through the entire process end-to-end, from tagging the right attributes, making sure the image is accurate, and then rewriting product descriptions and enhancing them to make sure that they’re unique compared to the 10 thousand other retailers that may have received the same type of data.
Andrew: Interesting. Let’s say I have 250 thousand SKUs and multiple tasks per each of those. We’re talking about potentially over a million tasks right there that you can handle.
Ryan: Exactly, and very quickly. It’s almost an assembly line, if you can imagine.
Ryan: We’ve created the workflows for specific types of work such as product descriptions and things that we’ve seen as being large opportunities. It’ll start at the very beginning of the process, and then it gets moved along. Throughout the process, it’s quality-assured along the way, both through AI and human review. The workflows are such that it doesn’t have to come back to the retailer each step, it just goes through that assembly line. They’re, at the end, assured that they have received the quality output that they were hoping to get.
Andrew: They don’t have to see the sausage being made. You guys handle that.
Andrew: It comes out as a finished product for them.
Andrew: Interesting. This has become, obviously, a successful business for you. I see that Highland Capital Partners and D of J are investors in this business.
Ryan: Exactly. Yes.
Andrew: You also own Write.com and Transcribe.com. These are part of that same business, right?
Ryan: Right. They’re both sister companies of CrowdSource. We use those domains as both recruiting vehicles for transcriptionists and writers, but also to market to those specific verticals. We consider them some of the major ones for what we’re trying to attack online.
Andrew: I’m trying to think here, because I know a lot of domain name investors will go get articles written for their sites that they’re creating and that kind of thing. Would Write.com or CrowdSource.com – is that a place where they can do that?
Ryan: It depends on the scale. We can do one-off articles, but we’ve really shied away from that now that we have more scale involved. We work with mainly Fortune 500 companies or the bigger internet publishers and e-tailers.
Andrew: Got it. Okay. That makes sense. Now, let’s talk about these domain names. Obviously CrowdSource.com, Write.com and Transcribe.com are all excellent domain names. How have these helped you grow the business?
Ryan: Sure. We consider them relative category killers in terms of the markets that we’re trying to get.
Ryan: There are a couple of others that would obviously be considered the same in each vertical. It’s just credibility, I think. I think that when we’re recruiting, there are a lot of companies that are out trying to recruit for writers or transcriptionists. I think the brand alone, whether it’s in ad words or in organic or whatever, I think it just adds credibility.
Andrew: Mm-hmm. It’s interesting that you said Transcribe.com, that you use that to attract people to do the transcription. I guess that’s how you say it.
Andrew: You’ve almost created – and obviously listeners can go look at this – but you’ve taken a segment of your business that’s important and then gotten a category killer domain name for that. You kind of use that as a separate lead gen, but for both sides of the funnel, the workers and the users. I don’t know what you call them.
Ryan: You’re quick to comprehend that. Exactly. It’s both a recruitment vehicle and also for lead gen. The transactions can take place on those sides, but then the actual work will be pushed through our scalable workforce on our CrowdSource platform.
Andrew: You’ve really simplified it. I’m looking at Transcribe.com right now, and it’s pricing right up there – turnaround time based on minutes and that sort of thing. Does this help you from a search engine perspective, having this one site dedicated to a particular topic as opposed to being buried on CrowdSource.com?
Ryan: I imagine so. Search engines and organic are obviously a moving target. I think any time you can create more of a niche topic or vertical, you’ll be placed in that community online in search engines’ minds. I do think we have a boost in organic relevance. That’s helped us for sure.
Andrew: Got it. Great. I find all of this fascinating, but what I’ve found really fascinating about what you guys have done in your business, is that you have a background in domain names. You have a domain name business, and you’re applying this sort of crowdsourcing to it. So first of all, tell us about that domain name business. The ClickableNames business.
Ryan: Sure. So in 2006, my business partner and I, well, we founded a company called MonsterCommerce in 2000, and ended up selling it to Network Solutions in 2006. We each worked there for 18 months following the acquisition. We learned a lot about domain names there. After 18 months, we took six weeks off and began writing business plans. When we settled on one, we stumbled across Mechanical Turk. Basically, we ran hundreds of thousands of different tasks through Mechanical Turk with no intention of using the outputs. More or less, it was an experiment. That’s basically where Mechanical Turk was at the time.
Using that domain expertise that we had from Network Solutions, we built basic software using the common hundred data points that a lot of software domainers use to detect dropping domains or domains that haven’t been registered. We began registering using that software. Then as we used Mechanical Turk and built our own platform, we differentiated ourselves from using software.
Clearly, we used software to reduce the number of names, like by using negatives such as trademark terms and bad language and things like that. By limiting the number of names, we could then send the names through real humans because of the subjective nature of domain names. A lot of domains were dropping through the cracks, because software can’t detect that subjective nature, but humans can.
We built a workflow that includes multiple humans reviewing names throughout the process, and eventually servicing, daily, names that we find of extreme value or of value that a lot of the domainers in the space are unable to detect with straight software.
Andrew: Are these primarily expired domain names, then?
Ryan: I would say they are expired or dropping names. A lot of times, we will, for either some of our clients, and even internally, we will say, okay, this is the vertical we are trying to reach. We’ll write a fairly comprehensive description of the type of business that we are looking to use. Before we bought the domain CrowdSource, we sent the crowd out to use Ideation to find different ideas for the actual business. It surfaced that ScalableWorkforce.com was available for general registration, so we ended up buying that, which we thought was a good name.
Ryan: It held its purpose at the time. Eventually, we upgraded to Crowdsource. The power of the crowd is pretty remarkable. When we create a contest like that, we’ll receive 20 thousand or so ideas, and then, again, if we’re using Mechanical Turk, it would be tough, because we wouldn’t have the multiple layers of workflows that we could send those domains through for filtering. We’ve built those workflows so that those 20 thousand will come in, but if we were trying to sort through those internally, it would take a long time. The power of the crowd and those workflows are able to sort through them and surface, of those 20 thousand or so, the most valuable.
Andrew: When I look at expired domain names, I use software, like you’re saying. I’ll filter based on how many top-level names that domain is registered in, how old they are, some keyword type stuff, some SEO data, that sort of thing. What you’re saying is you’re actually using some top-level filters to get rid of trademarks and such, and then you’re sending these through your workflows to all these workers, who then evaluate them on a number of different subjective measures.
Ryan: That’s exactly right. There are about a hundred data points that you’re aware of, everything from language, CPC, search volume - I mean, all the things that you’d expect – age and even category and word breakdown and so forth. We use those basic tools, then use the crowd to find the subjective words that software overlooks. We found that could be quite a few, as you can imagine. At the end of the day, domains are very subjective unless they’re generic keyword rich.
Andrew: Right. Are most of the ones you’re unearthing brandable domain names, then?
Ryan: In some cases. Then there are some that are generic, like category type names, that just have unique word breakdowns and engrams and things like that we’re detecting. It’s hit or miss in a lot of ways, but, yes, a lot of brandable names.
Andrew: Right. You know, I looked at ClickableNames, and I don’t know if all these domains were captured in that way, but something like Statster.com, for example. It’s a good domain name that I don’t think I would have picked up using software, but obviously would be great for a stats program or something like that.
Ryan: Right. That’s actually a good question. One of the things, and ClickableNames is a good example of that, but we have an R&D arm that’s separate from our core platform. I agree with you, I don’t think Statster’s a good name at all, truthfully.
Andrew: Well, no. I’m saying I think it’s pretty good, I just wouldn’t have captured it with my software.
Ryan: Oh, I’ve got you.
Andrew: I think it’s a good, brandable name.
Ryan: In the stats base, perhaps. We also, in an example of the R&D arm, we would buy names that are more brand-centric, if you will, and then we would use the crowd from a design perspective to design the logos for those domains. That was more of a use case, but we basically created a workflow and recruited designers, and then would send those types of domains that the workers would say are more brandable, that were good brands, and then send them off to be designed as part of the workflow. Then they come back and we’d post them to the site.
Andrew: That’s what I’m seeing, if you go to the ClickableNames.com website. A name like HigherLabs.com where there’s a logo around it, Spotro.com, which is obviously purely brandable, it doesn’t mean anything. Then you’re moving that through your workflow to create a brand around it that someone can buy.
Ryan: Right. We did so knowing that that wasn’t a business we necessarily wanted to get into, but from a design perspective, we wanted to understand from an R&D perspective what it would be like to onboard designers, and then add that portion of the workflow so that we could prove that concept. That’s what’s shown there.
Andrew: Right. I’m just looking here – WetDryVac.com, InsectRepellant.com, Straighteners.com – so you feel like you’ve proven that this method works, of uncovering these good domain names through human interaction as opposed to just the software and stats.
Ryan: It works. Yes.
Andrew: You guys are selling them primarily through the site, or you’re also selling them through third parties?
Ryan: We do both. We sell them through the site and third parties. Again, domaining isn’t a core competency of ours. We have one and a half, and sometimes just a half of a resource dedicated to doing renewals, managing which names we’re going to drop and doing the site sales. Clearly, when testing this type of thing, you also register a lot of horrible names along the way. [crosstalk]
Andrew: [laughing] Yeah.
Ryan: Clearly, there are a lot of drops along the way too.
Andrew: One other question around this - I think if I were working, and I saw a great domain name, that I’d just go ahead and register it before you could, right?
Andrew: How do you prevent your workers from registering domain names that they’re evaluating?
Ryan: Sure. There are two ways. One, we have multiple steps within the workflow. It happens from time to time, obviously, so we can just look at the analytics of where our workflows break down. The most specific and the most successful way we do it is to simply inject names that we think are good that are dropping, then we let them go through the workflows. If they are not detected, we train the workers where it wasn’t detected. If we find it was registered, we don’t have the same interests in mind from the workers, so we part ways in that case. It’s basically an admitting bogeys type of thing.
Andrew: They call that honeypot, or something like that.
Ryan: Honeypot, known answers, yeah, that type of thing. Exactly. It doesn’t happen very often. We also do, when a domain is found, and it’s of value, we know enough about domaining to be able to evaluate the value of it, and then the people involved within that process, we compensate for being a part of discovery. It’s rewards-based as well. There is incentive for them not to do that, but that doesn’t say that it keeps everyone from doing it.
Andrew: Right. It sounds like, though, you have a process for it. It runs almost on autopilot now, it sounds like. Half of a resource - this isn’t really a main focus for you going forward. It’s mostly on the CrowdSource business.
Ryan: That’s right. We do crazy R&D for a variety of types of tasks. It’s all about the differentiation from oDesk or Mechanical Turk. There are a lot of labor pools out there, but as I was saying earlier, it’s really difficult to get the work in. It’s difficult to create, in fact, in some cases you can’t create multi-step workflows. You don’t have visibility to the work that’s being created within the workflows. You can’t choose qualifications of the workers properly. The output is done at such scale that you need artificial intelligence, known answers, AB testing, having multiple workers doing the same work, making sure they have the same outcome. That’s the differentiator between platforms like ours and just labor marketplaces, if that makes sense.
Andrew: Yeah. I think anyone who has used oDesk or eLance to hire people to do things has discovered that. When I use Amazon Turk, even for simple things like capturing phone numbers or addresses, or checking for data on a website – it’s a lot of work just to get it set up. Then accuracy is an issue, unless you have a workflow in place for that. Then I found out once the hard way, that there are sites where the workers get together and say whether or not they feel the person that hired them is being fair judging their work, and that sort of thing. It’s a big, complicated process for someone who doesn’t do it every day.
Ryan: It definitely is. Actually, we’re partners with Mechanical Turk and oDesk and have a lot of respect for them. Our model is slightly different in that we do the workflows, but that really is the problem. Mechanical Turk was the pioneer of this, and clearly, when you’re defining a new space, it was heavily involved around microtasks. It almost had a negative stigma at first just because the creator of the task decided whether the worker was paid.
Ryan: There was a lot of spam. The pay was low. It has really over the years become more verticalized, and the types of work are becoming more complex, to where now it’s like the majority of our workers are making solid wages and performing tasks like writing for buying guides or whatever else. It’s not just microtasking like in 2004 or earlier than that when Mechanical Turk was founded.
Andrew: Interesting. Well, Ryan, this has been a very fascinating story. Obviously all my listeners are in the domain name business, so I don’t think people have heard this story before, and how you’ve created the Crowdsource platform, and also used that to kind of figure out which domain names are valuable. Ryan, thanks so much for being on the program and sharing your story.
Ryan: Hey Andrew, I really appreciate your time.
Andrew: My guest today has been Ryan Noble. He is the co-founder of Crowdsource.com. If you’ve enjoyed this program, please leave a positive review on iTunes, and of course you can find all previous episode of the Domain Name Wire podcast at DNW.com/podcasts, or on iTunes. Thanks for listening.
Announcer: Thanks for listening to the Domain Name Wire podcast. For the most up-to-date news and tips about the domain name business, visit us online at domainnamewire.com, or follow us on Twitter at DomainNameWire.