Startup Idea: The Future of Digital Work

future of work.jpg

I’ve been giving my startup ideas away because I’m too focused on YesGraph to work on other ideas. I try to describe the idea enough to give you a head start in taking on the problem. If you like it, share it, try YesGraph, and subscribe here to get future posts.

This latest idea has been brewing for a while. In short, I think we can change how we get work done using computers. To set the stage, here are a few trends that feed into the idea.

Trend 1: Technology as Substitutes or Complements

The frequent prediction that robots kill jobs misses the fact that not all technology is the same. Technology means doing more with less. This can mean people can get more done, or it can mean the technology replaces the people.

For example, a surgical robot makes a surgeon richer, not poorer. She isn’t replaced by the machine, but is made better by it. An ATM is a replacement for many rudimentary tasks at a bank, and so it replaces labor.

Replacing labor is temporarily disruptive. Eventually it frees people to work on more valuable things. This will never stop because human desire is limitless. [1]

The important lesson here is that complements make people richer without the disruption of replacement technology. Technology complements rock.

Trend 2: Labor Patterns and Income Distributions

A brief history lesson: we’ve seen rampant expansion of information technology over the last few decades. In the 90s, saw an explosion of productivity. Then globalization took over as a dominant trend, which dramatically increased the wealth of developing countries. This has left real wages largely stagnant for most people in developed countries.

Tyler Cowen just came out with a book: “Average is Over”. Read it. The idea is that high skilled labor is getting incredibly well compensated. Humans with machine complements are valuable and getting more so. Less skilled labor is getting more plentiful, including service jobs to serve the growing rich.

This book is a far more nuanced take on “the 1%” vs “the 99%”. The best part is that it avoids any premature judgement before fully exploring the ideas and their implications. Emotion gets in the way for most people before they even know what is actually happening.

What does online work look like right now? For the most part, you have incredibly skilled people like software engineers paid over $100K a year. On the low end, you have Mechanical Turk getting jobs done at $0.01 per task. What about in between?

Trend 3: Artificial Intelligence is Good at Only Some Things

Robots are incredibly dumb. They lack any awareness of their environment. They lack any way to represent common sense thinking. We don’t know how human brains work, so we can’t copy the only example we have of an intelligent machine.

But there are some things computers are incredibly good at. They can power through massive amounts of data quickly. That data needs to be well organized, but then the possibilities are almost endless.

Humans are incredibly good at some thing compared to computers. Where computer lack, we shine. Common sense reasoning and pattern recognition are how our brains work.

Computers can be much more powerful if they had humans as complements, to work together.

Another way to make technology smarter is to build intelligent systems in aggregate. The best example here is Kiva Systems. Kiva replaced incredibly difficult physical labor in shipping distribution centers with a system of dumb sensors and robots in an engineered environment. The humans in the system do the hardest parts: recognizing objects and handling them from one bin to another. No part of the system is that smart, but the overall system was worth close to $1B when Amazon bought it.

Trend 4: Mobile is Good Enough

The problem with smartphones and tablets is that they look like toys compared to desktops. Desktops can multitask better and have higher resolution inputs, faster too. Processing power is far better on desktop.

But this isn’t going to last. While a machine that needs a battery is inherently limited, the practical difference for many tasks is very small.

For example, text messaging on a phone isn’t much worse than on a desktop computer. It is easier to carry on multiple conversations on desktop, but that is a small difference.

Some tasks, like working with a spreadsheet, are so much better on a desktop that their replacement to get the job done on a mobile device just won’t look the same. Over the long term, the reason why you use a spreadsheet will be replaced with a tool dedicated to the task. Plus that tool is probably going to be faster at that specific job than a spreadsheet on a desktop.

Trend 5: Digital & Mobile, but Physical & Local

Many startups embrace the idea of the physical world coming online. This is the most direct response to the smart phone. If you have the internet in your pocket, what jobs can you do?

There are a few things to unpack here. First, all the labor is unskilled. As a result, wages aren’t particularly high.

The driving market is expanding into the car purchasing market. When a capital expenditure is replaced with service labor, you can expect wages of that labor to rise. This is the opposite of trends for pretty much every other industry, with falling or stagnant service wages. So driving might be the exception to labor trends because of this pent up demand.

The second thing to unpack is that all these applications barely involve using the supercomputer in your pocket. They almost all combine a messaging system with a localization system. That’s it.

We haven’t even come close to unlocking what it means to have a billion people on the planet connected constantly to the internet through computers that are more powerful than desktop machines from just 10 years ago. We’re hung up about the physical world, and not focusing on the idea that everyone has a powerful computer.

Trend 6: Online Education Suffers Low Completion Rates

Massively Open Online Classrooms (MOOCs) have unlocked an incredible amount of value in education. Our universities are good for batch learning over years with theory-heavy coursework. Putting this content online has unlocked an incredible amount of opportunity for people who didn’t have access to good education.

Completion rates of these courses are notoriously bad. Some argue that this is a rich-world problem. People that already have access to a university education don’t feel compelled to stick is through.

I think this response masks something more nuanced: the content and format copied from universities doesn’t actually match what people need. When I need to learn something, the ideal content, if I can find it, is typically no longer than a chapter in a book. 50 hours of lectures is a format that doesn’t match my needs.

Now that we’re moving online, I wonder how much more will be unlocked just from flexibility of the platform. You can draw an analogy to YouTube. With 30 minute TV shows and 90 minute movies representing 99% of video consumption in 1999, why would people want 3 minute movies? It turns out that the broadcast & film industry business models masks the huge demand for something different.

Universities are also biased against practical knowledge and vocational education. They don’t like industry to the point of believing business corrupts the very ideals of higher education. That is insane when you consider how much time we spend at work and how much better our lives are when we get better at our jobs. It is surprisingly common ivory tower thinking.

I think we’re going to see a lot of experimentation in teaching people the thing they need to know, right when they need it. The demand from people who want to do their job well is going to push this the hardest.

Putting It All Together

So we have phones in everyone’s pocket that are just as good as desktops, but no one is working on them yet. We have massive potential in online education, with the most potential in a format that fits learning something while standing in line at Starbucks. We have amazing machines that desperately need humans for some tasks. Humans that work with complementing machines get richer.

My idea is to build a platform for digital work that you can do on your smart phone. When you pick apart that sentence, you begin to see the power of the idea. So many problems can be solved by such digital work on your phone.

You tap a pool of people to get a job done. How do you coordinate that kind of project? Project management is digital work.

What if people are in different countries and speak different languages? Translation is digital work.

What if you don’t know how to get the job done? Education is digital work. Your guidance counselor helping you navigate education and work possibilities… that is digital work.

How do you recruit and filter a massive pool of people to get on the platform? Recruiting is interviewing and can be digital work.

If you want a job done, how do you tap the platform? Concierge is digital work. Customer support is digital work.

Ok, but how does the product work?

People put in jobs into the system. The jobs get farmed out to the labor force on the platform. The definition, devision, and coordination can be done in part with software, but with humans helping direct things where machines have trouble.

There are standard tools that can help: voice, messaging, googling & other research tools. Where there are focused tasks, the platform could make job-cards, which would be mini apps to get a specific job done.

For example, translating a document is probably best done with something besides google docs. Suggested machine translations, built in dictionaries and other language tools, and consensus editing are all very focused features of a mini translation app. Defining and building such mini apps is also digital work.

For another example, a list view of items doesn’t make sense when looking at a hierarchy. So if you’re not looking at a photo timeline but a file tree, the interface should match the task. Or, instead of looking through either you might be better served with faceted search. Tuning the software interface to the task will make people more productive.

Implicit is defining a stream of data and a work flow. This fits best for recurring tasks, rather than one off needs. Too many work platforms are trying to solve one off tasks that are, by definition, a small part of labor and harder to systematize.

Build It Using It

To help illustrate the idea, let’s walk through a few tasks you might need to get done in order to actually build it.

You want to wrangle a few test workers. Let’s assume you’ve manually recruited a few people to act as testers. To get more, you would create a task around each step of the process and put it out on the platform:

Post X to craigslist.
Look for X in the response emails.
Schedule a time to talk to the good candidates on the phone.
Ask these questions on a phone call or email thread with the candidate.

Your team might still have tasks that are too hard to explain to the platform. This is where specialization comes in. Recruiters often look for specific things, and hiring managers have a shorthand of how to communicate those things to the recruiter. There is nothing about the platform that would exclude specialists, though skills matching becomes a challenge.

Now you have some people on the platform that you’ve never met. This is great because their feedback is going to be essential to making the experience great. Everyone is using an app or mobile web site to interact with this alpha version. After every job they do, you have another job asking them and paying them to give their feedback. This way you identify the bottlenecks from the supply side right from the start. You could do the same for the recruiting process itself and refine your filters.

On the demand side, you want to find companies that are willing to test this out. Let’s say you want to target startups because they understand new technology, but you want them to have some size and some money. Create a task to kick this off, requesting a structured feed of TechCrunch posts to extract the company, the stage, and any people mentioned. Create another task to research who works at that company using Google, LinkedIn, and the company about-page. Find people that work in sales, and send an email template saying they should check out the platform to do market research. Find people that manage the office, and send them an email template about the million little tasks the platform could get done. When someone replies, if there are no questions, create a task for the platform to schedule a time to talk on the phone for a sales call.

Record how you’re using the platform, because these guides will help others use it, to understand the power. This a problem with concierge services: people don’t know how much they can offload, how to thinking about delegation.

Eventually you could train people be self sufficient at sales development research. The strategy of the initial targeting is something someone trained in sales can create and push out to the platform. With just a bit of training, someone can take instruction like “target early startups” and come up with the whole lead research pipeline I described. Just in time training makes specialists out of generalists very quickly. We find this uncomfortable because our own education took years, but most people don’t use most of what they learn directly.

Sales is a complicated process, but there are some standard questions someone might ask before you have a call. This is another call for specialization, where people on the platform learn more about your product to be able to provide customer support and answer basic questions. Part of the work here is creating and formatting documentation in a way that is digestible to learn. This is a great example of the kind of online education that would help someone learn a new job. Other bits of knowledge could be taught too, like how to treat customers during support requests or how to aggregate feedback in a way that is useful for a product team.

All this work is already being done within companies. Some tasks are already outsourced, and this platform would just make it easier. It would formalize many processes, which would implicitly benefit many companies.

It is really easy to come up with more examples of how to use the platform. Create pilot programs with these examples and find companies that would be willing to try it out:

Then you have a whole class of tasks that individuals might want to try on the platform. Travel planning, calling a business, restaurant research & reservations. If your digital personal assistant has done it, then this platform can do it.

I want to stress that this section of the post could far long. Once you start thinking about the processes you’d like to outsource, the world opens up in possibilities.

Living Wage + Automation

You could simplify the platform by making a minimum hourly rate. The app would track time in the app and charge appropriately. That fixed rate could be something higher than the median wage in the US, let’s say $20/hr. That is kind of high, which will help the platform avoid degrading to the least common denominator. That is why Amazon Turk is a bad fit for most of these tasks. They were so focused on scale to aid automation that they anchored the price per task to a point where their supply of labor is constricted.

What about tasks where you can’t create $20/hr of value? If they only created $10/hr then you could make it work on the platform by adding automation and a technological complement to the task. This is a way of defining what software to write. If something is done very frequently and could be made more efficient, write software for it. If in your market research you find a certain class of tasks isn’t done, figure out how to write software to make it better.

This could be an open software marketplace too. A good example of this is creating blog themes. There is a whole marketplace of rudimentary software development around customizing how a blog works. The output is often just a basic website and not even a blog. This works because a platform like Wordpress has created a way for a little bit of software to make an impact on a client’s work.

Similarly, the customized task cards to get jobs done more efficiently could open up new tasks to get done efficiently on the platform. You could use the platform itself to recruit and pay for people to develop cards for specific tasks. You could create a templating language and standard data pipes to make the job easier.

Sometimes you don’t need a custom interface, but software to learn and process some data. Maybe your market research has sub tasks where software can help, like scraping sites, and classifying the kind of data there. Maybe the software is general but needs a bit of training data. So the software feeds back tasks around defining training data to create an algorithm to take on the broader tasks efficient. Like Kiva Systems, the aggregate system gets the job done intelligently while each agent is relatively unaware of how that works.

The software developers could even get paid whenever their software is used. The platform knows exactly how much value is created from usage, and it could siphon a proportionate amount to developers. This turns fixed development into recurring revenue, which is incredibly motivating for developers.

The hourly wage being high is important here because lower cost labor displaces the work done on automation. Human labor is a local maximum of efficiency, and you want to force the automation onto the platform. A price floor is a great forcing function if it is implemented correctly.

Just Dive In

In the past, I’ve asked people to email me if they like the idea. I wanted to help generate of kernel of interest. Instead, I’d like the community to organize itself. So head over to HackerNews if you don’t mind the trolls.

If you want more ideas, subscribe to this blog. You should also check out YesGraph. Tell all your friends about us, and subscribe here to get future posts.

[1] The jobs that are left are better either way. Bank tellers work on more complex and valuable problems that an ATM can’t handle. The technology frees people to work on more valuable things even when it replaces labor. This post isn’t about short term labor market disruptions from technology, but the fixed-pie thinking is so common that we need to address it.

 
87
Kudos
 
87
Kudos

Now read this

The Story Behind @Dropbox’s 58,449 Retweets.

Before I started YesGraph, I was helping Dropbox on their growth team. I ran a quick campaign on Twitter for Dropbox that produced the most successful single tweet ever published from the @dropbox twitter account. Here is a quick story... Continue →