The Future of Work is Simpler and Scarier Than We Think
The season is last summer, 2018.
It’s about 3:30 AM. I’ve just finished reading 50 pages, starting at around 2:45 AM earlier that morning.
Over the preceding months during that time, from 3:30 onward to about 4:30, I focused on learning the skills of Python in an effort to learn what the hell Data Science, machine learning, and all that “jargon” that’s been thrown around in every latest article about the future of work.
Sometime in May of 2018 the latest Harvard Business Review appeared in the mail, and I turned it to one about the recently-turned public company Stitch Fix titled “How Stitch Fix Turned Personal Style Into a Data Science Problem”. Like many people, I’ve ready previous articles in HBR about Data Science being the sexiest job of the 21st Century and have seen the explosion of programs online centered around providing an approachable and accessible education on data science, such as the University of Michigan’s Online Data Science Masters.
Despite having read these articles and seeing the programs rise from obscurity, everything did not click for me in terms of the practicality of applying these data science tools to real-world business problems until I read the Stitch Fix article.
Data science isn’t woven into our culture; it is our culture. We started with it at the heart of the business, rather than adding it to a traditional organizational structure, and built the company’s algorithms around our clients and their needs. We employ more than 80 data scientists, the majority of whom have PhDs in quantitative fields such as math, neuroscience, statistics, and astrophysics. Data science reports directly to me, and Stitch Fix wouldn’t exist without data science. It’s that simple. — Katrina Lake, HBR, May-June 2018
What exactly is machine learning? How do data scientists work? How does personalization (the Stitch Fix way of doing business) work in the real world? I sought to find answers to these (if Data Science was indeed going to be as big of a deal in the 21st Century as people like HBR and Katrina Lake had implied it would, and if I intended to be a responsible and relevant business leader who could adapt to such a future, I sought out to learn a bit of the craft on my own.
This article isn’t about how I sought to and learned to code Python.
Such an article would just be a representation of the physical manifestation of the execution of using the plethora of online resources easily found through a Google search and applying the time it takes to actually learn, fail, succeed at such a task. If you want to learn how to code Python, I suggest you start with Udemy, Udacity, or Coursera, or Dataquest. I enjoyed the Coursera approach through U of Michigan initially as it took a more traditional academic approach and my learning was much faster than the to-the-point Udemy approach, but I found I needed the to-the-point approach at a certain skill level offered by Udemy to apply tactics vs learning the academic side. In short — explore all the options and see what works best for you.
Indeed, this article is about what I learned about The Future of Work through learning to code and learning about Data Science. The vast mis-conception I believe that is held in today’s business culture, and one that needs to be completely revered and requires extensive education for business executives to understand, is this:
The future of work isn’t an all-encompassing AI machine like Watson from IBM which will supposedly decimate a large percentage of white collar jobs. No,in fact it’s much more simple than that.
What I learned in learning to code, is that the future of work and the future of job automation is actually in the power of more simple, straightforward coded applications, or what I call Micro-Applications.
Micro-Applications are potent, powerful tools that are actually just an accumulation of just, say, 100 lines of code or so, that can do an incredible amount of work. These are very simple tools that can be highly leveraged by non-technical team members within an organization to automate somewhat complicated or mundane workflows to an incredible degree.
I stumbled upon this phenomenon when discussing my new-found coding skills with some work friends recently. We were discussing one particular workflow that someone on the finance team was executing on a weekly basis, which was essentially some data comparisons of two large data-sets.
So, for example, in one data set there would be entry “alpha one” and in the second data set there would be the entry “Alpha 1”. Or, for example, there would be “Beta 0” in one and “0 beta” in another. The corresponding data in each data set tagged to that specific entry needed to be compared for analysis purposes. So, instead of being allowed to use the all-powerful “v-lookup” tool in Excel where these entries need to be the same, the comparisons needed to be done manually because they were different.
This particular team member took quite a significant amount of time each week to do this work, resulting in an opportunity cost to the business since the skill set of this particular team member would have allowed him to work on much more complex and thought provoking work, but this work needed to be executed weekly for later analysis, and he was tasked to do it.
I therefore set out to create a tool that might alleviate the burden. I heard of a Python library called FuzzyMatching where two strings of text, so here “Alpha 1” and “alpha one”, are compared for what is called a match score.
The code below, which I wrote in Jupyter Notebook, outlines the process for those so inclined to read. In essence, it takes one list of presidential names in list A and finds the “best” match in the other list (list B). Thus, the name “Obama” was matched as the best fit to “B. Obama” in the other list through the fuzzy matching python module…so think of the output as at two column CSV where column A is the original text and column B is the best match text from file B. See “Obama: B. Obama” or “Trump: D. Trump” key reference pairs below, which is in essence a text version of the two-column csv.
Simple solutions to complex problems that have big impacts on peoples lives are the beauty of writing these snippets of code. The art of finding the solution in the least complex code possible is the magic of the process of creating it.
This piece of code, simple enough to execute by a non-technical individual (can be taught in an hour), results in hours saved in such a task, which can easily be applied to others use-cases with comparable requirements (i.e. data comparisons).
What struck me about this situation, where I was able to spend 2 hours to write code (probably longer than needed given my novice expertise) that saved approximately 6 hours a week, was how incredibly straight forward that solution was to build and how little time I had spent learning coding to build it. After only having learned the craft of Python for a few months, I had applied the skills to a real business problem to achieve significant ROI on a not-very-uncommon problem (comparing two similar but not exact data-sets).
Implications on the Future of Work
The problem with the current thinking about the future of work is that it is too complex, too high-level, and, consequently, too nebulous of most individuals to wrap their head around. If you put “Future of Work” into the Harvard Business Review website search engine, a case study on how Artificial Intelligence will impact the future of work is returned as the top 5 results or so. This is a mistake.
In fact, the future of work, as it relates to technology, has much more to do with the simple application of basic code to automate much of the basic technical work that has been created in our economy as a consequence of Work Tech 1.0 (manipulating spreadsheets, data entry, etc).
Work Tech 2.0 is the application of simple code blocks that connect to these applications (spreadsheets or data destinations) to automate specific work functions in order to alleviate the burden of the mundane on team members. The reason we have yet to see this happen, and the reason companies still employ team members to execute a lot of the work that can and should be automated with today’s tools, has to do with the marketplace of engineering talent.
There simply is no incentive for those with the skill-sets needed to solve a problem I mentioned above while the opportunity cost is employment at a large technology firm with big salaries. The economy is built on efficient application of resources, and in a market where engineering resources are scarce, they will be applied to larger problems unsolvable my manual work at a micro-scale. In short — people want to build Domo, not Micro-Apps.
The recent change however is how proliferated coding is becoming with new platforms such as Udemy, Coursera, and other tools, which is a relatively recent phenomenon. Another shift in the market is the rise of Python, which is a more approachable language than other coding languages, hence its rise in popularity verses alternatives.
As an economy, we could be on the precipice of a moment when supply of engineers will begin to exceed demand (maybe….). WIRED hinted at this phenomenon when it wrote “The Next Big Blue-Collar Job is Coding.” With the rise of accessible coding comes the rise of accessible coding jobs, and such jobs could be, not necessarily being the Chief Algorithms Office for Stitch Fix, but the guy that automates a workflow for an inventory planner at a company and saves 5 hours a week with 100 lines of code.
Economic Challenges
The real point here, or the real challenge, is how this impacts the economy. Will the same thing that happened to manufacturing jobs in the 20th century also happen to office jobs in the 21st century via small bits of python code eliminating redundancies across organizations in a simple, bespoke manner. All the time we spent re-training folks displaced by manufacturing automation could, again, become an outdated effort with outbreak of spreadsheet automation.
This economic shift is both a challenge and opportunity. It is an opportunity for business to see yet more untapped potential for efficiency and profitability improvements, which can be returned back to employees in terms of better benefits, higher pay, or to customers in the form of lower prices.
Limits
These micro-applications can’t be prioritized, designed, or implemented, however, without the management teams of companies understanding simple code capabilities and how relatively small investments can lead to large efficiency gains. The costs of these investments will continue to decline as the supply of engineers (even novice ones) goes up through the help of Galvanize and others (engineers for hire on Fiverr anyone?). Part of the slow adoption of micro-applications likely has to do with how little is understood by senior management on the power of simple code. Perhaps executive education courses at Harvard should focus here with, not teach code, but teaching what it can do.
Conclusively, however, there is a large, looming change for the workforce that probably feels the least threatened from automation, the person churning out weekly be-spoke reports for his or her boss every week that 100-lines of code can do in seconds. Perhaps a few courses on Udemy (company funded) on coding could help future proof an organization for the sake of employees as well as the company.
