The internet is bifurcating. And it is becoming clear there will be massive ramifications for how global citizens see the world and how companies operate within it.

The last decade was about bringing billions of individuals online. By the end of 2005 there were just over one billion internet users. By 2010, the number had surpassed the 2-billion mark. As we close 2019, the number of global internet users has surpassed four billion, just under 54 percent of the global population. In the early years of this transformation, the focus was on creating an internet that could be accessed in developing countries on developing networks. From Pranav Dixit reporting in Buzzfeed:

“The only times India came up during product discussions was customizing those products on slow and patchy internet networks in developing countries,” said a longtime Google executive who didn’t wish to be named. “What we think of as the ‘internet,’ even 10 years ago, was the American internet, and that is what everybody experienced.”

In recent years, there has been a move to create an internet increasingly customized to its users. As Caesar Sengupta wrote in a Google blog post in 2015:

“Our goal is to bring all Indians online — regardless of income, region, age, gender, or language — and as they come online, we want to make the Internet more relevant and useful for their needs.”

Again from Dixit:

“Over the last few years, Google has made its products available in more than a dozen Indian languages, reworked Android keyboards to work better with Indic language scripts, and even trained its voice assistant to understand Hinglish, a mixture of Hindi and English that millions of Indians use colloquially, which trips up Alexa and Siri regularly.”

The customization and personalization that was so attractive early on has revealed some downside risks as Eli Pariser wrote about in his 2011 book The Filter Bubble: What the Internet Is Hiding from You. As Bill Gates noted in 2017, technologies such as social media, “lets you go off with like-minded people, so you’re not mixing and sharing and understanding other points of view…It’s turned out to be more of a problem than I, or many others, would have expected.” In some ways, algorithms coupled with individuals’ own self-selection have created siloed pools of internet users and driven a type of bifurcation.

Local rules and laws create bifurcation. Europe’s right to be forgotten creates separate internets. China has long maintained state control over the internet there, while Saudi Arabia, Iran and other countries in the Middle East exert similar influence. Turkey’s two-year block of Wikipedia was just ruled unconstitutional by the Turkish Supreme Court. The court voted 10-6 in favor of Wikipedia so the mistake of the ban wasn’t as obvious as we might hope. And Russia has begun testing a national internet system that could enable the country to operate its own alternative sovereign internet.

Governments have obviously realized the power of a connected populous. Many watched the Arab Spring redefine nations in early 2010s. Today, nations like India and Hong Kong frequently turn off the internet in times of public uprising. And as governments turn off the internet to quell protests, users are turning to offline messaging solutions like Bridgefy and FirecChat which use the Bluetooth connectivity of mobile devices to create peer-to-peer mesh networks so that individuals can continue to communicate and coordinate even when communication networks are down. A bifurcating internet that is increasingly state-controlled will lead to greater offline communication and information dissemination, even if it is digital.

Companies also see the inevitability of a splintering internet. Google was widely criticized when it was revealed it was secretly working on a censored search engine project dubbed “Project Dragonfly.” Apple recently took heat for complying with Russia’s demand to show the annexed Crimean peninsula as Russian territory inside of Apple Maps and its Weather app, when viewed inside of Russia.

A bifurcating internet could mean smaller addressable markets for digital products and services. It likely means that local alternatives have longer runways. Most countries want to develop a domestic innovation agenda. A more state-controlled internet likely translates into support for domestic companies. This has long been true in China. Just this week, The Wall Street Journal reported that China’s Huawei received as much as $75 billion in tax breaks, financing, and create resources. A bifurcating internet will likely result in similar dynamics, though perhaps somewhat less pronounced, in more countries. It is a net negative for U.S. tech companies.

Last week I spent time with several teams at Paylocity – a cloud-based provider of payroll and human capital management (HCM) – at their HQ just outside of Chicago. But Paylocity isn’t just a producer, they are also a consumer. Paylocity uses their software platform for their own HCM.

In fact, Paylocity is also their own largest client. While an average Paylocity client has just over 100 employees, Paylocity has over 3,000 employees, making them the largest user on their platform. This creates an interesting environment for experimentation and exploration. It also presents an interesting approach for product development and highlights how cross-department collaboration can be operationally important.

While Facebook’s early motto “move fast and break things,” might now conjure negative sentiment, it also invokes an important Schumpeterian spirit – in order for companies to remain relevant, organizations must reinvent themselves from the inside out. Throughout my time at Paylocity I frequently heard references to, “drink our own wine” or #DOOW – the idea that Paylocity sticks by the quality of their products and services by using them as well. The subtle difference between, “moving fast and breaking things” and #DOOW is accountability. It isn’t enough to break things if that new knowledge isn’t poured back into the product.

Some observations related to #DOOW:

Product Development Can Take Place from the Inside Out

Historically organizations might build a product roadmap by eliciting input from clients. But when an organization “drinks their own wine” they create a feedback environment within the halls of the organization. End-users and developers are nested within the same organization. Developers can get real-time feedback from end-users while also more quickly adjusting to evolving requirements.

Celebrate Breaking Things

The teams I interacted with at Paylocity celebrate breaking new features being rolled out to customers. By catching bugs early in what amount to real-world tests, Paylocity is able to shorten innovation cycles and innovate iteratively.

Massive Cross Department Collaboration

Many organizations talk about cross department collaboration, but true coordination is often scarce. Most executives are experts within their own domains and cross department alliances don’t often transcend the products and services being rendered. But when organizations are building product features from the inside out and “drinking their own wine,” employees must rely more heavily on each other. Cross department collaboration in these instances will help the HCM team get the features they want and ensures developers are building the features their end-user will use.

The #DOOW philosophy has become a core aspect of Paylocity’s value proposition. Paylocity stands by their HCM offerings by relying on the tools they build for others. And as the largest company on the platform, it creates some wonderful dynamics for innovation for themselves and others. It is an organizational approach that is highly conducive to digital products and something we will see more frequently as a result.

It reminds of the origin of AWS. As Ron Miller writes, Amazon “was growing quickly and hiring new software engineers, yet they were still finding, in spite of the additional people, they weren’t building applications any faster. When [Andy] Jassy, who was Amazon CEO Jeff Bezos’ chief of staff at the time, dug into the problem, he found a running complaint. The executive team expected a project to take three months, but it was taking three months just to build the database, compute or storage component. Everyone was building their own resources for an individual project, with no thought to scale or reuse. (I think you can guess where this is going.) The internal teams at Amazon required a set of common infrastructure services everyone could access without reinventing the wheel every time, and that’s precisely what Amazon set out to build — and that’s when they began to realize they might have something bigger.”

By solving internal problems, companies like Amazon and Paylocity are creating solutions that are attractive to other organizations.

How does your organization, or other organizations you’ve seen, “drink their own wine?”

Yesterday Best Buy reported 4Q FY18 results well above Street estimates. Comparable store sales in the quarter were up 9% (domestic) and 9.9% (international), compared with a 3% Street estimate. US comps were especially strong for entertainment (16.8%) and appliances (20.7%). Entertainment was up 26% year-over-year (likely driven by gaming). Appliance sales were up nearly 30% year-over-year and broke a billion dollars in the quarter for the first time. Overall sales were up 13.4% year-over-year with consumer electronics growing 10.3% and the computer and mobile phone category growing 13.4%.  

Domestic online sales were up 21.9% year-over-year and were 20% of total domestic revenue in the quarter. We’ve long talked about online sales but at 20% of total domestic sales, it really starts to restructure the business. With 70% of Americans living within 15 minutes of a Best Buy store, in-store pick-up and same-day delivery are key components of growth moving forward. Best Buy has been expanding their same-day delivery services. They’re also expanding their in-home consultation services to help unlock what they see as latent demand.

Best Buy is also planning to close all of its 250 mobile-focused stores by the end of May 2018. These stores were built to leverage growing mobile phone adoption. Today, nearly every U.S. adult (95%) now owns a cellphone, and 77% own a smartphone. At the same time, these stores were also built to create customer convenience but the new convenience is online.

In a letter to employees,  Best Buy chief Hubert Joly noted that “the cost of operations in our Mobile stand-alone stores is higher than in our Big Box stores.” Some 55% of Best Buy’s phone stores are within three miles of one of its big-box stores. The larger stores are better positioned to handle the warehousing of online orders that will either be picked-up in store or delivered same-day.

So where’s Best Buy’s Achilles heel? It’s in services. Service revenue was down 9.3% on a year-over-year basis. As online sales increase as a share of total sales, Best Buy will need to grow higher-margin services in order to sufficiently differentiate themselves.   

Earlier this week Axios published the results from a recent Axios-SurveyMonkey poll. The results reveal an interesting dichotomy. There appears to be wide consensus that technology is making us better off and that this will continue into the future. Consider the following findings:

82% of American adults believe the success of U.S. technology companies has had a positive effect on the overall U.S. economy
71% of U.S. adults believe technology has had a positive effect on our society
74% of U.S. adults believe technology will make life better for themselves and their families when looking out 10 years

At the same time, the majority of respondents (55%) believe technology will take away more jobs than it creates over the next 10 years. While we are seemingly optimistic about technology’s role in our lives, the outlook dims when it comes to jobs. And amazingly this pessimism holds across all age cohorts. While you might expect pessimism to increase with age, the results of the poll suggest otherwise. Millennials appear to be just as pessimistic as older cohorts for example. And while the provided crosstabs don’t indicate statistical significance between age cohorts, the oldest cohort (65+) actually appears to be the least pessimist of the bunch. Only 49 percent of those over 65 believe technology will take away more jobs than it will create in the coming decade.

We’re left with a glaring juxtaposition. How can technology both make us better off and be a net destroyer of jobs? Possible in the short-run perhaps, but difficult to imagine that both of these outcomes could be sustained over the long-run.

Certainly innovation has destroyed its fair share of jobs. In the early 1900s, over 40 percent of Americans were employed in agriculture. By 2000, that figure had dropped to 2 percent. This phenomenon has played out across a variety of economic sectors. Have you seen an elevator operators recently? How about stagecoach drivers? It’s clear there will be job loss. Innovation has always and forever left an imprint on the labor market. But innovation has yet to be a net-destroyer of jobs over the long-run.

What we have seen, especially in recent decades is a polarization of the workforce. We’ve experienced growth in non-routine work at both ends of the income distribution, while automation and technological change have hallowed out the middle class. But changing jobs doesn’t translate into huge changes in the rate of employment. So again, innovation has been a net-creator of jobs over the long-run.

We’ve long had a delicate relationship with technological progress. Consider this New York Times headline from 1928. We seem to be able to easily see how technology improves our lives and we can see the jobs that are being lost to innovation and automation. But we’ve long lacked the creativity to imagine how technological transformation will translate into new jobs. This dichotomy plagues our ability to plan for the future because we have very little sense of what that future will entail. Our movies seem to do a good job of portraying future leisure but not a great job of deciphering future work. We seem to intuitively know that technology is making us better off but have little understanding of the employment implications of technology improving our lives. We see the first-order effects, but can’t seem to decipher the second-order effects.

Here’s just one example of what 10 years can bring us. There’s an estimated 22 million software programmers in the world, with maybe 8-10 million developers focused on mobile apps. The first iPhone was released in 2007, but it wasn’t until July 2008 that Apple opened the App Store. The app store as we know it today, wasn’t even in existence 10 years ago. And without it, few developers were focused on mobile development.

So how should we think about the future of work?

While we fear losing our jobs to automation, technological transformation has really been more focused on improving our productivity than on replacing us. Improving productivity rates enables businesses to produce more with less workers but also means they can expand production in other areas which in turn requires new workers. The net has always been positive.

As history has shown us, we should presume innovation will beget jobs that don’t yet exist today. To understand the future of work, we should look to where production might expand. Here’s one area to consider. Technology shifts consumption. Overtime technology has increased the share of luxuries in overall consumption. Jobs follow these shifting consumption patterns and will continue to do so into the future. innovation will continue to make us better off and be a net-creator of jobs.

The market for smart speakers, also commonly referred to as voice-enabled smart speakers and virtual personal assistant (VPA)-enabled wireless speakers, is still a very nascent market. Yes, the market has enjoyed tremendous growth, especially over the last six months, but there is still significant growth yet to be had.

I’m already beginning to see some analysts suggest we are close to peak adoption/sales and that annual sale volume will begin to decline as soon as 2020. I created a very quick household adoption model for smart speakers to illustrate why this is dead wrong.

I’ve modeled a simple s-shaped diffusion curve for smart speakers. You can see we are still in the early phase of adoption. My model makes some simple assumptions:

1. Smart speakers have a long-term (peak) household adoption rate of 60 percent
2. Smart speakers will reach 55 percent of their peak adoption rate by 2020
3. Smart speakers will reach 90 percent of their peak adoption rate by 2023

We can change any of these assumptions without fundamentally changing the outcome. The reality of the smart speaker market is probably close to my illustration. My simple model suggests roughly six percent of households owned a smart speaker by the close of 2015 and some 19 percent own one today and those seem to be consistent with what the analyst community is estimating.     

Analyst estimates suggesting annual smart speakers sales will peak in 2019 and begin to decline are difficult to defend. My simply illustration/model certainly doesn’t support that premise. Here are some narrative arguments in support of my perspective and the simple model I present:  

1. Apple’s HomePod hasn’t started to ship yet. Other competitors (i.e. Microsoft) could also enter the market. This will be a lift for 2018-2020 unit volume sales
2. Unit volume is inversely related to price. Any increase in competition will drive prices lower and increase unit volume. We’ve already seen that with the introduction of the Amazon Echo Dot and the Google Home Mini.
3. Google’s major investment at CES suggests Google is intent on making a strong stand in the smart speaker market. Neither Amazon nor Google seem concerned about margins on hardware sales. They both appear to be after a decidedly larger market and are willing to discount hardware prices in order to propel sales higher.   
4. Current ownership rates are only at 15-20 percent. Annual sales volume won’t turn negative until ownership rates get closer to peak household adoption levels. My simple illustration here suggests annual unit volume should continue to grow through at least 2021.
5. Density rates, the number of smart speakers per household, are still low, but I expect them to rise significantly. TVs are one of the most densely owned consumer technology products. Households own roughly three on average. Given how use-case scenarios are emerging, I could easily see smart speakers eclipsing that. We spend roughly 60 percent of our awake time at home in the kitchen. As households began adopting smart speakers, they naturally placed the first one they bought in the kitchen. From there households are adding smart speakers to secondary and tertiary rooms like bedrooms, offices, and living rooms.   
6. My simple model suggests a household adoption rate with a positive second derivative until 2020. In other words, adoption is increasing at an increasing rate until 2020. Even in a world where sales are only purchased by new households, in other words, there are no replacement purchases nor households buying more than one unit, unit sales volume should grow through 2020.    
7. Industry analysts seem to confuse s-shaped diffusion curves with annual unit sales occurring over the same time period. Remember, the s-shaped diffusion curve characterizes household adoption and ownership rates over time. It doesn’t explicitly say anything about sales. Sales are a function of three things: (1) new household adoption, (2) replacement purchases by households who purchased the product in a prior period, and (3) households buying more than one device (density rates expanding beyond one). Only #1 is directly influenced by the shape of the diffusion curve. And remember, my simple illustration shows household adoption increasing at an increasing rate into 2020.

The smart speaker market is nascent and we aren’t close to peak adoption, nor peak sales, anytime soon.

Artificial intelligence (AI) will reduce use of centralized services and give further rise to decentralized, distributed services. The social implications of this are just beginning to materialize. Consider the case of Uber. Recent academic research found Uber’s entry into new markets decreased ambulance rates in those markets by at least seven percent. In a world with only ambulances, individuals have few options when seeking medical attention. But not all medical needs are life threatening or require an ambulance. Calling an ambulance is a blunt instrument in a world with a wide spectrum of medical needs. The presence of ridesharing services offer an alternative that individuals can use in place of ambulances. The resulting decline in ambulance demand decreases wait times, improves patient survival rates and leads to broad societal welfare improvement. Moreover, when individuals can self-select less expensive alternatives to ambulances or specify which hospital to present at, it should free up scarce, expensive resources and enable these services to be used more efficiently.

The case of Ubers and ambulances is just a single example of things to come as more services are born from AI. Machine learning and other AI techniques are the underlying foundation for every part of ridesharing services and other companies like them. From figuring out the appropriate fare to providing estimated arrival times to ensuring vehicles are positioned in the right quadrants of cities. As self-driving vehicles become prevalent, AI will play a more active role in every aspect of ridesharing services and the services that develop around them.

AI is built on data and AI techniques like machine learning take into account a wide range of digitized information. Consider Uber’s use of machine learning to estimate how long a given Uber Eats delivery will take. Their models will naturally take into account drive times and traffic conditions for different parts of the city at different times of the day. But as both data and sophistication grow, these models will also recognize and take into account that the time needed to make noodles differs from that needed to make a hamburger.

Centralized services are often built on the economic premise of economies of scale. We can lower the cost of rendering services by scaling a service across a wider number of individuals because of the inverse relationship that exists between the per-unit fixed costs of production and the quantity produced. Economies of scale are also driven by a matching problem that exists in the analog world – how to get the right things to the right places for the right people at the right time. Digitization and datafication helps solve this inherent matching problem.

We are already seeing how machine learning techniques are lowering points of friction across a very wide service market. In the case of ridesharing, this might entail telling an individual to walk 15 yards to a nearby street corner to improve accuracy and reduce pick-up time. It would be impossible for humans to optimize this outcome for every street corner in every city of the world.

Cities like Washington, DC are exploring how to integrate ridesharing services or other alternative services into their city-wide EMS systems to lower costs and improve outcomes. Doing so would enable cities to broaden and expand the distribution of available services, and as a result they also would indirectly incorporate AI into their city-wide systems. In the future, 911 systems might include AI nurses who assess the situation in real-time and decide the appropriate response.

AI makes decentralized, distributed services more efficient and effective and in turn will make these services more applicable and important moving forward.