Writing for Machines, Not People

It is a truism that, over the last 20 years, news writing, reporting and even user-generated content have had to adapt to machine readers and crawlers that index online content and algorithms that rank display results. That now seems to be spreading to the writing of financial reports as well. 

“Corporate disclosure has been reshaped by machine processors, employed by algorithmic traders, robot investment advisors, and quantitative analysts,” say Sean Cao, Wei Jiang, Baozhong Yang & Alan L. Zhang in a paper written for the National Bureau of Economic Research. 

“Our findings indicate that increasing machine and AI readership, proxied by machine downloads, motivates firms to prepare filings that are more friendly to machine parsing and processing,” they say. Of course. 

To be sure, beyond writing for the algorithms, many more informal styles also have emerged.  But structural changes are important in view of the use of search engine translators, for example. Use of idioms arguably is discouraged in headlines, for example, as those might not translate easily and accurately into other languages. 

Headline writers often prefer to use puns, or word play. That might not work as well for crawlers indexing the content. So factual “indexable” headlines--if less clever, perhaps--are more important. 

So is the importance of “clickable” writing and storytelling: creating content that gets attention. Sensationalism still works, perhaps sadly. 

But the larger point is that writers increasingly write to be noticed by algorithms. Yes, people still write to be read by people, but to get an audience also means writing for the machines that index content. 

Reporters, meanwhile, either have chosen--or are forced to--step outside the bounds of traditional objective journalism, which creates many new problems for readers, viewers and listeners. 

How Deep a Revenue Dip; How Swift a Rebound for Telecom Revenues?

In April 2020, early in the Covid-19 pandemic, Analysys Mason forecast telecom revenue in developed markets would fall by about 3.4 percent in 2020, with less than one percent growth in 2021. 

Global revenue might not reach 2019 levels until 2023, Analysys said in December 2020. 

International Data Corp., on the other hand, predicted that global telecommunications and subscription TV services revenue would dip less than one percent in 2020.

Note that these forecasts were made in May 2020, and might well change. Here is Analysys Mason’s May 2020 forecast, with a range of cases from mild to severe impact. Also note that the "moderate" scenario tracks the "mild" scenario closely. Some might argue that means the "moderate" recovery case is almost the same as the "best case" forecast.

In the moderate case, we return to 2019 revenue levels by perhaps 2023. The "worst case" is almost impossible to anticipate, in terms of time of recovery.

source: Analysys Mason 

In the first quarter of 2020, revenue declined about 2.2 percent globally, according to ResearchandMarkets. Some service providers saw revenue “growth rate” declines as high as 12 percent in the first quarter alone, including telcos in the U.S. market. 

Even if it seems too optimistic, the IDC prediction is well within historical expectations. The Great Recession of 2008 caused a momentary flattening of revenue growth, with the prior pattern asserting itself quickly afterwards. A modest dip would not be without precedent, even if we fear greater damage. 

And though it is reasonable to expect a dip in business customer spending (with economies shut down and significant bankruptcies expected), consumer spending on telecom services might well increase, as it did in the United States in the aftermath of the 2008 Great Recession. 

source: Statista

IDC estimated global service provider revenue at nearly $1.6 trillion in 2020, a decrease of 0.8 percent compared to 2019. IDC expects the decline to continue in 2021, but at a somewhat lower degree. 

The mobile segment, the largest segment of the market, will post a slight decline in 2020 due to lower revenues from roaming charges, less mobile data overages due to the stay-at-home situation, and slower net additions, especially in the consumer segment, IDC argues.

Fixed data services spending was forecast to increase by 2.9 percent in 2020, however. In the U.S. market, that appeared to have happened. In the United States, in the third quarter of 2020, for example, net broadband additions in the first quarter were the highest since 2015, while in the second quarter, net additions were about triple the rate for the same quarter of 2019, according to Leichtman Research Group figures. 

Subscription video services were expected by IDC to be boosted by the lockdown, but also affected by the economic downturn, so the spending in this category is expected to decline slightly in 2020, IDC said.

The Americas market was forecast to see a tiny decline of 0.04 percent. Europe, the Middle East, and Africa (EMEA) and Asia/Pacific (including Japan) will dip more. Growth is not expected in EMEA or Asia/Pacific before 2022 as the users in emerging markets are expected to remain cautious about spending for some time, IDC estimates. 

source: IDC

It remains to be seen who is correct about the extent of the revenue dip in 2020. It also remains to be seen whether a historically-seen pattern of relatively-swift rebound happens once the pandemic is over.

How Long Before Telecom Revenue Returns to 2019 Levels?

How long will it take for global telecom revenue to return to 2019 pre-pandemic levels?

Perhaps the good news is that, right now, based on development of vaccines, it seems as though the second quarter of 2021 might see an end to most economy-affecting prohibitions and impediments to normal levels of commerce. 

The bad news?

If so, we might not see a return to early 2019 levels of revenue until perhaps 2023, according to Analysys Mason. A worst-case scenario delays recovery to 2019 levels until some time after 2024.

Note that these forecasts were made in May 2020, and might well change. Here is Analysys Mason’s May 2020 forecast, with a range of cases from mild to severe impact. Also note that the "moderate" scenario tracks the "mild" scenario closely. Some might argue that means the "moderate" recovery case is almost the same as the "best case" forecast.

In the moderate case, we return to 2019 revenue levels by perhaps 2023. The "worst case" is almost impossible to anticipate, in terms of time of recovery.

source: Analysys Mason 

The bright spots are residential broadband (assuming strength in internet access but weakness in linear video and voice), business unified communications and internet of things.

source: Analysys Mason 

Many seemingly believe that all the past year’s emphasis on remote work and remote learning “must” be good for connectivity providers, and to be sure, some evidence can be cited, such as an increase in broadband  subscription rates in the United States, in the third quarter of 2020, for example. Net additions in the first quarter were the highest since 2015, while in the second quarter, net additions were about triple the rate for the same quarter of 2019. 

The story overall is not that bright, as most service providers, in most markets (perhaps nearly every market) have seen a dip in revenues in 2020. Mobile operators, for example, have seen declining revenues for roaming, since people were not traveling as much as usual. With businesses operating in restricted mode, if at all, usage was down there as well. Many small businesses have gone bankrupt, so customer additions were negative in that sense. 

At the same time, costs increased as firms scrambled to support a sudden surge of at-home work patterns. In the first quarter of 2020,revenue declined about 2.2 percent globally, according to ResearchandMarkets. Some service providers saw revenue “growth rate” declines as high as 12 percent in the first quarter alone, including telcos in the U.S. market. 

source: TBR 

Profits might diverge from the revenue declines in some cases, as some firms take steps to restructure. But top-line growth will tend to suffer at most firms. In India, for example, fixed network subscribers dipped nearly seven percent, year over year, in September, while mobile accounts shrunk by about two percent.  

source: TRAI 

We might well be prepared for surprises, though. In past recessions, telecom revenue has rebounded relatively quickly, and revenue dips were shallow. But the pandemic has not been a "typical" recession, leaving open the possibility that recovery could be on a different pattern.

U.S. Cable Upstream Peak Traffic Grows 47% Since March 2020

U.S. cable TV networks, which have 70 percent of the installed base of broadband connections, have seen cumulative downstream peak traffic growth since March 1, 2020 of 27 percent, with peak upstream growth up 47 percent since March 1, 2020, says the NCTA.

Throughout the pandemic, provider backbone networks have shown no signs of congestion, the group also  notes. 

source: NCTA 

Can Overbuilder ISP Market Share Exceed 30 Percent?

One key metric for any internet access overbuilder (any internet service provider competing against both a cable operator and a telco) in a market is take rate: the percentage of addressable homes that sign up for service. 

The reason is that a rule of thumb suggests an ISP competing as an overbuilder has to get about 30 percent share to be sustainable. The corollary is that the business case for other incumbents can get perilous, as that leaves 70 percent share for all the other providers.

It might help if the incumbents have multiple revenue streams to rely on (voice and video entertainment being the most common, plus business services). Still, a viable overbuilder dramatically reshapes both market share and profitability profiles for incumbents. 

Ting Internet seems to have overall take rates of 28 percent. In the past, Ting has estimated it would get about 20 percent take rates  within the first 12 months of active marketing, with a goal of reaching 50 percent over five years. 

In the past, Ting Internet has reported 30 percent take rates, so the latest data suggests that is a reasonable expectation for the markets Ting chooses to address. 

As was the case for overbuilders in the video entertainment business, and now with triple-play and internet access overbuilders, it has proven difficult to reach 20 percent market share after several years of operation.

As even the most-successful overbuilders have discovered, being the market leader is difficult, as the leader’s market share, in a competitive three-way market, might not exceed 30 percent or so of homes (potential accounts).

EPB in Chattanooga, Tenn., for example, is the poster child for overbuilder hopes, having gotten as much as 45 percent of consumer market share in its service area (with share defined as revenue-generating units, not “accounts” or “homes”).

EPB’s internet access share might be about 27 percent, its video share lower than that. An interesting statistic, in that regard, is that about eight percent of EPB’s internet access customers buy the gigabit service.

EPB’s market share is highly unusual in most overbuilder markets, as EPB arguably competes with Comcast, not AT&T, which has negligible video share in Chattanooga.

In a triple-play market, that is important. Comcast might have 61 percent video share, while EPB might have 36 percent share, leaving only three percent video share held by AT&T. Essentially, EPB has become the number-two provider, relegating AT&T to third place, something that is not the case in most other U.S. markets where three mass market fixed network suppliers compete.

Telcos Not Much Closer to Success in Enterprise Solutions Market, ABI Research Says

Telcos might not be much closer in 2021 to a goal of becoming relevant in the cloud-based enterprise solutions market, Don Alusha, ABI Research senior analyst suggests. 

That should surprise almost nobody, as the application value chain now rests on the internet, software platforms, and the cloud. That allows near-zero distribution and near-zero transactional costs, Alusha says. 

Direct competition from hyperscalers is a factor as well. Amazon offers multiple devices for either edge or on-premises deployments: Snowcone, Snowball, Wavelength, Outposts, and Greengrass IoT, for example. 

Hyperscalers also are already deploying enterprise digital solutions, most of which are usage- or subscription-based instead of upfront, capex-based solutions that telco solutions arguably require, says Dimitris Mavrakis, ABI Research senior research director. 

Telco operators will need to adapt to opex models to survive, especially in the small and medium enterprise (SME) segments of the market, he says.

“2020 has seen AWS, Google, and Microsoft all advancing and underlining their telco ambitions to provide enterprise connectivity solutions” says Mavrakis.

“Their existing ties with enterprises for cloud storage, as well as their general openness toward service-based offerings, will make them particularly attractive to enterprises,” says Leo Gergs, ABI analyst. 

That might also include hyperscalers supplying access services as well, he says. 

Hyperscalers are leading the market in consumption economics, says Alusha. In a sense, that is a continuation of a trend we saw a decade or two ago, as the enterprise software business became “consumerized,” with many consumer apps adopted first by employees and then supported by enterprise IT. 

Enterprise IT of the past was based on high switching costs, relatively low volume, high price, and a pay-up-front capital investment model. 

The future purchase pattern will instead be based on high volumes, low pricing, and an opex model. 

Telcos and their suppliers are not yet ready to fully embrace consumption economics, says Alusha. 

Telecommunications is an asset-intensive industry with expertise in managing factories and supply chains, developing technologies, and understanding the cost of goods sold, inventory turns, and manufacturing. 

Human-intensive services are entirely different, Alusha says. For example, in services and opex-based models, technology providers do not manufacture a product to then sell it. Instead, they sell a capability or knowledge, created “at the same time they deliver it,” he says. In other words, software is sold as a service, not a product. 

Loon Uses AI to Steer Balloons More Effectively

Loon’s navigation system’s most complex task--steering balloons--now is assisted by artificial intelligence, specifically deep reinforcement learning, a type of machine learning, Loon says. 

Overall, the RL system kept balloons in range of the desired location more often while using less power, according to Loon. 

source: Loon 

Reinforcement learning enables an agent to learn by trial and error in an interactive environment using feedback from its own actions and experiences, Loon says. “As far as we know, this is the world’s first deployment of reinforcement learning in a production aerospace system.”

This contrasts against the conventional approach of the automated system following fixed procedures artisanally crafted by engineers, Loon notes. 

In this case, RL helps optimize flight in the face of time-varying winds, partial visibility of the wind field surrounding the balloon, not always having enough power for the ideal maneuver and frequent decision points. 

Google AI teams hope that beyond Loon, this work on stratospheric navigation can serve as a proof point that RL can be useful to control complicated, real world systems for fundamentally continual and dynamic activity. 

Global Telecom Revenue Will Reach 2019 Levels in 2023

The telecom industry will take until 2023 before its revenue exceeds that for 2019, according to Analysys Mason, with 2021 revenue growth of about one percent. 

The industry will see a US$43 billion fall in 2020 revenue, a 2.7 percent drop, driven by lower mobile roaming revenues, business services and prepaid mobility. 

“One third of this will be recouped in 2021,” said Stephen Sale, Analysys Mason research director for consumer services.

The company sees no significant revenue from 5G in 2021. “It’s not a game changer for consumers,” he said.

Global WAN Services Revenue $76 Billion in 2020

The global enterprise wide area networks business generates about US$75.9 billion, TeleGeography says. 

MPLS represents 43 percent of that revenue; dedicated internet access 16 percent; SD-WAN two percent and local access (leased line and Ethernet connections from customer premise to carrier point of presence) 38 percent.

source: TeleGeography 

Earlier in 2020 I estimated $48 billion in 2020 enterprise WAN (long haul only) spending, which is very close to TeleGeography’s latest estimates. Global public network service revenues are about $1.7 trillion or so. So public network WAN service revenue represents about 4.4 percent of total public network revenues, with the WAN portion (excluding local access) being $46.6 billion or about 2.7 percent of global public network revenues. 

Some estimates put global public network revenue at a higher level around $2 trillion annually, in which case WAN services represent about two percent of total public service revenues. 

Not included in such figures are private WANs operated by hyperscalers and application providers, as they build and own their own networks. Nor would it be surprising if such buyers had a preference for dark fiber purchases or leases, rather than “lit” services. 

Using 2025 as a starting point for carrier SD-WAN services, it has taken five years to reach two percent share of total WAN service provider sales, or about 3.4 percent of total long-haul revenue, excluding local access. 

The expectation is that SD-WAN will cannibalize MPLS. 

Among the trends the latest TeleGeography data cannot show is the global shift to private networks for enterprise WAN traffic. 

By 2016, more than 70 percent of all internet traffic across the Atlantic was carried over private networks, not on public WAN networks. Obviously, that also means no revenue was earned directly by public service providers for carrying that traffic.

On intra-Asian routes, private networks in 2016 carried 60 percent of all traffic. On trans-Pacific routes, private networks carried about 58 percent of traffic.

The "5G Race" Storyline is Wrong, As Are So Many Others

The “5G race” story framework seemingly is irresistible: there will be winners and losers, with the winners moving fastest to deploy the networks. The only problem with the storyline is that it likely will prove to be false. 

Does anybody really believe being “first” with analog mobile services, or any of the past digital generations (2G, 3G, 4G) has mattered? Has it changed the economic positions of nations, or industries, beyond what we would expect for other reasons?

In other words, does early or late adoption actually matter? A fair assessment might be that it could matter for industry suppliers, in terms of market share. Some might argue Huawei gained from early supply of some 5G infrastructure, while Nokia suffered. 

On the other hand, the new emphasis on open and virtual networks opens the door for new suppliers, which might make early victories by incumbents irrelevant over the longer term, as new firms enter the supply chain. 

“Early or late” might temporarily provide advantage or disadvantage for particular mobile operators in some markets. But making sense of the advantage also must include the momentum and growth profiles of each firm before 5G. Maybe a firm gains or loses share in 5G because it already had been gaining share in 4G, for reasons unrelated to 5G deployment. 

Among the historical examples of the irrelevance of the early-late paradigm is the development of several technologies in the U.S. market, where adoption always has been “late.” That is said to be true now of U.S. 5G speeds. It is true for the moment, but ultimately the relevant gap will disappear. 

That does not mean U.S. speeds, on average, will be among the top 10 globally, for example. U.S. mobile speeds are slow, and have been relatively slow, for 4G services, compared to many other markets. The point is that it will not matter, in user experience or other expected benefits (for industry, firms, economic growth, innovation). 

But the “U.S. is behind” storyline has been used often over the last several decades. Indeed, where it comes to plain old voice service, the U.S. is falling behind meme never went away.

In the past, it has been argued that the United States was behind, or falling behind, for use of mobile phones, smartphones, text messaging, broadband coverage, fiber to home, broadband speed or broadband price. 

In the case of mobile phone usage, smartphone usage, text message usage, broadband coverage or speed, as well as broadband prices, the “behind” storyline has proven incorrect, over time. 

Some even have argued the United States was falling behind in spectrum auctions. That clearly also has proven wrong. What such observations often miss is a highly dynamic environment, where apparently lagging U.S. metrics quickly are closed.

To be sure, adoption rates have sometimes lagged other regions. Some storylines are repeated so often they seem true, and lagging statistics often are “true,” early on. The story which never seems to be written is that there is a pattern here: early slowness is overcome; performance metrics eventually climb; availability, price and performance gaps are closed over time. 

The early storylines often are correct, as far as they go. That U.S. internet access is slow and expensive, or that internet service providers have not managed to make gigabit speeds available on a widespread basis, can be correct for a time. Those storylines rarely--if ever--hold up long term. U.S. gigabit coverage now is about 80 percent, for example. 

Other statements, such as the claim that U.S. internet access prices or mobile prices are high, are not made in context, or qualified and adjusted for currency, local prices and incomes or other relevant inputs, including the comparison methodology itself. 

Both U.S. fixed network internet prices and U.S. mobile costs have dropped since 2000, for example. 

The point is that the “U.S. is behind” storyline seems irresistible. But it also ultimately is meaningless. All the relevant gaps were eventually overcome. One possible explanation is that U.S. service providers, who earn high profit margins compared to most other countries, upgrade deliberately, to maintain margins, rather than necessarily rushing to “be first.”

Consumer demand also is an issue. It can be argued that U.S. consumers wait to see value before adopting new technology, instead of rushing to buy the latest technology “just to be early adopters.”

The point is that the “5G is a race” is an irresistible storyline. But it arguably will be proven false. Countries, firms and consumers will adopt 5G when it makes sense, when it offers value, or simply as a byproduct of buying some other product, such as a desired phone model. 

That is, in substantial part, related to another problem journalists face, namely the “next big thing” storyline that becomes news because that is why proponents and vendors are pushing. Many journalists would probably agree that they tire of writing stories about the next big thing, or the present big thing, over and over again. It seems to be an occupational hazard. 

But the point is that easy storylines are irresistible for possibly lazy journalists. To be sure, deadlines create the need for story construction tools, including the venerable “two sides” framework (he said, she said). We all use such tools. 

That is why elections are characterized as horse races, or why verification matters. Still, some might argue that a bit of  laziness is why verification is sorely lacking, or why more original stories are not routinely created. It is not easy to do so routinely. It is harder work. But sometimes it leads to “better” storytelling.