How Does ChatGPT "Daily" Usage Compare to Other Popular Apps?

According to altindex, fewer than 10 percent of potential users use ChatGPT daily, the highest usage being nine percent daily use among users 18 to 24; the lowest usage among those 55 or older, at about one percent. 

That might represent significant usage for an app that was released commercially in 2023. 

source: altindex

But determining “active daily use” is not easy. There is no universally agreed-upon definition for "daily active users.” Most apps track user actions to determine engagement and count users who perform certain actions within a 24-hour window. But that requires a judgment call.

Does one track when an app is opened; whether a significant interaction occurred or session length? And how are interactions or session length defined?

If a user opens the app at least once within a 24-hour window, that typically is counted as a DAU.

Does active use mean watching a video for a certain duration, making a purchase, sending messages, or completing a specific task within the app?

Engagement with core app features often matters when counting active usage. On a social media app, a user viewing posts, liking content, or messaging someone might be considered an active user. 

An e-commerce app might consider adding items to a cart or making a purchase as active use.

And how long does a session have to last to be counted? A user who opens the app multiple times but only for a few seconds each time might not be counted as a DAU, while someone who has a single, extended session might be.

Consider that perhaps 39 percent of Instagram “users” are considered daily active users Likewise, perhaps 30 percent of the TikTok audience are daily users, by some estimates. 

By other estimates, about 20 percent of YouTube users are considered daily active users. 

App Name	Daily Active Users (%)
Instagram	39
TikTok	29
YouTube	20
Amazon	15

But some estimates of daily usage among active users are much higher. As with all such estimates, it matters greatly which denominator is used: all persons; all registered users; all active users; all active users who have used an app within the past week and so forth. 

Also, it makes a difference when counting “monthly” active users versus “daily.” 

Platform	Percentage of Users Daily
Facebook	68.38%
Instagram	63%
Snapchat	61%
Twitter	42%
YouTube	51%

The point is that ChatGPT usage is not out of line for an app that has been in commercial use less than two years. 

AI Job Displacement Might Resemble Prior Manufacturing, IT and Service Offshoring and Outsourcing

Will artificial intelligence lead to a new wave of “offshoring” or “outsourcing” of jobs, as happened to manufacturing and then many information technology and service jobs? Studies suggest the possibility is real, as AI seemingly enhances the performance of lower-skill workers more than highly-skilled workers, closing the performance gap between high and low. 

While much of the concern about artificial intelligence impact on jobs focuses on job displacement, perhaps a greater issue will ultimately be to “devalue” the wage premium garnered by top performers over lower performers. 

The reason is that AI might benefit lower-skilled workers more than high performers, allowing employers to rely more on lower-skilled workers to perform tasks once requiring higher-skilled workers. 

Wage compression, in other words, can happen when technology allows less-skilled workers to perform expert-level tasks. That might imply faster wage growth for lower-wage workers compared to higher-wage workers.

A study by the Congressional Research Service suggests that between 2019-2023, U.S. real wage growth was dramatically faster for low-wage workers (around three percent annualized) compared to high-wage workers (around one percent annualized). This contrasts with the prior 40 years where high-wage workers saw faster growth.

But forces other than AI might also be at work. A CRS study notes that wage rates for workers with advanced degrees rose faster than for lower-skilled or mid-skill workers between 1979 and 2019. And a study suggests wage increases were fastest for lower-skill workers between 2020 and 2024. 

In other words, we might see another unfolding of job displacement from high-wage areas to low-wage areas if AI allows lesser-skilled workers to accomplish tasks formerly conducted by higher-skilled workers. 

Study	Key Findings
Stanford/MIT (2023)	AI disseminates knowledge of high-skill workers to less-skilled workers
Congressional Research Service (2024)	Wage compression from 2019-2023 with much faster growth for low-wage workers
Oxford University (Frey et al.)	Language AI benefits less-skilled workers more than highly skilled
Brekelmans & Petropoulos (2020)	- AI likely to significantly alter low and middle-skilled jobs - High-skilled jobs relatively less at risk, but impact still non-negligible

Indeed, that is a pattern already established in the 20th century as manufacturing jobs moved to lower-wage countries. Many manufacturing jobs, particularly in labor-intensive industries like textiles and electronics assembly, were outsourced from developed nations to developing countries.

In the late 20th and early 21st centuries, a "second wave" of outsourcing emerged, targeting white-collar jobs in the information technology and services sectors. Countries such as India, with large pools of English-speaking graduates and significantly lower wages, became major destinations for outsourcing IT services, software development, and business processes.

And there already is evidence that AI benefits lower-skilled workers more than highly-skilled workers, narrowing the performance gap between the groups. That should create the possibility of substituting use of AI-enhanced formerly-lower-performing workers in place of at least some highly-skilled workers. 

The issue is how much the performance gap is narrowed and to what degree tasks can be redesigned so that AI-enhanced workers of lower skill can handle tasks once performed by higher-skilled workers.  

A study by researchers from Stanford and MIT analyzed data from 5,179 customer support agents at a Fortune 500 company. The findings showed that access to an AI conversational assistant increased productivity (measured by issues resolved per hour) by 14 percent on average. However, the impact was greatest on novice and low-skilled workers, with minimal impact on experienced and highly skilled workers.

Another study by Oxford University's Carl Benedikt Frey and others found that the introduction of language-based AI software in the workplace benefits less-skilled workers more than highly skilled ones.

Disappointed by Apple Intelligence? Don't Be

Some might profess disappointment with Apple’s “Apple Intelligence” features including AI-assisted Siri, Writing Tools and Image Playground, as none arguably introduce new “killer app” features and mostly make existing use cases “smarter.” But that is likely to be the case for most AI implementations for a while. 

But there's a strong argument to be made that artificial intelligence will be deployed and experienced by most people primarily through their phones and internet experiences; primarily as enhancements or extensions of existing features. 

Most people already have experienced AI when issuing voice commands to their devices; shopping; using social media; smartphone cameras or any form of content recommendation. AI will be applied more often and more extensively for such use cases.  

As most people experience the internet using their phones, so most people are likely to frequently use AI-assisted phone experiences. And it is perhaps pointless to compare the value or importance of work and consumer AI experiences.   

It is hard to compare the “impact” of AI in consumer smartphone and internet interactions, compared to AI use in business or work applications, in part because the “value proposition" is different. “Fun” or “enjoyment” or “convenience” is often the expected outcome of a consumer AI use case, where “productivity” is typically the desired work outcome. 

The sheer ubiquity of the smartphone means it will be the most-common framework for AI encounters by consumers. Compared to other potential platforms for AI, like smart cars or refrigerators, smartphones are nearly universally carried and used. The constant accessibility makes them a prime real estate for AI.

Also, AI on phones is already pervasive. Many features we take for granted, like facial recognition device unlocking, spam filtering, and voice assistants, are powered by AI, and it seems reasonable to suggest that AI acting as a personal assistant will eventually be a major use case. 

With the caveat that individuals will vary in their usage of various apps, potential use of AI-assisted experiences is significant. 

Activity	Daily Usage	Monthly Usage
Virtual Assistants (Siri, Alexa)	15-30 minutes	7.5-15 hours
Smartphone AI Features (camera, autocorrect)	30-60 minutes	15-30 hours
Web Interactions (search engines, recommendations)	1-2 hours	30-60 hours
Productivity Apps (email, calendar, document editing)	2-4 hours	60-120 hours
Entertainment (music, video streaming)	1-3 hours	30-90 hours
Social Media (content curation, facial recognition)	1-2 hours	30-60 hours
Navigation and Maps	15-45 minutes	7.5-22.5 hours
Smart Home Devices (thermostats, security systems)	15-30 minutes	7.5-15 hours

Also, software interactions that are potentially AI-assisted are frequent for office or knowledge workers; less often the case for workers in some industries such as construction, agriculture or manufacturing. 

Office Worker Activity	Daily Usage	Monthly Usage
Productivity Apps (email, documents)	2 hours	60 hours
Virtual Assistants/Chatbots	30-60 minutes	15-30 hours
AI-powered Software Tools	1-3 hours	30-90 hours
Automated Tasks/Workflows	1-2 hours	30-60 hours
AI Skills Development/Training	15-30 minutes	7.5-15 hours

Also, interactions with software in many industries likely includes significant interaction with machines that might incorporate software and AI (cash registers, scanners, other machinery) rather than productivity apps, content or documents. 

Industry	Daily Usage	Monthly Usage
Construction	1-3 hours	30-90 hours
Agriculture	1-2 hours	30-60 hours
Manufacturing	2-4 hours	60-120 hours
Education	4-6 hours	120-180 hours
Retail	3-5 hours	90-150 hours
Finance	5-7 hours	150-210 hours

In the construction industry, some workers may spend one to three hours daily using project management software, computer-aided design tools, and productivity suites for documentation and communication.

At least some agriculture workers typically interact with farm management software, GPS/GIS mapping tools, and data analytics platforms for one to two hours per day.

Manufacturing employees often monitor or interact with computer-aided manufacturing software, enterprise resource planning systems, and productivity tools, accounting for two to four hours of daily usage.

Educators, including teachers and administrators, might use software, productivity suites and so forth when planning lessons, grading papers or tests or doing other support work, with an estimated four to six hours of daily usage, assuming roughly half the day is spent in actual instruction. 

In the retail industry, workers interact with point-of-sale (POS) systems, inventory management software, and productivity tools for three to five hours per day.

Finance professionals, such as bankers, accountants, and analysts, spend a significant portion of their day using financial software, data analysis tools, and productivity suites. 

But the point is that perceptions of Apple Intelligence as “underwhelming,” compared to the potential of artificial general intelligence, for example, miss the point. Most AI implementations will make existing experiences more useful, more fun, more entertaining or more productive, but perhaps not so noticeably at first. 

Over time, more-developed capabilities will emerge, often in the form of fully-autonomous  apps, actions or machines. But that is a way off. Right now, most AI implementations will “make things better” by making them more predictive, more accurate or faster.

The big leaps will come later. 

When Will it Make Sense to Build a Custom Generative AI App?

Yes, it is becoming easier for larger enterprises to build custom generative AI applications. As always, that will not necessarily mean it is a wise use of time and resources to build "all" GenAI apps and features "custom." In many cases it will make more sense simply to use the features existing or new suppliers can support. 

A new survey of 1,300 CEOs by TCS suggests 72 percent already are retooling their firms to support use of artificial intelligence. 

Some 51 percent of surveyed firms are planning to build their own generative AI implementations. Or at least that is what the CEOs believe. 

Whether those beliefs are ultimately borne out in reality or not is not so clear, but it would be reasonable to suggest that where it does make sense, it will most often be in cases where a particular enterprise has some highly-specific tasks of identifiable importance and a particular set of existing software systems that must be supported.

In principle, executives might push in this direction if unusual security or privacy needs exist, or when such custom development offers the chance of gaining a competitive advantage over competitors. 

In most cases, those “custom” implementations are likely to use small language models more often than not, some might argue, as the cost and complexity are more manageable. Also, there are lots of plausible reasons why a respondent might agree with a question “are you planning to create your own enterprise-specific LLMs for use in Generative AI implementations?” 

Not all “yes” answers might mean full intention to create a custom language model. Many respondents might interpret the question more in the sense of “are you planning to customize a LLM for your own company’s use?” Most respondents could respond affirmatively in that sense, at least in the sense of using proprietary company data. 

Remote Work is an Issue for Younger Worker Engagement, Gallup Study Finds

Every technology or trend has externalities--unanticipated consequences--and remote work seems no exception. According to a new Gallup poll, remote work also leads to loneliness.

The issue is whether, and to what extent, that translates into lower productivity. It might already be clear remote work contributes to less employee engagement. 

In 2023, global employee engagement stagnated and overall employee well being declined, Gallup says in a new report.  While both measures are at or near record highs, the lack of improvement is notable, as they follow multiple years of steady gains, the report says. 

“The result is that the majority of the world’s employees continue to struggle at work and in life, with direct consequences for organizational productivity,” the Gallup study suggests. 

Gallup estimates that low employee engagement costs the global economy US$8.9 trillion, or nine percent of global gross domestic product.

source: Gallup 

And the problem seems squarely among younger workers. 

source: Gallup 

Some might argue that older workers have higher self-assessed well being than younger workers, but that seems not to be the case. “A decade ago, younger workers had consistently higher life evaluations than older workers; therefore, the difference in perspective is unlikely to be a product only of life stage," the Gallup report says.

And Gallup finds that 70 percent of the variance in team engagement can be attributed to the manager. 

And for whatever reason, the “New World” countries (North America and South America) have higher rates of “engaged” workers than the “Old World.” 

source: Gallup 

The point is that if there is a causal relationship between remote work; managerial effectiveness; engagement and productivity, then remote work plausibly has some causal relationship with work productivity and a direct causal relationship with managerial skill.