Friday, May 29, 2026

AI Undermines "Answer Questions" Business Models

Chegg is one of the clearest early examples of a public company whose core business was rapidly undermined by generative AI.


But those of you who have worked in any content production industry have seen this before, in the impact of the internet on content business models.


Industry

Traditional Revenue Model

AI Threat

Risk Level

Newspapers

Ads + subscriptions

AI summaries replace clicks

Very High

News websites

Programmatic ads

Search traffic declines

Very High

Magazines

Ads + subscriptions

Commodity lifestyle content

High

Trade journals

Subscription + data

AI-generated research summaries

High

Music labels

Streaming + licensing

AI-generated songs and voice clones

High

Stock photography

Licensing fees

Text-to-image generation

Very High

Online reference sites

Ads + subscriptions

Direct AI answers

Very High

Product review sites

Affiliate commissions

AI recommendation engines

Very High

Educational publishers

Textbook sales

Personalized AI tutoring

High

Local journalism

Ads + classifieds

Reduced traffic and lower cost AI content

Very High


Chegg built a subscription business around three core assets:

  • A massive library of solved textbook problems and Q&A

  • Human experts and tutors

  • A recurring subscription model (students paid monthly for homework and study help). 


So the business moat was built on:

  • Proprietary content accumulated over years

  • Search traffic from students looking for specific solutions

  • Willingness to pay for reliable, structured answers


Generative AI changed all three assumptions, and undermined the business model.


Tools like OpenAI ChatGPT and Google Gemini offered:

  • Instant answers

  • Natural-language explanations

  • Low or zero cost

  • Broad subject coverage

  • Personalized tutoring


This turned Chegg’s premium service into a commodity.


Metric

Peak / Before AI Shock

After AI Disruption

Market capitalization

~$14.7 billion (2021 peak)

~$150–200 million (2025–2026)

Share price

~$113/share

Around $1/share

Revenue trend

Strong pandemic growth

Sustained year-over-year declines

Subscribers

Multi-million paid base

Persistent subscriber losses

2025 layoffs (May)

~248 employees (22%)

2025 layoffs (October)

~388 employees (45% of remaining workforce)


But Chegg likely will not be alone. Other lines of business might have similar characteristics:

  • Sell information rather than physical goods

  • Depend on labor-intensive expert work

  • Have low switching costs

  • Offer outputs that can be generated in text, image, audio, or code. 


Industry

Traditional Value Proposition

AI Substitute

Risk Level

Examples

Homework help

Solved problems and tutoring

ChatGPT-style tutoring

Very High

Chegg

SEO content agencies

Human-written articles

Automated article generation

Very High

Jasper

Translation services

Human translation

Neural machine translation

Very High

DeepL

Basic legal drafting

Contracts and standard documents

AI document drafting

High

Harvey AI

Tax preparation

Form completion and guidance

AI tax copilots

High

Intuit

Customer service BPO

Human support agents

AI chat and voice bots

Very High

Zendesk AI

Coding contractors

Routine software work

Code generation assistants

High

GitHub Copilot

Graphic design for simple tasks

Logos and ad creatives

Image generators

High

Adobe Firefly

Market research summaries

Analyst reports

Automated synthesis

High

AlphaSense

Recruiting screening

Resume review

AI candidate matching

High

LinkedIn Talent Solutions

Medical transcription

Dictation and coding

Speech-to-text + AI coding

Very High

Nuance Communications

Stock photography

Generic images

AI image generation

Very High

Shutterstock


As was the case when the internet disruption began, content suppliers will be in the line of fire:

  • Research report subscriptions

  • Professional tutoring services

  • Basic legal document preparation

  • Simple coding tutorials

  • Generic content websites

  • Q&A platforms charging for access

  • Standardized test prep companies. 


If you can describe your service as “We answer questions about X,” danger clearly exists, as AI will provide a substitute.


Maybe AI is Not Such a Big Job Killer

“The extant empirical evidence does not suggest AI is leading to a large-scale replacement of workers by machines in either output or knowledge production,” argue economists Ajay K. Agrawal, University of Toronto; John McHale, University of Galway and Alexander Oettl, Georgia Institute of Technology. 


“Instead, the evidence seems more consistent with AI as a productivity-augmenting tool used by workers,” they argue in a new paper


And they argue there are important policy implications. If AI actually displaces workers, then alternative income distribution systems (universal basic income) make sense. 


Human capital-focused responses would be largely futile if AI is going to replace most workers anyway. 


Conversely, human-capital investment policies take on greater importance if AI mostly augments what workers do. 


The public policy implications are important, they argue.


A major implication is that human capital determines whether AI produces broadly-shared prosperity or rising inequality.


If more AI expertise increases the productivity gains from AI, while also reducing some inequality effects, 

policy should emphasize:

  • AI literacy across the workforce

  • vocational retraining

  • continuous mid-career education

  • managerial capability to integrate AI into workflows

  • higher-order cognitive skills that complement AI.


Also, productivity impact depends heavily on “thinly staffed tasks” (areas where too few skilled workers). 


Policy should therefore:

  • identify labor bottlenecks

  • accelerate training in constrained occupations

  • improve mobility into high-leverage roles.


In other words, AI inequality effects are not mechanically determined, but depend on:

  • workforce skill distribution

  • task allocation

  • education systems

  • how AI tools interact with existing human capabilities.


The paper argues AI systems:

  • help lower-skilled workers improve faster

  • diffuse expert knowledge

  • increase output without eliminating all human roles.


If AI augments human knowledge and skills rather than displacing humans, then public policy goals, then AI literacy will matter more than income replacement strategies in general.


Also, specific blockages in some fields hinge on issues other than AI. 


The obvious point, if the trend continues, is that AI might not be the job killer many fear.


Thursday, May 28, 2026

Uses and Misuses of Price's Law or Pareto Principle

The notion that a five percent to 10 percent reduction in force at a large organization might be "productivity-neutral" or even "productivity-positive" rests on the premise that large organizations eventually suffer from organizational entropy or “slack.”


In massive corporations, the individual contribution of an individual is notoriously difficult to measure, if it can be measured at all.


The Pareto Principle (80/20 rule) suggests that 80 percent of the value is produced by 20 percent of the employees. If this holds true, a 10-percent layoff that misses the "vital 20 percent" would, mathematically, have a negligible impact on total output.


Price's Law likewise suggests that half of organizational output is created by just 10 percent of workers.


But the idea can be carried too far. In other words, Pareto might suggest  where value is concentrated, but it does not tell us what parts of the value chain, in what quantities, we can try to remove.


In other words, complex products often require many value chain contributions whose contributions are outside the “80 percent of value” attribution, but are still essential for product success. 


The logic of eliminating most of the other value chain elements outside the “80 percent of value” only works only when inputs are independent and optional. That is rarely, if ever, the case for most products. 


For example, a particular  product ships only if every step is completed. So even low-value steps are non-optional constraints. 


Also, some value chain operations represent option value or risk reduction. They might not show up in the completed product, but might instead provide protection against product failure. 


Quality assurance efforts, regulatory compliance or maintenance might not be direct value creators, but might be necessary to deliver a final product. 


So many functions might be structurally necessary but individually have low marginal impact. And even that does not help address the question of staffing levels to support such processes. 


Were that not the case, competitive markets would force firms toward the Pareto-suggestion of minimal staffing. And we do not see that. 


Study / Source

Domain

Key Finding

Implication for Workforce

Pareto principle overview

Operations / Engineering

80% of outcomes driven by ~20% of causes

Output concentration is real

Pareto in supply chains (Slimstock)

Supply chain

20% of products drive most revenue

Inventory focus is uneven

Lean operations Pareto usage

Manufacturing

Majority of defects from small set of causes

Useful for prioritization

Juran Pareto analysis guide

Quality management

“Vital few and useful many” distinction

Many low-impact roles still necessary

IMD Pareto analysis strategy article

Strategy

Pareto helps focus leadership effort, not eliminate complexity

Tool for prioritization, not simplification

Value chain simulation research (VCS)

Manufacturing systems

Output depends on interdependent processes across cost, quality, delivery

System requires multiple linked roles

Pareto distribution empirical study (arXiv)

Economics/statistics

Heavy-tailed distributions common in real systems

Inequality of contribution is structural, not eliminative


That noted, Pareto does make sense for:

  • Prioritizing improvements (fix top 20 percent of defects)

  • Sales focus (top 20 percent of customers or products)

  • Time allocation (focus on high-leverage activities)


Still, Pareto notwithstanding, large organizations might often be less productive than imagined because of:

  • Social Loafing: In large groups, individuals often work less hard than they would alone because their lack of effort is easily hidden by the group's overall performance

  • Bureaucratic Friction: Beyond a certain size, organizations require "coordinators for the coordinators." Removing a layer of these roles can actually speed up decision-making, allowing the remaining staff to be more productive because they spend less time in meetings.


So when an organization has excess personnel, it can often absorb a reduction in force without losing core output, effectively "trimming the fat" to improve the output-per-employee ratio.


Study / Source

Key Focus

Source Link

Love & Nohria (2005)

Reducing Slack: Found that downsizing improves performance specifically when the firm has "excessive" resources (high slack).

Read at ResearchGate

Cascio, Young, & Morris (1997)

Financial Consequences: A landmark study showing that layoffs alone rarely boost ROA, but asset restructuring combined with cuts does.

Read at ResearchGate

Guthrie & Datta (2008)

Industry Context: Demonstrates that the negative impact of layoffs is significantly higher in "knowledge-intensive" (R&D) industries.

Read at ResearchGate

McKinsey & Company

The Productivity Imperative: Analyzes how technology and "de-layering" (removing management tiers) can boost service-sector productivity.

Download PDF (McKinsey)

Zyglidopoulos (2005)

Corporate Reputation: Examines how the market and stakeholders perceive downsizing as a signal of "efficiency-seeking" behavior.

Read at ResearchGate


The caveats are several.


Guthrie and Datta warn that in organizations with high innovation requirements, a 10-percent cut can remove critical "intangible assets" whose loss will not be seen until later. 


The Cascio study suggests that many firms fail to see long-term productivity growth after layoffs because they lose the ability to innovate


In other words, if the ideal is “cutting fat but not muscle,” the danger is “cutting some muscle as well.”.


If “productivity” is defined as total output divided by total input, then a smaller denominator “automatically” raises productivity, assuming output remains the same. 


The argument is that many large firms have enough "operational slack" (excess resources) that a five-percent to 10-percent  cut acts as a "forcing function," requiring the remaining staff to automate or abandon low-value tasks.


Large organizations are complex, so it might not be easy to determine how much, and where, to make layoffs. If one assumes every existing function actually is essential, then an “across the board” approach actually makes some sense.


The organization keeps the function, but possibly operates more efficiently. 


No Supplier Likes Customer Concentration, But Sometimes It Cannot be Helped

Customer concentration in the hyperscaler segment is practically unavoidable, when a handful of customers represent such a large percentage...