New Technology Often Requires Inventing New Interim Proxies for Financial Potential

When a new technology such as artificial intelligence creates new kinds of value, the traditional financial metrics (revenue, profit, return on investment) often fail to capture progress in the early years.

Instead, industries invent intermediate operating metrics: proxies that signal whether the new model is working before the business model is fully proven. Sometimes it works; sometimes it doesn't.

Lots of dot-com firms touted "eyeballs" as a measure of attention. Many competitive telecom firms used metrics such as "access line equivalents" (taking total bandwidth and breaking it into voice grade "line" equivalents) as an example of potential revenue upside.

These metrics usually measure one of three things:

• Adoption (how many people use it)

• Engagement or usage intensity

• Network growth or installed base. 

Stage	What Firms Measure
Early technology adoption	Installed base, users, traffic
Network growth	Engagement, interactions, ecosystem size
Monetization phase	Revenue per user, margins
Mature industry	Standard financial metrics

In the computing business, there are many examples. 

Technology Wave	Era	Early Operating Metric	What It Measured	Later Financial Metric That Replaced It	Firms
Personal computers	1980s	Installed PC base	Growth of computing platform	Software and hardware revenue	Microsoft, Apple
Dial-up Internet	Early 1990s	Subscribers / online accounts	Growth of consumer internet access	ARPU and subscription revenue	America Online
Web portals	Late 1990s	Page views	Traffic volume and advertising potential	Ad revenue per user	Yahoo
Dot-com era websites	1998–2001	“Eyeballs” (unique visitors)	Audience reach	Advertising revenue	Netscape ecosystem sites
Telecom data services	1990s–2000s	Access Line Equivalents (ALEs)	Aggregate network demand	ARPU and service revenue	telecom carriers
Search engines	Early 2000s	Queries per day	Demand for information retrieval	Revenue per search / ad revenue	Google
Social media	2005–2015	Monthly Active Users (MAU)	Network size and engagement	Ad revenue per user	Meta Platforms
Cloud computing	2010s	Compute instances / workloads	Adoption of cloud infrastructure	Revenue growth and margin	Amazon Web Services
SaaS software	2010s	Annual Recurring Revenue (ARR)	Predictable subscription base	Free cash flow and margin	Salesforce
Sharing economy	2010s	Gross bookings / rides	Platform usage volume	Take rate and net revenue	Uber
Streaming video	2010s	Subscribers	Platform scale	ARPU and operating margin	Netflix
Cryptocurrency	2015–2022	Wallets, hash rate, total value locked	Network security and participation	Transaction fees and financial services revenue	Coinbase ecosystem
Generative AI	2023–present	Tokens processed / active developers / API calls	Real workload demand	Revenue per model usage	OpenAI

Many could note a similar pattern for AI. New metrics emerge because we cannot typically measure early impact using traditional financial measures:

• Monetization lags adoption

• Network effects require scale first

• Investors need forward-looking signals, so usage metrics answer that question before profits exist.

Phase	Typical Metric
Technology novelty	Install base
Early growth	Users or traffic
Platform stage	Engagement
Business model maturity	Revenue per user
Mature industry	Profitability

The AI economy therefore creates new metrics in the interim:

• Tokens processed

• Active developers

• Inference workload

• Model training compute. 

These resemble earlier indicators in the early internet era such as:

• page views (web)

• queries (search)

• Monthly active users (social media)

Eventually the industry will likely shift to measures more closely tied directly to firm profits and revenues:

• revenue per AI workload

• enterprise productivity gains

• profit margins on AI services.

Eventually, we’ll learn which operating metrics actually have higher predictive value, and which have less. 

During the dot-com bubble around the turn of the century, some metrics turned out to have near-zero predictive value.

Company	Metric Highlighted	What the Metric Measured	Why It Was Misleading
Pets.com	Website traffic / brand awareness	Consumer interest in online pet supplies	Traffic did not translate into profitable orders because shipping costs exceeded margins
Webvan	Number of cities launched	Geographic expansion of grocery delivery infrastructure	Massive capital spending occurred before proving unit economics
eToys	Revenue growth rate	Rapid expansion of online toy sales	Sales were heavily subsidized by marketing and discounting
TheGlobe.com	Registered users	Size of social community platform	Users were mostly non-paying and generated little revenue
Boo.com	Site engagement and global launch presence	Interest in online fashion retail	Extremely expensive website technology created slow performance and high operating costs
Excite	Page views	Web portal traffic volume	Advertising demand could not support the scale of infrastructure spending
Lycos	Unique visitors	Audience size of web portal	Monetization per visitor was extremely low
Broadcast.com	Streaming traffic and media partnerships	Growth of internet audio/video streaming	Technology and bandwidth costs exceeded realistic revenue models
Priceline (early phase)	Gross travel bookings	Total value of transactions handled	Gross bookings overstated the company’s actual revenue capture
Drkoop.com	Health site visitors	Consumer interest in medical information	Advertising revenue insufficient to support operations