AI Valuation Two-Path Test
Test whether AI spending creates new revenue or defensible cost savings
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 5
- Confidence
- 94%
Scott Galloway frames current AI valuations as requiring one of two broad outcomes: customers must generate substantial incremental revenue from AI-enabled products, or they must create large, durable cost savings that flow to profit. The test begins with the return implied by the investment, then looks separately for evidence on each path. Adoption, impressive capability, or layoffs alone do not establish that the required return exists. Galloway also models falsifiability in his employment argument by naming sustained net job destruction as evidence that would prove him wrong. The method therefore combines valuation math, operational evidence, and a predeclared disconfirmation test. It does not imply that AI lacks importance; his larger point is that a seminal technology and an attractive shareholder return are different claims.
Origin
Extracted from The Diary of a CEO
Core principles
- 01Large valuations require corresponding economic returns
- 02Revenue growth and cost savings are distinct proof paths
- 03Executive forecasts should be checked against observed outcomes
- 04Important technology can still be a poor investment
- 05A thesis needs a stated condition that would prove it wrong
How to run it
- 1
State the implied return
Translate the valuation or spending commitment into the scale of revenue, margin, or cash flow needed to justify it. Use a defined time horizon rather than an open-ended promise.
Pro tip Write the required outcome before reviewing management's narrative.
Watch out Do not treat technical importance as proof of shareholder value.
- 2
Test the revenue path
Look for new products, higher sales, or other incremental revenue that customers can reasonably attribute to AI. Distinguish revenue creation from merely adding AI to an existing workflow.
Pro tip Ask what customers can sell now that they could not sell before.
Watch out Usage and site-license growth do not by themselves prove customer return.
- 3
Test the efficiency path
Measure labor, cycle-time, or operating savings and whether they persist. Check whether savings reach the bottom line or are reinvested elsewhere.
Pro tip Use completed work and audited costs rather than announced intentions.
Watch out A layoff announcement does not establish that AI caused the reduction.
- 4
Check outside indicators
Compare the story with employment, productivity, business formation, and customer outcomes. Look for both confirming and contradictory data.
Pro tip Separate industry-specific disruption from economy-wide effects.
Watch out The episode cites figures without providing underlying sources, so verify them independently.
- 5
Declare the falsifier
Record the observable outcome that would change your conclusion, such as sustained net job destruction or revenue failing to materialise. Revisit the thesis when that evidence arrives.
Pro tip Set a review date alongside the falsifier.
Watch out Do not move the test after the evidence turns against the thesis.
In the wild
A consumer company considers a large AI license. The team calculates the revenue or recurring savings required over three years, then tracks whether AI launches genuinely new products or reduces verified operating costs. If neither path approaches the required return by the review date, it reduces the commitment rather than defending it with adoption metrics.
→ The purchase is judged by attributable economic return rather than the prestige of using AI.
Galloway says his view that AI will create more jobs than it destroys would be wrong if job destruction persists and new businesses and roles fail to keep pace. He contrasts that condition with the employment data he cites during the conversation.
→ A broad forecast becomes a testable claim with evidence that could overturn it.
Common mistakes
Equating importance with returns
A technology can transform society while competition prevents a small group of companies from capturing the value.
Counting adoption as proof
Licenses, model use, and capital spending are inputs; the test requires attributable revenue or savings.
Leaving the thesis unfalsifiable
Without a stated disconfirming condition, every outcome can be reframed as support for the original story.
Is it for you?
Best for
People assessing an AI company, enterprise license, or infrastructure investment with ambitious growth assumptions.
Not ideal for
It cannot predict every second-order benefit or replace detailed financial, technical, and competitive due diligence.
From the transcript
“There either needs to be a trillion dollars in incremental revenue from new products from companies that have licensed AI.”
“The technology will absolutely survive. I do think it's seminal. It's breakthrough.”
From the episode
Scott Galloway: AI Wasn’t Built For You. The Rich Don’t Need You Anymore!