AI Pricing Experiments
AI pricing experiments test how to price AI products themselves and how to use AI to test pricing on non-AI products. The two are different sports. For pricing AI products: the canonical pattern is OpenAI's tier experimentation — Free, Plus ($20), Pro ($200), Enterprise. Each tier tests willingness-to-pay against feature differentiation. For using AI to optimize pricing on other products: the pattern is Adobe-style ML personalization, where prices are tested per segment with bandits or A/B tests against a holdout. In both cases the trap is changing pricing without measurement infrastructure to detect cannibalization.
The Trap
The trap is launching a new AI tier (e.g., 'Pro at $200/mo') without a measurement plan to detect cannibalization from the existing tier. OpenAI's $200 Pro plan was an experiment to extract surplus from power users — but if 30% of $200 buyers were prior $20 Plus subscribers churning UP, that's cannibalization to net positive; if 30% of NEW $20 Plus signups were diverted DOWN from Pro because of confusion, that's cannibalization to net negative. Without segment-level cohort tracking, you don't know which.
What to Do
Before any AI pricing change, instrument three measurement layers: (1) Cohort revenue per visitor — total revenue / visitors who saw the new pricing, (2) Tier mix shift — % of new signups by tier vs. baseline, (3) Net new MRR — accounting for cannibalization. Run the change as a 50/50 split for 4 weeks minimum. Real signal usually emerges at week 3 because of consideration cycles. Don't trust week 1 numbers — they over-index on power users.
Formula
In Practice
OpenAI launched ChatGPT Pro at $200/month in December 2024 as a tier experiment, sitting alongside Plus at $20. Sam Altman publicly stated they were losing money on Pro because power users consumed more inference than expected — demonstrating the value of price experimentation: even at a 10x price multiple, willingness-to-pay among heavy users exceeded marginal cost in ways the company hadn't fully modeled. The experiment generated learning that retroactively reshaped capacity planning.
Pro Tips
- 01
Test price up before testing price down. It's almost always reversible — you can drop a price next quarter — but raising prices that you've previously dropped destroys trust.
- 02
Bandit algorithms (Thompson Sampling, contextual bandits) are appropriate for personalized pricing on commodity SKUs but inappropriate for SaaS subscriptions because the customer notices the variation. Use bandits for one-time purchases; use A/B tests for subscription tiers.
- 03
Always grandfather existing customers. A 'price up' that backfills to existing customers triggers churn that wipes out the new pricing's gains. Adobe learned this in 2012 with the Creative Cloud transition — the customer outrage was a multi-year headwind.
Myth vs Reality
Myth
“AI lets you optimize pricing in real time per customer”
Reality
It can — but most contexts don't reward it. SaaS customers compare notes; B2B customers procurement-audit your pricing. Real-time personalized pricing is appropriate for hotels, airlines, and ride-share — almost nothing else.
Myth
“A pricing test has clear winners after a week”
Reality
Pricing has long consideration cycles. A 1-week test over-weights impulsive buyers. Minimum 4 weeks for SaaS; 8+ weeks for enterprise. Anything shorter is noise.
Try it
Run the numbers.
Pressure-test the concept against your own knowledge — answer the challenge or try the live scenario.
Knowledge Check
You launch a $200 'Pro' tier alongside your existing $20 'Plus' tier. After 30 days: 200 Pro signups, but Plus signups dropped from 5,000/mo to 4,200/mo. What's the most likely net MRR impact?
Real-world cases
Companies that lived this.
Verified narratives with the numbers that prove (or break) the concept.
OpenAI
December 2024
OpenAI launched ChatGPT Pro at $200/month, a 10x premium over the existing $20 Plus tier. Sam Altman publicly admitted on X that the company was losing money on Pro because power users consumed more inference than the price covered — a candid signal that the tier was an experiment in extracting surplus from the heaviest users, not a settled pricing decision. The experiment yielded reusable knowledge: capacity planning changed, and subsequent product tiers (Sora, Operator) priced into the Pro bundle reflected the learning.
Pro Tier Price
$200/mo
Plus Tier Price
$20/mo (10x lower)
Disclosed Outcome
Losing money — tier underpriced for heavy users
Pricing experiments are valuable even when they 'fail' financially short-term — they reveal willingness-to-pay segments and consumption patterns that are otherwise invisible.
Adobe (Sensei AI in Creative Cloud)
2017-present
Adobe uses ML (Sensei) to test promotional pricing and bundle composition across Creative Cloud customer segments. Personalized retention offers, free-month promos for at-risk users, and bundle upgrades are tested against held-out controls before being rolled out broadly. The discipline came from the painful 2012 Creative Cloud transition, when a forced subscription move triggered widespread customer backlash — Adobe rebuilt its pricing testing infrastructure to never repeat that.
Pricing Test Cadence
Continuous
Methodology
Personalized offers + holdout
AI-driven pricing personalization works best for retention offers, not net-new acquisition pricing. Customers don't compare retention offers; they do compare list prices.
Related concepts
Keep connecting.
The concepts that orbit this one — each one sharpens the others.
Beyond the concept
Turn AI Pricing Experiments into a live operating decision.
Use this concept as the framing layer, then move into a diagnostic if it maps directly to a current bottleneck.
Typical response time: 24h · No retainer required
Turn AI Pricing Experiments into a live operating decision.
Use AI Pricing Experiments as the framing layer, then move into diagnostics or advisory if this maps directly to a current business bottleneck.