AI Product Psychology

The Psychology of Pricing: Anchors, Decoys, and Decent Price Discrimination

The previous lesson covered how to charge without pain; this one covers how much. First remember a counterintuitive fact: users have no objective sense of price—every judgment comes from a reference. References can be arranged; this craft is a century old, and AI products just got a tool unique to them.

Economist experimentAnchor builderPrice-discrimination red linesPrice-hike email duel
Hands-on · First, subscribe to The Economist

This is the most famous pricing experiment in behavioral-economics history—Predictably Irrational opens with it. The Economist actually ran this subscription price list. First pick the tier you'd subscribe to, then see what MIT students chose.

Three tiers—pick one Round 1
Digital

Full-year online reading on Economist.com

$59
Print

Print magazine mailed all year

$125
Print + Digital

Print magazine mailed all year, plus online reading

$125
Theory in a minute · Anchors and decoys

Kahneman and Tversky showed that judgments about numbers get pulled by the number you saw first—even when that number came from a spinning wheel. That's anchoring. The decoy effect goes further: put an obviously bad deal among the options; nobody picks it, but it changes the relative attractiveness of the others. Print-only at $125 is the decoy: its only job is to make the “add $0, get digital” bundle look like a steal.

Every row on a pricing table is working—including the one nobody buys. It isn't dead stock; it's a salesperson.
Hands-on · Build anchors for your AI product

Your AI writing assistant has only one tier at ¥69/mo. Conversion is mediocre; ARPU won't climb. Flip the three switches on the right one by one and watch the pricing page grow an anchor structure step by step—the mid-tier pick rate below moves with it.

Pricing-page builder 0 / 3
Flip the switches on the right; the pricing page on the left and the metrics below change live
writer.example.com/pricing
Starter¥29/mo Base model · 200 runs/mo
3k chars per doc
Most popular · Chosen by 78% of users Pro¥69/mo Flagship model · Unlimited
Long docs · Style customization
Max¥299/mo Strongest reasoning model · Team collab
API access · Dedicated support
Users picking mid-tier (¥69)41%
Average ARPU¥47
Only two tiers for now: ¥29 and ¥69 reference each other—no anchor. Half of users feel ¥69 is “a bit expensive” and shrink down to Starter.
Hands-on · Judging the red lines of price discrimination

In economics, price discrimination is a neutral term: different willingness-to-pay pays different prices—welfare can be win-win. In user perception it splits two ways: the decent form is versioning; the career-ending form is screwing regulars. Five practices—judge one by one whether you'd ship them.

Five practices: ship it, or don't 0 / 5
The one-line test: if this became a news headline, could you stand behind it?
spot-the-difference · Price-hike email duel

Model costs fell, but you added features and decided to raise ¥69 to ¥89. Two notice emails—pick the one that won't trigger a cancel wave.

spot-the-difference: which email can you send and survive Pick a side
Email A
Price adjustment notice

Dear user:

Due to rising operating costs, starting on the 1st of next month, Pro pricing will change from ¥69/mo to ¥89/mo, effective from your next billing cycle.

Thank you for your understanding and support.

—— Product team

Email B
Pro is going up—but you get a 90-day grace period

Old friend:

Over the past six months Pro added three things: a reasoning-model upgrade, 200k-character long docs, and a style library. New-user pricing will move to ¥89/mo.

You're a long-time user: renew within the next 90 days and lock ¥69 for a year; do nothing, and after 90 days the new price applies.

—— Product team

Key Takeaways

Price feel is all reference: Nobody knows what an AI assistant “should” cost—judgment comes from the anchors you place. The row nobody buys is still working; it's a salesperson.

Three tiers is the standard structure: Starter catches the hesitant, the top tier is the anchor, mid-tier takes the bulk. The “Most popular” label is social proof—another nudge.

AI versioning is natural: Model tiers (base / flagship / strongest reasoning) have real cost differences users can feel—far more decent price discrimination than “feature castration.”

The red line is non-chosen identity: What users actively choose (student verification, annual pay, lower-tier) is decent; quietly repricing by user profile waits for the headline. Always give existing users a grace period on hikes.

Source: Original to Xiaoshan Academy's AI Product Psychology series; Economist decoy experiment from Ariely, Predictably Irrational (2008) ch.1, MIT classroom data (100 students); anchoring from Tversky & Kahneman (1974).