AI Product Psychology

Honeymoon Cliff: Hype Raises Expectation—Retention Pays It Back

Keynote demos look miraculous; day one is thrilling; day three starts nitpicking; day thirty, uninstall. You’ve seen this curve on countless AI products. Two causes: hype pulls expectation up, and novelty fades on its own. Both can be designed.

GachaHype-intensity sliderExpectation-curve editorExpectation-confirmation theory
One formula · Satisfaction is subtraction
Satisfaction = actual experience − prior expectation
Expectation-confirmation theory (Oliver, 1980)—forty years of consumer-satisfaction research stand on it.
Experience is built with fixed cost; expectation is lifted by one marketing line. Raising expectation is free; paying it back is costly.
Hands-on · Gacha: demo slot vs. real use—where’s the gap

AI products are born behind on this subtraction problem, and the reason hides in probability. Left is the keynote demo slot; right is real user use. Both hit the same model—tap Generate five times and see what each side draws.

Same model, two viewpoints 0 / 5 draws
Tap “Generate once” below—five times in a row
launch-keynote.example.com/demo
Keynote demo slotEdited for broadcast
S-tier output
Waiting to generate…
app.example.com/workspace
Real user useWhat you see is what you get
Waiting to generate…
Demo S-tier 0 · Real-use S-tier 0
See it now. The demo slot lights up S-tier every time—because it’s picked from dozens of runs, then edited. Real use gets the whole distribution: S, B, and faceplants. Marketing shows P99; users experience P50; the gap all lands on “this product doesn’t work.” Traditional software doesn’t have this: demo click → report; user click → same report. Marketing a probabilistic good comes with an expectation bubble built in.
Hands-on · Hype-intensity slider: conversion vs. retention seesaw

If expectation is lifted by marketing, how high becomes a product decision. Left is your landing page—drag the slider through five copy tiers; right, three metrics move live. Find the tier where conversion × retention peaks.

Five hype tiers, three metrics Drag me
Tier 3
meetnote.example.com
Free trial
Signup conversion
Day-1 satisfaction
30-day retention
Novelty fades · The other half of the cliff

Even with expectation managed, another drop awaits: novelty fades on its own. Ed-tech research calls it the novelty effect: a new tool looks strong when it first enters the classroom, then falls back in weeks—because part of the gain was “new” itself. AI honeymoons are especially short: first poem is magic, tenth is a feature, hundredth is owed, one miss is garbage.

Magic is a consumable. Propping retention on first-wow is heating your house with fireworks.
Hands-on · Expectation-curve editor: three levers you flip live

This is the 30-day satisfaction curve after signup. Current state: hype maxed, day-one capability fully lit, improvements dripped quietly—classic honeymoon cliff. Flip the three levers on the right one by one and watch the curve get caught, segment by segment.

30-day satisfaction curve 0 / 3
Flip switches on the right—the curve redraws live
Happy Let down Day 1 Day 15 Day 30 Churn danger line
Pick one · Which line doesn’t overdraw
Homepage copy for an AI legal assistant—which least overdraws expectation? Single choice
One test only: can the product’s P50 output carry the expectation this line sets
A“Your AI lawyer—knows the statutes better than humans”
B“Contract first-pass 10× faster; risk clauses auto-flagged; final call stays yours”
C“Legal AGI is here—say goodbye to lawyer fees”
D“Smart legal assistant—empowering a new paradigm for legal ops”
Key Takeaways
✓

Satisfaction = experience − expectation: raising expectation is free; paying it back is costly. Forty years of expectation-confirmation theory haven’t been overturned.

✓

Demos cherry-pick the distribution’s tip: marketing shows P99, users get P50—probabilistic-good marketing carries an expectation bubble by nature, and the gap all lands on the product.

✓

Magic is a consumable: novelty fades on its own; propping retention on first-wow is heating with fireworks.

✓

Three levers catch the curve: promise one notch below (copy writes P50), unlock capability gradually (advanced features schedule-unlock), bank improvements into felt releases (changelog is a free second honeymoon).

Source: Original to Xiaoshan Academy's AI Product Psychology series; expectation-confirmation theory from Oliver, A Cognitive Model of the Antecedents and Consequences of Satisfaction Decisions (1980); novelty effect is a standard finding in educational-technology research.