Market & Consumer Insights · Dialogue AI × The Observatory

Sell the rulebook, not the personas.

A segmentation study that did not produce segments. Here is the structure that is actually in the data, sized where the data earns it, and what it means for positioning and messaging.
FLIGHT TUP-495 · 495 PAX · IQS 3.76/5 · 3 ROUNDS · BOARDING
01 · The proof, first

There are no natural segments.

Cluster validity never clears the floor
k-means silhouette, 5 calibration items, n=365
0.000.250.500.25 REAL-CLUSTER FLOOR2345678SEGMENTS (k)

At every segment count from 2 to 8, the separation score sits near 0.20, under the 0.25 line that distinguishes a real group from an arbitrary slice of a continuum. Ward linkage and BIC agree. Adding more variables makes it worse. Why the Ns differ: clustering uses complete-case ratings (n=365); each rating chart shows its available cases (n=384 to 425); prevalence uses all coded transcripts (n=486).

0.21
peak silhouette (need ≥ 0.25)
56%
variance in just two continuous dials
For a data audience: this is the headline. Any four-box segmentation built from data like this is lines drawn on a cloud.
02 · What is really there

The shapeTwo dials, one cloud.

Every traveler, placed on the two dials
PCA · n=365 · colored by imposed quadrant (see §04)
STAKES INTENSITY →GUARD ↔ DELEGATE

A continuous smear with no breaks. People sit everywhere along both dials, and the same person slides along them by trip context.

The variation that exists is two continuous dimensions on a gradient:

DIAL 1: STAKES INTENSITY

How much any of this matters to a traveler on this trip. 34% of variance.

DIAL 2: DELEGATE ↔ GUARD

Lean toward letting the airline handle it, or toward guarding and keeping score. 22% of variance.

03 · The shared rulebook

Prevalence floors · n=486The loud signal is shared across everyone.

Predictability premium70%Refuses control/delegate binary62%Loyalty = they didn’t fail me61%Violation = not-knowing59%Movable dial (context-switch)57%All-in-price / fee morality42%Two-option load cap40%

These are presence floors (how many surfaced each), not survey incidence. The top five are cross-cutting rules, which is exactly why clustering on them fails: almost everyone follows them.

"I want like the airline to take care of it for me, but I also want to be presented with options [...] So I want to have both."P480 · 35-44 · Wichita, KS · refuses the binary
"Just tell us the truth [...] instead of just keeping everyone in the dark."P490 · 55-64 · Overland Park, KS · the violation is silence
04 · The calibration battery

Where it converged instead of separatingTwo of five questions ceiling out.

Pay more to avoid uncertainty
n=425 · 1-7 scale
1234567mean 5.47
Too many options = stressful
n=410 · 1-7 scale
1234567mean 3.62
Trust airline to decide
n=405 · 1-7 scale
1234567mean 3.69
Hidden fees > upfront price
n=392 · 1-7 scale · CEILING
1234567mean 6.09
Trust, once lost, hard to regain
n=384 · 1-7 scale · CEILING
1234567mean 5.65

The two coral histograms are pinned to the ceiling: almost everyone agrees. A question everyone answers the same way cannot separate a market. Only options-stress and trust-airline carry real spread, and even those move within a person.

05 · The navigable overlay

Use with the caveat attachedIf you must have a 2×2.

These four quadrants are an imposed median split on the cloud, offered only as a creative-routing convenience. They are not validated segments (just over half individual-assignment reliability), and the near-even sizes with no gaps are themselves a sign of a gradient. Never present them as real.
99/365 · 27%

Trusting Delegator

Creative cue: “We've got it handled, here's the plan”

96/365 · 26%

Detached Pragmatist

Creative cue: “No drama, no surprises, just get there”

86/365 · 24%

Easygoing Delegator

Creative cue: “Relax, we'll sort it”

84/365 · 23%

Vigilant Guardian

Creative cue: “Full transparency, you stay in the loop”

06 · Message × need

Messages map to needs everyone shares.

MessageGrounding needSizeWhy it works
You'll always know what's happeningViolation = not-knowing59% floorThe single biggest emotional lever
One all-in price, no surprisesFee anger ceilings 6.09/7near-universalCustomers read it as trust
Pay a little, skip the hassleBounded predictability premium70% floorThe default mode, price-capped
We’ve never let you downSpecific retrospective trust61% floorLoyalty = non-failure + status
Two or three clear options, not tenTwo-option load cap40% floorMore choice reads as abandonment
07 · The recommendation

Two promises, aimed at everyone.

Bet 01

End the silence

Own the disruption-comms moment: proactive, honest, frequent. It is universal, so you target everyone, and it is the largest single emotional lever in the dataset.

Bet 02

One all-in price

Reframe transparent pricing as a trust play, not a discount. Hidden fees read as betrayal at a near-universal rate.

Check out other slices
Traveler Under Pressure · Dialogue AI × The Observatory · 495 AI-moderated interviews