How buying decisions actually form

Daniel Kahneman's framing of cognition splits it into two coupled processes:

Routine purchases — a coffee, a familiar brand of soap — run almost entirely on System 1. High-stakes purchases — a car, a mortgage — start on System 2 but slip back into System 1 as comparison costs rise.

A caveat that matters. The two-system metaphor is a useful description of behavioral patterns, not a literal claim about brain anatomy. The neural-localization evidence behind it has weakened in replication. What survives is the behavioral prediction: automatic responses dominate when stakes are low, time is short, or comparison is hard. The metaphor's value is for reasoning about when a buyer will deliberate and when they will pattern-match — not for diagramming neurons.

A heuristic is a rule that gives good-enough answers most of the time and predictably wrong answers in identifiable cases. Three durable ones from the Tversky-Kahneman tradition:

• Availability. Judging frequency by ease of recall. Plane crashes feel more common than car crashes because they make headlines.
• Representativeness. Judging probability by resemblance to a stereotype. The Linda-the-bank-teller problem: people rate "feminist bank teller" higher than "bank teller", violating basic probability.
• Affect. Judging by gut feeling first, then reasoning to support it. Positive affect raises perceived benefit and lowers perceived risk; negative affect does the opposite.

Heuristics are not bugs. They are the only way a brain decides thousands of times a day within its energy budget. The useful part is the predictable failure modes — knowing where a heuristic breaks tells you where a decision is manipulable, deliberately by an actor or accidentally by an interface.

Anchoring is the empirically robust observation that an arbitrary starting number shifts subsequent quantitative estimates — even when the anchor is obviously unrelated to the task.

The classic Tversky-Kahneman experiment: spin a wheel labelled 1–100 (rigged to stop at 10 or 65); ask subjects "what percentage of UN member states are in Africa". Subjects who saw 10 averaged 25%; subjects who saw 65 averaged 45%. The wheel carried no information about Africa, and subjects knew it.

Applied to commerce:

• MSRP framing. "$999 list,$799 today" anchors comparison at $999. A standalone$799 sticker invites a different anchor.
• Decoy options. Three plans at $9 /$19 / $29 anchor the middle as reasonable. Drop the$29 and the middle moves to the top of the visible range.
• Negotiation asymmetry. Buyers anchor on list, sellers anchor on cost. Outcomes track the first credible number on the table.

Anchoring replicates across cultures, expertise levels, and explicit warnings. Knowing about anchoring does not immunize against it — which is why it survives in consumer interfaces.

Kahneman and Tversky's 1979 prospect theory replaced expected-utility theory as the standard descriptive model of choice under risk. Three claims, all well-replicated:

1. Reference dependence. People evaluate outcomes as gains and losses relative to a reference point, not in absolute wealth. Your current salary anchors every other salary.
2. Loss aversion. Losses hurt roughly twice as much as equivalent gains feel good. The empirically fitted coefficient $\lambda$ sits between 1.5 and 2.5 depending on stakes and domain.
3. Diminishing sensitivity. Same delta hurts less at higher magnitudes; the value function is concave for gains and convex for losses.

A compact value function:

$V(x) = \begin{cases} x^{\alpha} & x \geq 0 \\ -\lambda (-x)^{\beta} & x < 0 \end{cases}$

Typical fitted values: $\alpha \approx \beta \approx 0.88$, $\lambda \approx 2.25$. The S-shape explains two real behaviors at once: risk-aversion in the gain domain (skip the coin flip for $100) and risk-seeking in the loss domain (gamble to avoid a sure loss). Casino design, insurance pricing, and product warranty framing all live in this asymmetry.

The same dollar amount feels different at different baselines. This is Weber-Fechner in the perceptual domain applied to money: subjective magnitude is roughly logarithmic in the stimulus.

$\text{perceived change} \;\propto\; \log(x_\text{new}) - \log(x_\text{old})$

A jump from $0 to$10 is one log unit. A jump from $1000 to$1010 is essentially zero log units, even though the dollar amount is identical. People haggle 20 minutes to save $10 on a$50 toaster and add a $10 cable to a$1000 laptop without thinking.

This compounds with reference dependence. Once the buyer has anchored at $1000, every additional dollar is evaluated relative to that frame — not relative to the buyer's bank balance. This is the mechanism behind upsell at checkout (the cart is the anchor) and bundle pricing (one big number is psychologically smaller than the sum of small ones).

The mechanism is neutral — the same logic makes buyers accept legitimate small extras (extended warranty, faster shipping) that they would refuse priced standalone. Awareness is the only neutralizer.

Iyengar and Lepper's 2000 jam-tasting study reported that fewer choices on display (6 jams) led to higher purchase rates (30%) than more (24 jams, 3%). The finding entered the canon as choice overload.

The replication record is mixed. A 2010 meta-analysis by Scheibehenne, Greifeneder, and Todd of 50 studies found a near-zero overall effect, with strong moderation: overload appears when options are similar, when stakes are high, when buyers lack expertise, and when the assortment is poorly organized. When options are differentiable (clearly varied flavors, price tiers) or the buyer is an expert, more options can help.

Three implications survive the replication critique:

• Filter, don't truncate. Faceted filters preserve agency without forcing exhaustive comparison.
• Default + extend. A curated short list with an explicit "see all" beats both a flat full list and a hidden one.
• Category structure beats raw count. 50 wines organized by region beat 12 wines listed alphabetically.

Frame choice overload as a contested-but-actionable mechanism — not a universal law.

The default-effect literature is among the most robust in behavioral economics. Johnson and Goldstein's 2003 cross-country comparison of organ-donation consent: opt-in countries (Germany 12%, Denmark 4%) vs opt-out countries (France 99%, Austria 99%). Same human biology, same medical systems — different default on the form.

Three reasons defaults work, each independently replicated:

1. Implied endorsement. A default reads as "the recommended option" — the Status Quo Bias (Samuelson & Zeckhauser, 1988).
2. Effort cost. Changing a default requires reading, deciding, and clicking. Each step shaves participation.
3. Loss-frame on departure. Moving off the default frames the change as a loss of the current state. Combined with loss aversion, the asymmetry is large.

The lever is small in proportional terms — one form field — and enormous in absolute terms. Defaults govern 401(k) enrollment (opt-out enrollment ≈ 3x opt-in participation), browser homepages, app-permission grants, and email-subscription tickboxes. In every case, the population that actively chose the default is a small fraction of those who landed there.

Thaler and Sunstein's term: choice architecture is the design of the option set. Order, framing, defaults, decoys, prompts — the structural decisions about how options are presented, before any of them is chosen.

The uncomfortable honest claim: there is no neutral choice architecture. Some option must be listed first. Some price must be the default tier. Some checkbox must start checked or unchecked. The designer either decides these deliberately or decides them by accident.

A short audit list for any decision interface:

• Default. What option is selected if the user does nothing?
• Anchor. What number or option is most prominent, and what does it imply about the range?
• Differentiation. Is the option set differentiable enough that adding choices helps?
• Frame. Are losses and gains presented symmetrically?
• Structure. Are there filters and sorts, or just a flat list?

Every pricing page, signup form, settings panel, and checkout flow is choice architecture. The mechanisms in this lesson are the vocabulary for reasoning about which architectures serve the chooser and which serve the architect — a distinction that matters for designers, buyers, and regulators alike.