Field Note · No. 01 · decision science · 12 min read
The brain assigns value before the thinking begins. An agent can be taught the same arithmetic, and the clearest returns show up in pricing.
abstract
Neuroeconomics is the study of how brains actually assign value and make choices, measured directly rather than assumed. Fifty years of this work, from prospect theory to the dopamine reward signal, has produced a small set of findings that replicate well and matter commercially: people feel losses about twice as strongly as gains, they judge every price against a reference point, and the first number they see moves the last number they accept.
Most companies know none of this is in their software. Their pricing pages, offers, renewal emails, and negotiations are written as if customers were calculators. This note explains the science in plain terms, then shows how I encode it into AI agents through three layers: the instructions an agent carries, the worked examples it learns from, and the evaluations that score its output against the research.
The return is measurable and shows up fastest in pricing, where decades of published effects translate directly into offer structure, price presentation, and framing. I include the numbers, the sources, and the honest limits.
01 · what neuroeconomics actually is
Economics long assumed that people compute value rationally and choose whatever maximizes it. Neuroeconomics went looking for that computation in the brain and found something different. The field sits at the junction of economics, psychology, and neuroscience, and its founding observation is that valuation is a biological process with measurable, predictable quirks.1
Two findings anchor everything else. First, the brain carries a dedicated valuation signal: dopamine neurons fire not to reward itself but to the difference between the reward you got and the reward you expected.2 Expectations are not commentary on value; they are the substrate of it. Second, choices are made relative to a reference point, not on absolute outcomes. Daniel Kahneman and Amos Tversky demonstrated this in 1979 with prospect theory, which is still the most cited paper ever published in Econometrica.3 A $100 loss and a $100 gain are the same size to an accountant and very different sizes to a brain.
The practical summary is one sentence: the brain decides before the thinking begins, and it decides using reference points, expectations, and an asymmetric fear of loss. Everything commercial in this note follows from that sentence.
figure 1 · the prospect-theory value function · interactive
Drag λ to see how much steeper losses feel than gains. Tversky and Kahneman measured the median person at roughly 2.25.4 At λ = 1 the curve would be symmetric, and nobody’s brain works that way.
02 · the findings that matter commercially
The literature is large, but the effects that survive replication and show up in revenue are a short list. Loss aversion: losses are weighted about 2.25 times as heavily as equivalent gains, which is why “don’t lose your progress” outperforms “keep your progress” and why cancellation flows behave so differently from signup flows.4 Anchoring: the first number in a negotiation or on a pricing page drags every later judgment toward it, even when the number is arbitrary.5 Framing: describing the same outcome as a loss instead of a gain flips a majority of people’s choices, and the flip is visible in the amygdala while it happens.6
Three more complete the commercial set. Fairness is priced in the brain: people reject profitable deals that feel insulting, and the disgust region that lights up when they do is the same one that responds to bad smells.7 Price itself changes the experience: when researchers told subjects a wine cost $90 instead of $10, the brain region that encodes pleasantness responded more strongly to the identical wine.8 And relative position beats absolute value: adding a deliberately inferior option to a menu reliably shifts choices toward the option it was designed to flatter.9
figure 2 · the decoy effect · interactive
With the useless middle option present, 84 of 100 MIT students chose the premium bundle. Remove it and watch the same menu sell the opposite way. Classroom experiment reported by Dan Ariely.10
03 · how you train an agent on this
A clarification first, because I promised you plain language: nobody is retraining a frontier model’s weights here, and anyone who tells you otherwise is selling something. The model already read the literature. Training your agent means making the science operational, so the agent applies it by default instead of knowing it in the abstract. I do that in three layers.
The first layer is the instructions. An agent’s standing instructions carry the decision rules as procedure: establish the customer’s reference point before writing a price, present the anchor before the ask, frame retention in losses avoided and acquisition in gains, check every offer set for its relative structure, and flag any message that could read as unfair before it ships. The second layer is worked examples. A rule tells; an example trains. Each agent carries a small library of before-and-after cases from real engagements: a pricing page restructured around an anchor, a renewal email reframed from discount to loss avoided, a proposal whose three tiers were rebuilt so the middle one wins. The third layer is evaluation. Every output is scored against the science before a human sees it: did it anchor, what is the implied reference point, where is the decoy, what does the fairness read say? Anything that fails is regenerated with the failure named.
figure 3 · one rule from a pricing agent’s instruction set
pricing_rules:
reference_point: >
Never present a price without first establishing what the
buyer currently pays, loses, or forgoes. The price is judged
against that number, not against zero. If no reference
exists, build one from their own data before the quote.
anchor_order: state the anchor before the ask, always
frame_check: retention copy frames losses avoided;
acquisition copy frames gains
fairness_gate: if the offer could feel insulting to a
reasonable buyer, it fails before it ships
04 · what you get back, with pricing as the sharpest case
Pricing is where the training pays first, because pricing is where the published effects are largest and the change is cheapest to make. An agent trained this way audits a pricing page in minutes against effects that took researchers decades to establish: whether the anchor appears before the ask, whether the tier structure gives the brain a relative comparison it can win, whether the highest price is doing its silent work as a reference point, and whether the number itself fights the left-digit effect, which is why $49 and $50 perform differently by far more than a dollar.11
The same training compounds beyond the pricing page. Renewal and cancellation flows get rewritten around loss aversion, which is where the 2.25 multiplier lives. Proposals get anchored and tiered instead of quoted flat. Negotiation prep gets a fairness read before the first call. Product pages inherit the price-as-signal finding: a premium price presented without premium framing is a contradiction the buyer’s brain resolves against you. None of this requires new products or new customers. It re-prices attention you already have, which is why it is usually the first system I point an agent at inside a build.
05 · the honest limits
Three cautions, stated plainly because the science deserves them. The effects are real but context-dependent: published sizes come from specific populations and settings, and your customers are neither MIT students nor wine tasters in a scanner, so every change we make is measured against your own baseline rather than assumed from the literature. The ethics line matters: these findings clarify value or they manipulate, depending entirely on whether the product delivers what the framing promises, and I only build the first kind, partly because the second kind also destroys lifetime value once buyers notice. And measurement is not optional: any agent I train on this science is also scored by it, with changes tested against pre-registered baselines, because a persuasion technique you cannot measure is a story, not a system.
references
Put it to work
If you want an agent trained this way on your own numbers, the first step is the same as every Modven engagement: one conversation and a written brief, $500, credited in full if we build.