An independent study reference written by Dr Phuc V. Nguyen. It is not official subject material — for assessment requirements always follow your subject outline and vUWS.

Decisions and Modelsbeginner

What a decision actually is

Q: A team can reverse a website layout change in an afternoon. The proposal is to spend fifteen working days modelling its effect first. The strongest objection is that

the cost of the analysis is out of proportion to a choice that can simply be undone. Analytical effort should scale with the stake and with how hard the decision is to reverse. A change that can be rolled back in an afternoon is cheaper to try in the wild than to model at length, so the modelling buys certainty nobody needed and delays every other project in the queue.

Q: An analyst estimates a decision is worth A80,000 in expected value as it stands, and would be worth A200,000 if the outcome could be known in advance. A research firm offers a study for A$150,000. The right response is

decline, because even flawless information adds at most A120,000 of value. The value of information is the gain over the best decision you would already make, not the total payoff. Here the ceiling is A200,000 minus A80,000, which is A120,000, and no real study reaches the ceiling. A price of A$150,000 destroys value no matter how good the research turns out to be.

A decision is a commitment of resources to one course of action when other actions were available. Three consequences follow. If nothing you do changes depending on the answer, there is no decision, and analysis of it is decoration. A decision is made before the outcome is known, so a sound process can still produce a bad result and a reckless call can get lucky. Decisions also differ in structure and in reversibility, and those two features settle how much analysis is worth buying. Analytics earns its place only when it changes which option gets chosen.

Why it matters

Picture two doors. Walk through the first, dislike the room, walk back out. It cost you a few minutes, so try it and see. The second door locks behind you. Most choices at work are the first kind, and speed beats certainty. A few are the second kind, and that is where careful work pays for itself. Judging a choice only by how it turned out is like judging a bet after the card is turned over.

Before you read on — recall

A team can reverse a website layout change in an afternoon. The proposal is to spend fifteen working days modelling its effect first. The strongest objection is that

Formulas

Expected value of an alternative

EV(a) = \sum_{i=1}^{n} p_i \, v_i(a)

State

i

occurs with probability

p_i

and pays

v_i(a)

if you chose alternative

a

. Launching a product with a 40 per cent chance of earning A$500,000 and a 60 per cent chance of losing A$200,000 gives

EV = 0.4 \times 500 - 0.6 \times 200 = 80

thousand dollars. Doing nothing gives zero, so on this arithmetic you launch.

Expected value of perfect information

EVPI = \mathbb{E}\big[\max_a v(a)\big] - \max_a \mathbb{E}\big[v(a)\big]

The ceiling on what any study, survey or model can be worth. With perfect foresight you would launch only in the good state, worth an expected A$200,000. The best you can do without foresight is the A$80,000 above, so perfect information is worth at most A$120,000. Pay more than that for research and you have destroyed value before the work begins.

Worked examples

Scenario

A retail chain asks for a full customer profitability model before deciding whether to keep Sunday trading at four quiet stores. The team quotes thirty working days. Is that the right call?

Solution

Ask what the model would have to say to change the answer. Sunday closure is reversible within a quarter and the annual amount at stake is roughly A$300,000. A two-day check of Sunday margin per store, rostered staff cost, and the share of Sunday customers who shop no other day will settle it. The thirty-day model would earn its cost only if the decision were larger, harder to undo, or repeated across hundreds of stores. Matching the depth of analysis to the size and reversibility of the decision is the actual skill.

Scenario

A pricing analyst recommended a discount that lifted volume but cut total margin. A colleague acted on an unmodelled hunch the same quarter and it paid off. Who made the better decision?

Solution

On the information available at the time, the analyst may well have made the better decision and still lost. Judging quality by results alone rewards luck, and it teaches teams to stop writing down their reasoning. The repair is cheap. Record the options considered, the numbers assumed and the reason for the choice before acting, then review the process and the outcome as two separate questions. Across many decisions the good process wins, but you cannot see that unless you kept the record.

Common mistakes

✗A good decision is one that turned out well. Outcome quality and decision quality are different things, because the outcome also depends on chance and on information nobody had. Judge the process on what was knowable at the time, then judge the result separately.
✗Every decision deserves rigorous analysis. Effort should scale with what is at stake and with how hard the choice is to undo. Spending fifteen working days on something you can reverse next Tuesday is itself a poor decision about how to decide.
✗More data always improves a decision. Extra data helps only if it could change which option you pick. If your recommendation is identical under every plausible reading of the new evidence, that evidence carried no decision value.
✗Choosing not to act is not a decision. Doing nothing is an option with its own costs and risks, and it belongs on the same table as the active alternatives rather than being treated as the neutral default.

Revision bullets

•A decision commits resources to one option when others were open
•No alternatives means no decision, and nothing for analysis to change
•Decision quality and outcome quality are separate judgements
•Scale the analysis to the stake and to how reversible the choice is
•Data has value only when it could change the option you pick
•Doing nothing is an option and belongs in the comparison

Quick check

A team can reverse a website layout change in an afternoon. The proposal is to spend fifteen working days modelling its effect first. The strongest objection is that

An analyst estimates a decision is worth A$80,000 in expected value as it stands, and would be worth A$200,000 if the outcome could be known in advance. A research firm offers a study for A$150,000. The right response is

Connected topics

What Analytics Is Why Analytics Stalls Decision Process Decision Support A/B Testing KPI Design Data Storytelling Structuring a Problem Business Model Mapping Human-AI Agency

More in Decisions and Models

Models & Abstraction Supervised vs Unsup.

Sources

Simon (1955)
Simon, H. A. "A Behavioral Model of Rational Choice." Quarterly Journal of Economics, 69(1), 1955.
Sets out bounded rationality: real decision makers work under limits of information, time and attention rather than optimising over every alternative.
Gorry & Scott Morton (1971)
Gorry, G. A., & Scott Morton, M. S. "A Framework for Management Information Systems." Sloan Management Review, 13(1), 1971.
Classifies decisions as structured, semi-structured or unstructured, and matches each type to a different kind of system support.
Amazon.com 2015 Letter to Shareholders
Bezos, J. Letter to Shareholders. Amazon.com, Inc., 2015.
Popularised the one-way door and two-way door distinction: irreversible decisions deserve slow deliberation, reversible ones deserve speed.

How to cite this page

Dr. Phil's Quant Lab. (2026). What a decision actually is. Derivatives Atlas. https://phucnguyenvan.com/concept/ba-what-is-a-decision