“How to Measure Anything: Finding the Value of Intangibles in Business” is Douglas Hubbard’s argument that everything you’d want to know—employee morale, the value of an IT project, brand reputation, risk of a project failure—can be measured, at least well enough to inform a decision. The subtitle carries the thesis: measurement doesn’t require perfect precision, just enough reduction in uncertainty to make a better choice than you’d make without it.
Core ideas
Measurement is uncertainty reduction, not perfect knowledge. Hubbard defines measurement as a quantitatively expressed reduction in uncertainty based on observation. You don’t need to know something to the decimal point; you need to know it well enough that your range of estimates narrows and your decision improves. This reframes “immeasurable” things — customer satisfaction, the value of flexibility, quality of life — as measurable in principle, because if something matters to a decision, it has observable consequences, and if it has observable consequences, there’s a way to detect and quantify those consequences.
Three “measurement myths” he tries to knock down:
- It’s been done already (someone likely has a method — literature exists on “soft” variables like happiness, freedom, or security).
- You need lots of data (Hubbard leans hard on how much a single data point, or a very small sample, reduces uncertainty compared to zero data — his “rule of five” is a good example: five random samples from a population give you a 93% chance the median falls between the smallest and largest values).
- You need very precise data (in most business decisions, reducing uncertainty a little is worth far more than the cost of achieving it — the expected value of information often plateaus quickly).
Calibrated estimation. A big chunk of the book is about training people (via calibration exercises — trivia questions with confidence intervals) to give honest subjective probability ranges instead of false precision or vague hand-waving. Most people are either overconfident or underconfident estimators until they practice; Hubbard claims this skill is trainable in a few hours.
Applied information economics. Hubbard’s broader consulting framework: model the decision, quantify current uncertainty as a range/probability distribution, calculate the “expected value of perfect information” (or partial information) for each variable, then focus measurement effort only on the variables where more information would actually change the decision. This is the book’s most practically useful move — it stops people from measuring things that don’t matter and ignoring the one or two variables that do.
Monte Carlo simulation gets used throughout as the tool for propagating uncertainty through a model rather than relying on single-point “best guess” estimates that hide the range of outcomes.
Reception
It’s popular in tech, product management, and risk analysis circles (people building out estimation or decision-analysis practices cite it constantly). Critics say the calibration/EVPI machinery is more straightforward to describe than to execute well in a messy organization, and that some of his “we measured X” anecdotes are thinner on methodological detail than the book’s confidence suggests. Still, as an introduction to decision analysis and a corrective to “that’s just a soft/intangible thing, we can’t measure it” thinking, it’s held up well and spawned a lot of applied work (Hubbard has follow-on books on risk management and the “Failure of Risk Management”).
Part 1: The Measurement Solution Exists
Chapter 1-2: “The Challenge of Intangibles” / “An Intuitive Measurement Habit”
Hubbard opens with the claim that sits under the whole book: anything you care about has some observable effect on the world, otherwise you wouldn’t care about it or know it exists. If it has effects, those effects can be observed, and if they can be observed, they can be measured in some way, even imperfectly. He introduces the “Ultimate Measurement Scale” — a joke that lands a serious point: even “impossible” measurements like the number of fish in a lake, or the value of a human life used in policy decisions, get estimated routinely using indirect methods.
He also draws a hard line between measurement and precision. His formal definition: measurement is “a quantitatively expressed reduction of uncertainty based on one or more observations.” Note what this doesn’t require — it doesn’t require a single, exact number. A reduction from “somewhere between 0 and 10 million dollars” to “somewhere between 2 and 4 million dollars” is a measurement, even though it’s still a range.
Part 2: Before You Measure
Chapter 3: “The Illusion of Intangibles”
This is where he dismantles four specific reasons people give for why something can’t be measured:
- Concept (we don’t even know what “quality” means) — solved by clarifying what you actually mean and what decision hinges on it.
- Object (we don’t know what thing we’re measuring) — same fix.
- Method (we don’t know how to measure it) — this is what most of the book addresses.
- Error (we’re worried about being wrong) — his point: any reduction in uncertainty beats none, and decision-makers already act on far worse information daily.
Chapter 4: “Clarifying the Measurement Problem”
Key tool: “What is your uncertainty about, and what would you do differently depending on the answer?” If you can’t answer that, you don’t need to measure it. This chapter’s real deliverable is the idea that you already have more data than you think, and the measurement should be aimed squarely at the decision, not at academic completeness.
Part 3: Measurement Methods
This is the technical core.
Chapter 5: “Calibrated Estimates: How Much Do You Know Now?”
People are asked to give 90% confidence intervals on trivia questions (e.g., “the year the Great Wall of China was started”). Most people’s intervals are way too narrow — they’re overconfident, and their true “hit rate” is often 50% instead of the claimed 90%. Hubbard’s team has run these exercises for years and found calibration training (with feedback, repeated practice, techniques like the “equivalent bet test” — would you rather bet on your range being right, or spin a wheel with a 90% chance of winning?) reliably improves people’s estimates within a few hours.
Chapter 6: “Quantifying Risk Through Modeling”
Instead of single-point estimates (“this project will take 6 months”), use ranges and probability distributions, then run a Monte Carlo simulation — sampling thousands of times from each input’s distribution to see the resulting distribution of outcomes. This surfaces risk (e.g., “10% chance this project loses money”) that a single-point spreadsheet estimate hides entirely.
Chapter 7: “Quantifying the Value of Information”
This is arguably the book’s most distinctive contribution: the Expected Value of Perfect Information (EVPI). For each uncertain variable in a decision model, you can calculate how much it would be worth to know that variable perfectly. Variables with high EVPI are worth measuring further; variables with near-zero EVPI (because they wouldn’t change the decision even at their extremes) aren’t worth another dollar of research. This inverts how most organizations approach measurement — they tend to measure what’s easy or customary, not what’s decision-relevant.
Chapter 8: “The Transition: From What to Measure to How to Measure”
Bridges theory to practice — a checklist-style approach to picking a measurement method given what you now know about your uncertainty and the value of reducing it.
Chapter 9: “Sampling Reality: How Observing Some Things Tells Us about All Things”
Covers basic statistical sampling but leans on his Rule of Five: take five random samples from any population, and there’s a 93.75% chance the true median lies between the smallest and largest of those five values. It’s a rhetorical device as much as a statistical one — meant to show that even tiny samples beat zero samples by a lot, countering the “you need a huge dataset” myth. He also covers simple techniques like the “Rule of Three” (roughly 95% chance the true population median lies below the max of three random samples, useful for one-sided bounds).
Chapter 10: “Bayes: Adding to What You Know Now”
Introduces Bayesian updating as a formal way to revise estimates as new data comes in, rather than treating each new measurement as if you started from zero knowledge. He covers subjective Bayesian methods and the idea of using your calibrated prior as the starting point.
Chapter 11: “Preferences and Attitudes: The Softer Side of Measurement”
Applies the toolkit to genuinely “soft” things — happiness, preferences, brand value — often via revealed-preference methods (what people actually choose) and stated-preference/contingent-valuation methods (surveys designed to elicit trade-offs, like willingness-to-pay).
Chapter 12: “The Ultimate Measurement Instrument: Human Judges”
Covers how to use human judgment itself as a measurement instrument — with techniques to reduce the noise and bias inherent in expert judgment, including structured comparison methods (like the Lens Model / Rasch model in later editions) and simple rules for aggregating multiple experts’ estimates.
Part 4: Beyond the Basics
Chapter 13ish: “New Measurement Instruments for Management”
In later editions, this covers things like prediction markets, and more advanced applications — collaborative online estimation, using software tools Hubbard’s firm built (like his Monte Carlo add-ins) to operationalize the whole pipeline.
The book closes with a case for building an organizational habit: treat every “we can’t measure that” as a challenge rather than a stopping point, and always tie measurement effort to the value of the decision it informs, not to a hunger for data itself.
A few real running examples he uses
- Measuring the value of “IT security” investment by decomposing it into probability and cost of breaches.
- Estimating the value of a research library to a company (he/his firm actually did this consulting engagement) by measuring how often staff used it and what alternative cost they’d have incurred.
- Estimating “public value of a park” using revealed preference (nearby property values, travel-cost method).