Investment Mosaic Analysis
Combine disparate clues into a probabilistic picture of a company’s future
- Difficulty
- Advanced
- Time to result
- ~weeks to results
- Steps
- 6
- Confidence
- 97%
Miller connects investment analysis with his military-intelligence training: gather disparate pieces of lawful information, determine how they relate, and assemble a probabilistic picture of what may be happening. A single observation may appear meaningless, but several independent clues can jointly support a useful inference about a company, industry, or economy. The process starts with a defined uncertainty, collects varied evidence, weighs source reliability, and constructs multiple scenarios rather than one deterministic prediction. Analysts then identify which additional observations would strengthen or weaken each scenario and update probabilities as evidence changes. Its output is not certainty but a better-supported range of outcomes that can feed valuation, risk assessment, and position sizing. The method is especially valuable when conventional financial statements describe only part of a changing business.
Origin
Bill Miller learned the method through military-intelligence training and later recognized that the same synthesis of disparate evidence applied directly to stock selection.
Core principles
- 01Individual clues can be weak while their combination is informative
- 02Build probabilistic pictures rather than claiming certainty
- 03Use legally available information from diverse sources
- 04Test several scenarios instead of committing to one forecast
- 05Search for relationships among apparently unrelated facts
How to run it
- 1
Frame the intelligence question
Specify what you need to infer about the company, industry, or economy. Keep the question narrow enough that evidence can alter the answer.
Pro tip Phrase the question around an uncertain future development.
Watch out A vague question encourages indiscriminate information gathering.
- 2
Gather disparate clues
Collect lawful public information from financial reports, competitors, customers, suppliers, industry data, and observable behavior. Seek independent evidence rather than repeated versions of one source.
Pro tip Include clues that initially appear unrelated to the central question.
Watch out Never use material nonpublic information.
- 3
Grade the evidence
Assess each clue for reliability, relevance, timing, and possible bias. Separate direct observations from interpretations.
Pro tip Give greater weight to evidence that would be costly or difficult for the source to fake.
Watch out Several low-quality clues do not automatically create a high-quality conclusion.
- 4
Build competing pictures
Combine the evidence into multiple plausible scenarios explaining what is happening and what could follow. Avoid forcing every clue into a preferred thesis.
Pro tip Include at least one scenario that contradicts your initial view.
Watch out Prematurely settling on one narrative creates confirmation bias.
- 5
Assign probabilities
Estimate the relative likelihood of each scenario and identify the observations that would change those estimates. Translate the scenarios into possible business and valuation effects.
Pro tip Use probability ranges rather than false precision.
Watch out A most-likely scenario can still be wrong.
- 6
Update continuously
Incorporate new evidence and revise the scenario probabilities. Record why the picture changed so that learning is preserved.
Pro tip Track disconfirming evidence explicitly.
Watch out Do not reinterpret every new fact to protect the original thesis.
In the wild
An analyst combines hiring patterns, supplier commentary, customer reviews, shipment data, and a competitor’s capacity expansion. None proves that demand is accelerating, but together they support a higher-probability growth scenario that can be tested against the current valuation.
→ The analyst forms an evidence-based forecast without relying on inside information.
Miller described being trained to combine unrelated information into a probabilistic picture of what might be occurring. He applied the same discipline to companies and the economy, where isolated facts often become meaningful only when connected.
→ The training became a reusable approach to security analysis.
Common mistakes
Treating inference as certainty
A mosaic supports probabilities, not guaranteed outcomes. Position size should reflect the remaining uncertainty.
Collecting only confirming clues
Evidence selected to support an existing thesis creates a persuasive story without improving its accuracy.
Double-counting one source
Several reports derived from the same original information are not independent confirmation.
Is it for you?
Best for
It is best for analysts evaluating complex businesses, industries, or economic developments under uncertainty.
Not ideal for
It is not ideal when decisions require certainty or when the available information is unlawfully obtained or fundamentally unreliable.
From the transcript
“you can have what they call a mosaic approach where you put pieces of information together and figure something out.”
“taking disparate bits of information and using them to create an essence of picture, probabilistic picture of what might be going on and what could…”
“There were unrelated on their own, but you could put them together and get get the beginning of a picture when each individual piece by…”
From the episode
Barry Ritholtz's Masters in Business: Bill Miller Interview