THE MACHINERY OF BASE RATES
A Complete Guide to What Is Actually Happening
Why Most Decisions Are Wrong Before They Begin
What follows is not advice.
It is not a statistics lesson. Not a forecasting model. Not a framework for better predictions. Not an argument for being more “data-driven.”
It is mechanism.
The actual machinery that determines why most business decisions are made against a background that does not exist. The structural property of human cognition that causes operators to build strategy on stories rather than rates, on salient examples rather than underlying frequencies, and on the most recent data point rather than the distribution that generated it.
Most operators encounter this as a surprise. The hire who seemed perfect fails. The market that seemed hot collapses. The investment that seemed conservative loses. They blame the outcome. The machinery was running before they decided.
This document is a description of that machinery.
What the operator reading it does next is their business.
PART ONE: THE RATE BENEATH THE STORY
What a Base Rate Is
In 1973, Daniel Kahneman and Amos Tversky ran an experiment that would eventually win a Nobel Prize. They gave subjects a personality description of a man named Tom W. Neat, orderly, detail-oriented, no warmth toward other people, interested in puzzles and mathematical order. Then they asked: is Tom W. more likely to be a computer science student or a humanities student?
Most subjects said computer science. The description fit the stereotype.
But the experimenters had given them another piece of information that most subjects ignored. At the university in question, humanities students outnumbered computer science students roughly 5 to 1.
The base rate. The underlying frequency in the population from which Tom was drawn. Before you know anything about Tom, the probability that any random student is in humanities is 80%. The description has to be very diagnostic to overcome that 5:1 ratio.
Most subjects did not compute this. They matched the description to their prototype and answered. The base rate was invisible.
This is not a mistake confined to psychology experiments. It is the default mode of human cognition.
BASE RATE vs. NARRATIVE
┌──────────────────────────────────────────────────────┐
│ │
│ WHAT THE BRAIN DOES: │
│ │
│ Story arrives → matches to prototype → confidence │
│ │
│ "This candidate is great. She reminds me of our │
│ best performer." (prototype matching) │
│ │
│ WHAT THE BRAIN SKIPS: │
│ │
│ What percentage of candidates with this profile │
│ actually succeed? │
│ │
│ If the base rate of success for all hires is 30%, │
│ and this candidate's profile raises it to 45%, │
│ you are still more likely to fail than succeed. │
│ │
│ But the story said "great." So the brain says 90%. │
│ │
└──────────────────────────────────────────────────────┘
A base rate is the frequency with which an event occurs in the relevant population before any specific information is considered.
The base rate of startup survival past year five. The base rate of a new product succeeding. The base rate of a new hire performing above expectations. The base rate of a restaurant in a ghost kitchen maintaining 4.8 stars for six consecutive months.
These numbers exist. They are knowable. They are almost never looked up. And without them, every judgment floats unanchored in narrative space.
PART TWO: THE NEGLECT
Why the Brain Prefers Stories to Rates
Kahneman identified two cognitive systems. System 1 operates automatically, matches patterns, generates intuitions. System 2 is slow, deliberate, computational. Base rate reasoning requires System 2. Pattern matching requires System 1.
The problem is not that humans cannot do base rate reasoning. The problem is that System 1 answers first, and it answers with such confidence that System 2 never activates.
The mechanism is called substitution. When faced with a hard question (what is the probability of X given the base rate and the evidence?), the brain substitutes an easier question (does this look like X?) and answers that instead. The subject does not know the substitution happened.
THE SUBSTITUTION MECHANISM
HARD QUESTION (System 2):
"Given that 70% of restaurants fail in year one,
and this restaurant has factors A, B, C,
what is the posterior probability of survival?"
│
│ (too hard, too slow)
│
▼
EASY QUESTION (System 1):
"Does this restaurant FEEL like a winner?"
│
│ (instant, confident)
│
▼
ANSWER: "Yes, it feels right."
CONFIDENCE: High.
ACCURACY: Uncalibrated.
This is not laziness. It is architecture. The brain was designed for an environment where base rate data did not exist in statistical form. In the ancestral environment, “how does this situation match my past experience” was the only computation available. Explicit base rate reasoning requires external data, mathematical processing, and the willingness to override a confident intuition with a number. The brain does not do this naturally. It must be trained to do it. And even after training, the intuitive answer still arrives first.
The Four Forms of Base Rate Neglect
Base rate neglect is not a single error. It takes four distinct forms in business, each with its own mechanism and its own cost.
FORM 1: PURE NEGLECT
─────────────────────────────────────────────────
The base rate is available but never consulted.
"We're launching a new product."
Q: What percentage of new products succeed?
A: (never asked)
The operator proceeds on narrative alone.
The base rate would have changed the plan,
the investment size, or the contingency.
FORM 2: DENOMINATOR BLINDNESS
─────────────────────────────────────────────────
The base rate is consulted but the wrong
population is used.
"Our success rate on launches is 60%."
Q: Of all companies who try this, what is the rate?
A: 12%.
Q: Why do you think you are 5x better?
A: "We're different." (the universal answer)
The operator used a self-selected denominator
(their own history, which is survivorship-biased)
instead of the population denominator.
FORM 3: ANCHORING OVERRIDE
─────────────────────────────────────────────────
The base rate is known but a salient anchor
overrides it.
"I know 70% of restaurants fail, but this
location did $2M last year." (anchor)
The anchor (one data point, recent, vivid)
overwrites the rate (70%, abstract, boring).
The brain gives the anchor 90% weight and
the rate 10% weight. The correct weighting
is closer to 50/50.
FORM 4: NARRATIVE DISPLACEMENT
─────────────────────────────────────────────────
The base rate is acknowledged but a story
explains why it does not apply.
"I know most hires don't work out, but
this candidate has a unique background."
Every exception narrative feels compelling.
That is the mechanism. The narrative displaces
the rate by providing a causal story for why
THIS case is different. The problem: every
failed case also had a causal story for why
it would succeed.
PART THREE: THE MATH OF UPDATING
How Base Rates and Evidence Combine
Thomas Bayes, an 18th-century Presbyterian minister, published a theorem that describes how rational agents should update beliefs when new evidence arrives. The theorem is simple. Its implications are not.
BAYES' THEOREM (plain language):
Your updated belief =
Your prior belief (base rate)
× How likely the evidence is if you're right
÷ How likely the evidence is overall
EXAMPLE: HIRING
Base rate: 30% of hires succeed (prior)
Evidence: Candidate passed a skills test
How likely is passing the test if they'll succeed? 80%
How likely is passing the test overall? 50%
Updated belief:
P(success | passed test) = 0.30 × (0.80 / 0.50) = 0.48
The candidate who "passed the test" still has less
than 50% chance of succeeding. The base rate dominates
because the test is not very diagnostic.
┌──────────────────────────────────────────────────────┐
│ │
│ START: 30% (base rate) │
│ │
│ + Passed skills test → 48% │
│ + Strong reference → 62% │
│ + Cultural fit interview → 58% (noisy signal, │
│ pulls BACK toward base)│
│ + 90-day trial results → 85% (strongest signal) │
│ │
│ EACH EVIDENCE UPDATES FROM THE CURRENT POSITION. │
│ Not from zero. Not from the narrative. From the │
│ prior probability, which started at the base rate. │
│ │
└──────────────────────────────────────────────────────┘
The critical insight: weak evidence barely moves the needle. Strong evidence moves it substantially. But nothing makes the base rate irrelevant. A 30% base rate requires very strong, very diagnostic evidence to reach 80%+. Most business evidence (interviews, market research, case studies) is moderate at best.
This is why operators who skip the base rate and go straight to “the evidence says yes” are making decisions as if the prior is 50/50. They are giving themselves a 20-percentage-point gift that does not exist.
The Diagnostic Ratio
Not all evidence is created equal. The diagnostic ratio (also called the likelihood ratio) measures how much a piece of evidence should update your belief.
DIAGNOSTIC RATIO =
P(evidence | hypothesis true)
÷ P(evidence | hypothesis false)
RATIO = 1.0 Evidence is useless. Equally likely
whether the hypothesis is true or false.
RATIO = 2.0 Evidence is moderate. Doubles your odds.
RATIO = 5.0 Evidence is strong. 5x your odds.
RATIO = 10.0+ Evidence is diagnostic. Substantially
shifts the probability.
BUSINESS EVIDENCE, TYPICAL DIAGNOSTIC RATIOS:
┌─────────────────────────────────┬──────────┬──────────┐
│ EVIDENCE │ RATIO │ QUALITY │
├─────────────────────────────────┼──────────┼──────────┤
│ "Gut feeling" about a hire │ 1.0-1.2 │ Noise │
│ Unstructured interview │ 1.2-1.5 │ Weak │
│ Market research survey │ 1.3-1.8 │ Weak │
│ Competitor case study │ 1.2-1.5 │ Weak │
│ Structured behavioral interview │ 2.0-3.0 │ Moderate │
│ Work sample / skills test │ 2.5-4.0 │ Moderate │
│ 90-day trial period │ 5.0-8.0 │ Strong │
│ Multi-year track record in role │ 8.0-15.0 │ Strong │
│ Financial audit of claims │ 10.0+ │ Diagnost │
│ Validated psychometric tests │ 2.0-3.5 │ Moderate │
│ Customer pre-orders (paid) │ 4.0-7.0 │ Strong │
│ Customer "interest" (surveys) │ 1.1-1.5 │ Noise │
│ Pilot with real revenue data │ 6.0-12.0 │ Strong │
│ "Everyone says it's a good idea"│ 0.8-1.0 │ Anti-sig │
└─────────────────────────────────┴──────────┴──────────┘
ANTI-SIGNAL: When the diagnostic ratio is below 1.0,
the evidence actually makes the hypothesis LESS likely.
Social consensus on untested ideas is anti-diagnostic
because groupthink and social desirability inflate
positive responses.
PART FOUR: NORMALIZATION
Making Numbers Comparable
Base rates are useless if the numbers they are based on are distorted. This is where normalization enters.
Normalization is the process of adjusting raw data to remove distortions so that the underlying rate becomes visible. You cannot find the base rate of a business’s performance if the numbers contain one-time items, seasonal effects, accounting choices, and structural changes that make each period incomparable to the next.
RAW DATA vs. NORMALIZED DATA
┌──────────────────────────────────────────────────────┐
│ │
│ RAW: "Revenue was $500K last quarter." │
│ │
│ DISTORTIONS: │
│ - One-time $80K catering order (won't repeat) │
│ - Seasonal Q4 holiday bump ($60K above run rate) │
│ - Price increase took effect mid-quarter │
│ - One location was closed 2 weeks for renovation │
│ │
│ NORMALIZED: $340K run rate │
│ │
│ The raw number says "growing." │
│ The normalized number says "flat." │
│ Every decision based on the raw number is wrong. │
│ │
└──────────────────────────────────────────────────────┘
There are four types of normalization in business. Each removes a different category of distortion.
Income Normalization
Adjusting earnings for items that do not represent ongoing operations.
ITEMS TO NORMALIZE OUT:
ADD BACK:
+ Owner salary above market rate
+ One-time legal settlement
+ Startup costs (first-year only)
+ Discretionary expenses (owner's car, travel)
+ Non-cash charges (depreciation above economic)
REMOVE:
- One-time revenue windfalls
- Insurance settlements
- Gain on sale of assets
- Revenue from discontinued products
PURPOSE: What would this business earn under
a neutral operator with no unusual events?
That is the base rate of its earning power.
Revenue Normalization
Adjusting top-line for seasonality, one-time contracts, and structural changes.
ADJUSTMENTS:
SEASONAL: Ghost kitchen revenue peaks in winter
(more delivery orders) and dips in summer (more
dine-out). A quarter-over-quarter comparison
without seasonal adjustment is meaningless.
METHOD: Compare to same quarter last year,
or compute a 12-month rolling average.
ONE-TIME: A large corporate catering order,
a promotional partnership, a viral social media
moment. Strip these to see the organic rate.
STRUCTURAL: A new delivery platform added.
A menu overhaul. A price increase. These are
permanent changes to the base rate itself.
Do not strip them. Recalibrate the base rate
with a new starting point.
┌──────────────────────────────────────────────────────┐
│ THE CRITICAL DISTINCTION: │
│ │
│ NOISE: Variance around a stable rate. │
│ Strip it. The rate is the truth. │
│ │
│ STRUCTURAL BREAK: The rate itself changed. │
│ Do not strip it. The new rate is the truth. │
│ Recalibrate. │
│ │
│ Confusing noise for a structural break causes │
│ you to chase every uptick (false signal). │
│ │
│ Confusing a structural break for noise causes │
│ you to miss a real change in the business │
│ (missed signal). │
│ │
└──────────────────────────────────────────────────────┘
Expense Normalization
Separating ongoing operating costs from investment and one-time items.
ONGOING (the base rate of cost):
- Rent, utilities, insurance
- Staff at normal staffing levels
- Food cost at standard recipes and portions
- Platform fees at current rates
- Supplies at normal usage
INVESTMENT (strip for run-rate analysis):
- Equipment purchases
- Build-out costs
- Training programs (one-time ramp)
- Technology implementation
- Menu development R&D
ONE-TIME (strip completely):
- Equipment repair after failure
- Legal fees from a specific incident
- Emergency staffing costs
- Renovation downtime
PURPOSE: What does this business cost to run
on a normal month? That is the cost base rate.
Every month that deviates, ask: noise or structural?
Balance Sheet Normalization
Adjusting assets and liabilities to reflect actual economic value.
Equipment at book value vs. replacement cost.
Inventory at cost vs. realizable value.
Receivables at face value vs. collectible value.
Lease obligations at contract terms vs. market.
PURPOSE: What is this business actually worth
right now, stripped of accounting conventions?
The base rate of value, not the reported number.
PART FIVE: REGRESSION TO THE MEAN
Why Extreme Performance Normalizes
Francis Galton discovered this in the 1880s by measuring parents and children. Tall parents tend to have children who are shorter than them. Short parents tend to have children who are taller than them. Not because of some force pulling toward the average. Because extreme observations contain an element of luck, and luck does not persist.
This is regression to the mean. It is not a force. It is a mathematical inevitability. And it is the most reliably ignored base rate phenomenon in business.
THE MECHANISM:
Observed performance = True ability + Luck
When performance is extremely good:
True ability was probably above average
+ Luck was probably above average
Next period:
True ability: still above average (persistent)
Luck: resets to average (not persistent)
Result: Next period looks worse.
Not because anything changed. Because luck
returned to baseline.
┌──────────────────────────────────────────────────────┐
│ │
│ EXAMPLE: LOCATION PERFORMANCE │
│ │
│ South Loop had a record month: $85K revenue. │
│ Operator concludes: "We figured it out." │
│ │
│ Decomposition: │
│ True run rate: $68K (stable trend) │
│ Seasonal effect: +$8K (winter delivery surge) │
│ Luck: +$9K (viral TikTok, one-time) │
│ │
│ Next month prediction (base rate): $68K │
│ Operator prediction (narrative): $85K+ │
│ │
│ Next month actual: $71K │
│ Operator reaction: "What went wrong?" │
│ Correct reaction: "Nothing. Regression." │
│ │
└──────────────────────────────────────────────────────┘
The business implications are severe:
REGRESSION TRAPS IN BUSINESS:
1. REWARDING LUCK
Employee of the month after a peak period.
The peak was partially luck. The employee
"regresses" next month. Manager concludes
they got complacent. They did not. They got
average after being lucky.
2. PUNISHING BAD LUCK
Worst performer gets put on a PIP after a
bad month. The trough was partially bad luck.
They "improve" next month (regression). Manager
concludes the PIP worked. It did not.
3. CHASING HOT STREAKS
Expanding the menu item that sold best last month.
Increasing inventory on the product that moved
fastest. Hiring more of the profile that just
succeeded. If the hot streak was partially luck,
the expansion is built on a base rate that does
not exist.
4. ABANDONING COLD STREAKS
Killing the product that underperformed last month.
Firing the person who had a bad quarter. Exiting
the market that dipped. If the cold streak was
partially bad luck, you are cutting at the trough.
THE FIX: Never make structural decisions based
on a single period's data. Normalize for luck
by requiring 3+ periods before concluding a trend
is real. The base rate emerges over time, not in
a single observation.
PART SIX: SURVIVORSHIP BIAS
The Base Rate You Cannot See
Abraham Wald was a mathematician working for the Statistical Research Group during World War II. The military wanted to know where to add armor to bombers. They examined planes returning from missions and found bullet holes concentrated on the fuselage and wings. The initial recommendation: armor the fuselage and wings.
Wald said the opposite. Armor the engines.
The planes that came back had been hit in the fuselage and wings and survived. The planes hit in the engines did not come back. The data was collected from survivors, not from the full population. The bullet hole distribution was inverted from what the full base rate would show.
This is survivorship bias. The systematic error produced by studying only the cases that made it through a selection filter, while the cases that did not make it are invisible.
SURVIVORSHIP BIAS IN BUSINESS:
┌──────────────────────────────────────────────────────┐
│ │
│ WHAT YOU SEE: │
│ Successful companies that did X │
│ │
│ WHAT YOU DON'T SEE: │
│ Failed companies that also did X │
│ │
│ CONCLUSION FROM SURVIVORS: │
│ "X causes success." │
│ │
│ TRUE BASE RATE: │
│ X has no causal relationship to success. │
│ Both successful and failed companies did X. │
│ You only studied the ones that survived. │
│ │
└──────────────────────────────────────────────────────┘
EXAMPLES:
"Successful entrepreneurs dropped out of college."
Base rate: Most dropouts do not become successful
entrepreneurs. You are seeing the survivors.
"Great companies have strong cultures."
Base rate: Many failed companies also had strong
cultures. You are studying the survivors.
"This ghost kitchen model works. Look at the ones
that scaled."
Base rate: How many ghost kitchens opened with
the same model and closed? You only see the open ones.
"Top performers work 80-hour weeks."
Base rate: Many people work 80-hour weeks and fail.
Burnout is the base rate. Success is the exception
you study.
PART SEVEN: STRUCTURAL BREAKS
When the Base Rate Itself Changes
The most dangerous moment in base rate analysis is when the underlying rate changes and you do not notice. This is a structural break. The old base rate is no longer the truth. A new base rate has replaced it. But because the old rate is familiar and the new rate is unfamiliar, the brain defaults to the old one.
STRUCTURAL BREAK DETECTION:
┌──────────────────────────────────────────────────────┐
│ │
│ Performance over time: │
│ │
│ $70K ─── $68K ─── $72K ─── $71K ─── $69K ── │
│ │
│ Base rate: ~$70K/month. Noise: ± $3K. │
│ │
│ Then: │
│ │
│ $70K ─── $68K ─── $72K ─── $58K ─── $55K ─── $57K │
│ │
│ NOISE OR STRUCTURAL BREAK? │
│ │
│ TEST: Does the deviation persist for 3+ periods? │
│ $58K, $55K, $57K = 3 periods below old range. │
│ This is a structural break. New base rate: ~$57K. │
│ │
│ COMMON ERROR: Treating the first $58K as noise. │
│ "Bad month." Waiting for regression to old mean. │
│ The old mean is dead. Waiting for it costs $13K/mo. │
│ │
└──────────────────────────────────────────────────────┘
CAUSES OF STRUCTURAL BREAKS IN BUSINESS:
EXTERNAL:
- New competitor enters the market
- Platform algorithm change (delivery app ranking)
- Regulatory change (health code, labor law)
- Demographic shift in delivery zone
- Macroeconomic change (recession, inflation)
INTERNAL:
- Key employee departure
- Menu change
- Price change
- Supplier change
- Process change (intentional or accidental)
- Quality drift (normalization of deviance)
DETECTION PROTOCOL:
1. Define the base rate range (mean ± 2 standard deviations)
2. Any single observation outside the range = flag
3. Two consecutive observations outside = investigate
4. Three consecutive observations outside = structural break
→ Recalibrate the base rate
→ Identify the cause
→ Decide: accept the new rate or intervene to change it
PART EIGHT: SIGNAL EXTRACTION
Separating What Matters from What Does Not
Signal extraction is the discipline of identifying which data points contain information about the base rate and which are noise. In business, the volume of data available has increased by orders of magnitude. The amount of signal has not increased proportionally. Most of the increase is noise.
SIGNAL vs. NOISE IN BUSINESS METRICS:
┌─────────────────────────────────┬──────────┬──────────┐
│ DATA POINT │ SIGNAL │ NOISE │
├─────────────────────────────────┼──────────┼──────────┤
│ Daily revenue │ Low │ High │
│ Weekly revenue │ Moderate │ Moderate │
│ Monthly revenue (normalized) │ High │ Low │
│ Rolling 3-month revenue │ Very high│ Very low │
├─────────────────────────────────┼──────────┼──────────┤
│ Single customer complaint │ Low │ High │
│ Complaint rate per 1000 orders │ High │ Low │
│ Complaint rate trend (3 months) │ Very high│ Very low │
├─────────────────────────────────┼──────────┼──────────┤
│ One employee's performance today│ Low │ High │
│ Employee's 90-day average │ High │ Low │
│ Team's 90-day average │ Very high│ Very low │
├─────────────────────────────────┼──────────┼──────────┤
│ One star rating │ Low │ High │
│ Average rating this week │ Moderate │ Moderate │
│ Rating trend over 4 weeks │ High │ Low │
└─────────────────────────────────┴──────────┴──────────┘
PRINCIPLE: Signal increases with aggregation and time.
Noise decreases with aggregation and time.
The operator who reacts to daily data is reacting
to noise 80% of the time.
The operator who reacts to monthly trends is reacting
to signal 80% of the time.
The Nate Silver Framework
In signal extraction, there is a useful distinction between three types of information:
TYPE 1: KNOWN SIGNAL
The underlying rate is stable and measurable.
Example: your food cost as a percentage of revenue,
measured monthly, normalized for menu changes.
This IS the base rate. Act on it.
TYPE 2: KNOWN NOISE
Variance that is random and meaningless.
Example: daily fluctuations in order volume.
Do not act on it. Wait for the base rate to
emerge over time.
TYPE 3: UNKNOWN SIGNAL
Something real is happening but you cannot yet
tell if it is noise or structural break.
Example: Error rate jumped 30% this week.
Is it a new hire learning? A process failure?
A measurement artifact?
PROTOCOL FOR TYPE 3:
1. Flag it. Do not act.
2. Wait one more period.
3. If it persists, investigate the cause.
4. If the cause is structural, recalibrate.
5. If the cause is temporary, wait for regression.
PART NINE: COMPARABILITY
Making Different Things Measurable on the Same Scale
Comparability is the condition under which two numbers can be meaningfully compared. Without it, base rate analysis is impossible because you are comparing distorted numbers to distorted numbers and calling the result a rate.
THE COMPARABILITY CHECKLIST:
Before comparing two numbers, verify:
1. SAME TIME PERIOD?
Q1 vs Q1 (seasonal adjustment built in).
Not Q1 vs Q4 (seasonal noise embedded).
2. SAME SCOPE?
Both locations? Same menu? Same hours?
If one location was closed for renovation,
the comparison is meaningless without adjustment.
3. SAME ACCOUNTING?
Same cost allocation method? Same depreciation?
Same treatment of owner compensation?
Different accounting = different numbers,
same reality.
4. SAME EXTERNAL CONDITIONS?
Were delivery app algorithms the same?
Was weather comparable? Was there a local event
(convention, holiday, construction) affecting one
period but not the other?
5. STRIPPED OF ONE-TIME ITEMS?
Both periods normalized for non-recurring items?
A $20K equipment failure in one month makes that
month incomparable without adjustment.
Cross-Location Comparability
For a district manager running two ghost kitchens, cross-location comparison is essential but dangerous without normalization.
SOUTH LOOP vs. AVONDALE (raw):
┌──────────────┬───────────┬───────────┐
│ METRIC │ S. LOOP │ AVONDALE │
├──────────────┼───────────┼───────────┤
│ Revenue │ $72K │ $58K │
│ Error rate │ 0.8% │ 1.2% │
│ Star rating │ 4.85 │ 4.72 │
│ Labor % │ 16.8% │ 18.9% │
└──────────────┴───────────┴───────────┘
CONCLUSION FROM RAW: South Loop is better.
NORMALIZED (adjusting for conditions):
┌──────────────┬───────────┬───────────┬───────────┐
│ METRIC │ S. LOOP │ AVONDALE │ ADJ │
├──────────────┼───────────┼───────────┼───────────┤
│ Revenue │ $72K │ $58K │ Avondale │
│ (per order) │ $28.40 │ $29.10 │ higher AOV│
├──────────────┼───────────┼───────────┼───────────┤
│ Error rate │ 0.8% │ 1.2% │ Avondale │
│ (per new │ 0.8% │ 0.7% │ had 2 new │
│ hire adj) │ │ │ hires │
├──────────────┼───────────┼───────────┼───────────┤
│ Star rating │ 4.85 │ 4.72 │ S. Loop │
│ (adj for │ 4.82 │ 4.75 │ had promo │
│ promo bias) │ │ │ driving │
│ │ │ │ favorable │
│ │ │ │ selection │
├──────────────┼───────────┼───────────┼───────────┤
│ Labor % │ 16.8% │ 18.9% │ Avondale │
│ (adj for │ 17.2% │ 17.8% │ had extra │
│ training) │ │ │ training │
│ │ │ │ shift │
└──────────────┴───────────┴───────────┴───────────┘
CONCLUSION FROM NORMALIZED: Gap is much smaller.
Avondale may actually be performing better on
some metrics when conditions are equalized.
The raw comparison would have led to: reward S. Loop,
pressure Avondale. The normalized comparison leads to:
both are near target, Avondale's gap is structural
(new hires), which will self-correct.
PART TEN: THE OPERATOR’S PROTOCOL
How to Actually Use Base Rates
Theory is mechanism. Protocol is execution. Here is how an operator applies base rate thinking to decisions.
THE BASE RATE PROTOCOL (5 steps):
STEP 1: FIND THE RATE
─────────────────────────────────────────────────
Before making any decision, ask:
"What is the base rate of success for this type
of decision, in this type of business, at this scale?"
Sources: Industry data. Your own historical data
(normalized). Competitor data (if available).
Academic research. Trade publications.
If you cannot find the rate, estimate conservatively.
Unknown base rates are almost always worse than
operators assume.
STEP 2: NORMALIZE YOUR DATA
─────────────────────────────────────────────────
Strip one-time items. Adjust for seasonality.
Separate noise from structural breaks.
Make the numbers comparable before computing
any rate from them.
The base rate of a distorted dataset is a
distorted rate. Garbage in, garbage out.
STEP 3: ASSESS YOUR EVIDENCE
─────────────────────────────────────────────────
For each piece of evidence that makes you believe
this case is different from the base rate, ask:
"What is the diagnostic ratio of this evidence?"
Gut feeling: ~1.0 (useless)
Interview impression: ~1.3 (weak)
Pilot with real data: ~8.0 (strong)
Only strong evidence (ratio > 3.0) meaningfully
updates the base rate. Everything else is noise
dressed as insight.
STEP 4: UPDATE, DO NOT REPLACE
─────────────────────────────────────────────────
Use the base rate as your starting point.
Update it with evidence. Do not replace it.
"The base rate says 30%. My evidence moves it
to 48%. I will plan for 48% but prepare for 30%."
NOT:
"My evidence says 90%. The base rate is irrelevant."
STEP 5: WATCH FOR REGRESSION AND BREAKS
─────────────────────────────────────────────────
After the decision:
- If results exceed expectations for one period,
check for luck before celebrating.
- If results trail expectations for one period,
check for bad luck before panicking.
- If results deviate for three consecutive periods,
the base rate has changed. Recalibrate.
The base rate is alive. It moves. Your job is to
track it, not to set it once and forget.
PART ELEVEN: THE MACHINERY IN MOTION
Where Base Rates Run the Business
DECISION DOMAIN BASE RATE QUESTION WHAT MOST OPERATORS DO
HIRING What % of hires at this Hire based on interview
level succeed past 90 days? impression (ratio: 1.3).
(probably 25-40%) Plan as if 90%.
NEW PRODUCT What % of new menu items Launch based on taste test
LAUNCH succeed past 3 months? and social consensus
(probably 20-35%) (ratio: 0.8-1.0).
Plan as if 70%.
LOCATION What % of ghost kitchens Open based on one successful
EXPANSION at this scale hit year-2 location and a narrative.
profitability? (probably Plan as if 80%.
30-45%)
VENDOR What % of vendor switches Switch based on better
CHANGE improve cost without quoted price. Plan as if
quality degradation? 100% savings realized.
(probably 40-60%) (Hidden costs emerge later.)
PROCESS What % of process changes Implement based on logic.
CHANGE achieve intended outcome Plan as if 90%.
within 30 days? Actual: 30-50%.
(probably 30-50%)
PROMOTION What % of internal Promote based on individual
DECISION promotions succeed at the performance (which includes
next level? (probably luck). Plan as if 85%.
40-60%)
MARKETING What % of marketing Spend based on projected
SPEND campaigns achieve positive ROI from the agency.
ROI within 90 days? Plan as if 70%.
(probably 15-30%) Actual: 15-30%.
┌──────────────────────────────────────────────────────┐
│ │
│ PATTERN: Operators systematically overestimate │
│ success probability by 30-50 percentage points. │
│ │
│ The base rate says 30%. They plan for 70%. │
│ The gap between reality and plan is where │
│ businesses die. │
│ │
│ Not from bad ideas. From miscalibrated plans │
│ built on ignored base rates. │
│ │
└──────────────────────────────────────────────────────┘
PART TWELVE: THE SINGLE INSIGHT
The base rate is not a ceiling. It is not a prediction. It is not an argument against ambition.
It is the starting point from which all honest analysis begins.
When an operator knows the base rate, they can choose to beat it. They can deploy evidence, effort, and structural advantages to move their probability above the population mean. But they start from truth, not from narrative.
When an operator ignores the base rate, they cannot beat it because they do not know where they stand relative to it. Their plans are built on a confidence that was never earned. Their surprises are predictable to anyone who knew the rate.
The machinery is simple: the base rate exists whether you look it up or not. It will express itself in your outcomes whether you planned for it or not. The only question is whether you enter the decision with the rate as your foundation or with a story as your foundation.
One of these produces calibrated plans with appropriate contingencies. The other produces confident plans that fail at the base rate.
The base rate does not care which one you choose. It runs either way.