Decision IntelDecision Intel
Pricing
Sign InBook a callTry the Demo
The Bias Genome · Quarterly

What the audit catches
before the deal sponsor can hide it.

The constraints that fire while the deal sponsor is still drafting the thesis — not after the IC memo lands. Drawn from 143 real strategic decisions: 127 failures and 16 successes across 11 industries. Methodology open. Data cite-able.

Baseline failure rate in this dataset: 89%. Every “failure lift” below is multiplied against this baseline.

Named biases
22
DI-B-001 → DI-B-022
Case studies
143
hand-curated
Industries
11
with meaningful coverage
Calibration baseline
Brier 0.258 ± 0.012
fair band · 95% CI
Data source
Seed
live on customer opt-in
Refreshed
2026-05-30
ISO date
n = 143 historical corporate decisions · mean Brier 0.258 ± 0.012 (95% CI, 10,000-iteration bootstrap, seed 17039507) · methodology v2.0.0-seed · computed 2026-05-30. Tetlock superforecasters land at ~0.13 · CIA analysts at ~0.23 · coin-flip at 0.25.
Most dangerous
GroupthinkDI-B-004
1.1x
Failure lift vs baseline · n=56
Most prevalent
Overconfidence BiasDI-B-007
59%
Appears in 84 of 143 cases
Most costly when uncaught
Cognitive MiseringDI-B-016
88
Avg impact score across failures · n=33
Most underestimated
Recency BiasDI-B-015
1.1x
Low prevalence (10%) but outsized failure lift

How the genome is built.

Four steps from a documented real-world decision to a ranked bias, with an outcome loop that recalibrates the weights as consenting customer orgs share results.

1
143 strategic decisions
SECNTSBFDAPost-mortem

Real, documented calls drawn from SEC filings, NTSB reports, FDA actions, and academic post-mortems. Every case cites its primary source.

2
22 biases per case, mapped
DI-B-003DI-B-007DI-B-012

The Decision Intel taxonomy (DI-B-001 → DI-B-022) assigns biases to each case with an excerpt and an academic citation.

3
22×22 interaction matrix
toxicwarm

Every bias pair is scored against the others. Context amplifiers (stakes, dissent, time pressure) multiply the score; false-positive damping kicks in when a pattern fires but the outcome succeeds.

4
Targeted mitigation per pattern
Echo Chamber
Synergy Mirage
Yes Committee

Each toxic pattern gets a named playbook (a devil’s advocate, a pre-mortem, a forced counterfactual). When real outcomes come back, the weights recalibrate on your own data.

Outcome loop. When a decision lands, its result flows back into step 3 — the interaction weights recalibrate on real data, not just the seed set.

The risk landscape.

Every bias plotted on two axes: how often it appears, how much it lifts the failure rate. The top-right quadrant is where your audit energy pays off.

The risk landscape
Prevalence × Failure lift · 23 biases
Bubble size reflects sample size (trust the dot as it gets bigger). Quadrants are directional — treat small-n points as signal, not statistic.
Common & dangerous
Rare but deadly
Common, containable
Low concern
0%15%29%44%59%Prevalence — share of cases containing the bias0.0x0.5x1.0x1.5x2.0xFailure lift vs baselinebaseline 1.0xGroupthinkAuthority BiasConfirmation BiasOverconfidence BiasPlanning FallacyCommon & dangerousprioritize theseRare but deadlywatch for n growthCommon, containablesurfaced often, usually caughtLow concernrare & non-dangerous
Max lift in view: 2.2x · Max prevalence: 59%Hover any bubble to see n and exact lift.

The leaderboard.

Sorted by failure lift: how much more often a decision fails when this bias is present, relative to the baseline. Filter by industry to narrow the slice.

Scored across all 143 seed cases. n = sample size; ⚠ marks biases with n<3 (directional only).
#BiasFailure liftPrevalencenInsight
01
GroupthinkDI-B-004
Peaks in financial services
1.1x
39%56modest failure lift vs baseline · n=56 · often paired in Echo Chamber
02
Recency BiasDI-B-015
Peaks in financial services
1.1x
10%14modest failure lift vs baseline · n=14
03
Cognitive MiseringDI-B-016
Peaks in financial services
1.1x
23%33modest failure lift vs baseline · n=33
04
Gamblers FallacyDI-B-018
Peaks in financial services
1.1x
2%3modest failure lift vs baseline · n=3
05
Halo EffectDI-B-017
Peaks in financial services
1.1x
8%12modest failure lift vs baseline · n=12 · often paired in Conglomerate Fallacy
06
Selective PerceptionDI-B-014
Peaks in financial services
1.1x
11%16modest failure lift vs baseline · n=16
07
Bandwagon EffectDI-B-006
Peaks in technology
1.1x
13%18modest failure lift vs baseline · n=18
08
Zeigarnik EffectDI-B-019
Peaks in energy
1.1x
5%7modest failure lift vs baseline · n=7
09
Availability HeuristicDI-B-003
Peaks in aerospace
1.1x
9%13modest failure lift vs baseline · n=13
10
Optimism Bias
Peaks in government
1.1x
17%24modest failure lift vs baseline · n=24
11
Paradox Of ChoiceDI-B-020
Peaks in government
1.1x
2%3modest failure lift vs baseline · n=3
12
Hindsight BiasDI-B-008
Peaks in financial services
1.1x
10%15modest failure lift vs baseline · n=15
13
Survivorship Bias
Peaks in retail
1.1x
3%5modest failure lift vs baseline · n=5
14
Narrative Fallacy
Peaks in retail
1.1x
2%3modest failure lift vs baseline · n=3
15
Authority BiasDI-B-005
Peaks in financial services
1.1x
39%56modest failure lift vs baseline · n=56 · often paired in Yes Committee
16
Framing EffectDI-B-013
Peaks in technology
1.1x
17%24modest failure lift vs baseline · n=24
17
Confirmation BiasDI-B-001
Peaks in technology
1.1x
47%67modest failure lift vs baseline · n=67 · often paired in Echo Chamber
18
Overconfidence BiasDI-B-007
Peaks in technology
1.0x
59%84no clear failure signal · n=84 · often paired in Optimism Trap
19
Planning FallacyDI-B-009
Peaks in technology
1.0x
32%46no clear failure signal · n=46 · often paired in Blind Sprint
20
Status Quo BiasDI-B-012
Peaks in technology
0.9x
34%49no clear failure signal · n=49 · often paired in Status Quo Lock
21
Sunk Cost FallacyDI-B-011
Peaks in technology
0.9x
25%36no clear failure signal · n=36 · often paired in Sunk Ship
22
Anchoring BiasDI-B-002
Peaks in technology
0.9x
52%75no clear failure signal · n=75 · often paired in Optimism Trap
23
Loss AversionDI-B-010
Peaks in technology
0.9x
27%38no clear failure signal · n=38 · often paired in Status Quo Lock

Toxic combinations.

Named patterns where two biases compound. Detection in live memos is 8x worse than either bias alone. Includes the three M&A workflow-native patterns — Synergy Mirage, Conglomerate Fallacy, and Winner’s Curse — that account for the 70-90% of acquisitions that miss projected synergies (McKinsey + KPMG). This is the product category our toxic-combination engine was built for.

Toxic network
How the biases combine
Inner ring: biases that participate in multiple toxic patterns. Edge color = pattern. Hover a pattern below to isolate its edges.
4Overconfidence Bias2Confirmation Bias2Groupthink2Anchoring BiasLoss AversionPlanning FallacySunk Cost FallacyStatus Quo BiasAuthority BiasIllusion of ValidityHalo EffectDecisionsat compound risk
11 biases · 10 named patterns · 10 toxic edgesHub numbers = pattern participation count.
Toxic combination
Echo Chamber
n=47

Confirmation bias amplified by unchallenged consensus. Teams hear what they already believe.

Confirmation BiasGroupthink
Appears in
The Coca-Cola Company· 1985U.S. Navy· 1988NASA / Perkin-Elmer· 1990
Toxic combination
Optimism Trap
n=43

Favorable initial estimates become reference points; downside scenarios are discounted.

Anchoring BiasOverconfidence Bias
Appears in
Long-Term Capital Management· 1998Blockbuster· 2000AOL Time Warner· 2000
Toxic combination
Status Quo Lock
n=34

The fear of loss from any change outweighs the documented cost of inaction.

Status Quo BiasLoss Aversion
Appears in
Johnson & Johnson· 1970Eastman Kodak· 1975Xerox· 1979
Toxic combination
Yes Committee
n=30

Deference to authority suppresses dissent; decisions ratified rather than debated.

GroupthinkAuthority Bias
Appears in
Continental Illinois National Bank· 1984NASA· 1986Barings Bank· 1995
Toxic combination
Blind Sprint
n=29

Overconfidence meets systematic underestimation of time and complexity.

Overconfidence BiasPlanning Fallacy
Appears in
Continental Illinois National Bank· 1984The Coca-Cola Company· 1985U.S. Navy· 1988
Toxic combination
Sunk Ship
n=24

Past investment justifies continued commitment — the 'we're too deep to stop' pattern.

Sunk Cost FallacyConfirmation Bias
Appears in
Eastman Kodak· 1975Xerox· 1979US Department of Defense / Lockheed Martin· 2001
Toxic combination
Conglomerate Fallacy
n=9

Far-adjacency acquisition justified by target growth and brand halo, with no answer to "why us as parent" — Porter parenting-advantage absent.

Illusion of ValidityHalo Effect
Appears in
AOL Time Warner· 2000WorldCom· 2002Myspace (News Corp)· 2005
Toxic combination
Synergy Mirage
n=5

Synergy claims without a named operational mechanism, accountable executive, or 90-day milestone — the canonical M&A failure mode (70-90% of acquisitions miss projected synergies).

Overconfidence BiasPlanning Fallacy
Appears in
AOL Time Warner· 2000WorldCom· 2002Hewlett-Packard· 2011
Toxic combination
Winner's Curse
n=5

Auction-dynamic anchoring drives bids above intrinsic value; "strategic necessity" and "competitive process" language flag the deal-fever pattern.

Anchoring BiasOverconfidence Bias
Appears in
Hewlett-Packard· 2011Yahoo· 2013General Electric· 2015
Toxic combination
Doubling Down
n=3

Escalating commitment to a losing course to avoid realizing the loss.

Sunk Cost FallacyLoss Aversion
Appears in
Barings Bank· 1995Long-Term Capital Management· 1998Amaranth Advisors· 2006
M&A workflow-native coverage · 14 anchor cases

The three M&A patterns we audit for, with their anchor cases.

McKinsey + KPMG track 70-90% of acquisitions missing their projected synergies. The patterns that drive most of those failures are nameable, repeatable, and detectable before the IC vote. Each column below shows the named pattern, its mechanism, and every deal in our 143-case library tagged with it.

70-90%
of M&A misses synergy projections
McKinsey + KPMG
9 types
of M&A artefacts recognised
CIM · IC memo · QofE · synergy model · integration plan · term sheet · model · DD · counsel review
5 toxic combos
fire on M&A workflows
3 M&A-specific (Synergy Mirage / Conglomerate Fallacy / Winner’s Curse) + Sunk Ship + Yes Committee
EU AI Act
Art. 14 record-keeping
Aug 2026 enforcement — DPR maps onto Art. 14 by design

Synergy Mirage

5 cases
Overconfidence × Planning Fallacy

Synergies without mechanism, owner, or 90-day milestone

Fires when projected synergies lack a NAMED OPERATIONAL MECHANISM, NAMED ACCOUNTABLE EXECUTIVE, and MEASURABLE 90-DAY MILESTONE. Per BCG integration best-practices, 70-90% of acquisitions miss projected synergies for exactly this gap.

Tagged cases · sorted recent first
  • General Electric2015
  • Microsoft2013
  • Hewlett-Packard2011
  • WorldCom2002
  • AOL Time Warner2000

Conglomerate Fallacy

9 cases
Illusion of Validity × Halo Effect

Far-adjacency without parenting thesis

Fires when an acquisition is justified primarily by target growth and brand halo, with no answer to Porter’s "why us as the best parent" question. Anchor failures: AOL-Time Warner, Daimler-Chrysler, Bed Bath & Beyond + Container Store.

Tagged cases · sorted recent first
  • Bed Bath & Beyond2023
  • Sears Holdings (Sears/Kmart)2018
  • Carillion plc2018
  • Steinhoff International2017
  • Microsoft2013
  • General Electric2008
  • Myspace (News Corp)2005
  • WorldCom2002
  • AOL Time Warner2000

Winner's Curse

5 cases
Anchoring × Overconfidence

Auction-dynamic anchoring above intrinsic value

Fires on competitive-process language ("strategic necessity", "preempting competitor B", "we cannot let X get this asset") combined with monetaryStakes=high + timePressure=true. The deal-fever signal that pushed WeWork’s S-1 valuation and the post-2010 SPAC wave above defensible levels.

Tagged cases · sorted recent first
  • Quibi2020
  • WeWork2019
  • General Electric2015
  • Yahoo2013
  • Hewlett-Packard2011

Coverage grows as the 143-case library expands. New patterns and anchors land in lockstep across the audit pipeline, the DPR, and this page. When a pattern’s case count shows fewer entries than you expect, that’s a tagging gap, not a detection gap — the engine catches the pattern; the library is still being annotated.

Biological-state amplifiers · unique to Decision Intel

Memos written under stress or after a winning streak fail differently.

Two language signals that materially change how bias load translates to outcome. Detected automatically from the memo text, applied as a multiplier on the toxic-pattern detection score. The behavioral-finance literature on these is solid; we’re the only decision-intelligence platform we know of that operationalises them.

↑ 1.20× multiplier
Winner Effect

Success-streak language (“our last three deals closed,” “we haven’t missed in two years,” “the team is in flow”) amplifies overconfidence and disposition effects. The neuroendocrine literature links winning streaks to elevated testosterone and risk preference. We surface the signal so the audit committee sees what the memo’s authors were riding when they wrote it.

↑ 1.18× multiplier
Stress / Cortisol

Crisis language (“urgent,” “regulator deadline,” “we have to move now,” “the window is closing”) amplifies System 1 decisions and shortens the deliberation horizon. Cortisol-driven cognition collapses the option set the memo writer considered before recommending. The multiplier surfaces that compression so the reader knows the recommendation came from a narrowed search, not an exhaustive one.

Implementation: detectWinnerEffect + detectStressSignals in the compound-scoring engine. Multipliers compound with the bias-load and noise terms, capped at 3.0× total to avoid runaway scores.

Method · How this dataset is built
  • Each case is a real, documented strategic decision drawn from SEC filings, NTSB reports, FDA actions, post-mortems, or academic case studies.
  • Biases are assigned per-case by applying the Decision Intel taxonomy (DI-B-001 → DI-B-022). Every named bias links to peer-reviewed academic sources at /taxonomy.
  • Failure lift = failure rate among cases with this bias ÷ baseline failure rate across the full dataset (89%).
  • Sample-size gate:headline rankings require n ≥ 5. Rows with n < 3 are shown dimmed with a ⚠ (directional only).
  • Honest selection bias: famous strategic failures dominate the public record. Industries with small coverage (aerospace, entertainment) should be read as signal, not statistic.
  • As consenting customer orgs opt into anonymized outcome sharing (see Settings → Privacy when logged in), their data supplements this seed. The live genome endpoint at /api/intelligence/bias-genome will take over once n ≥ 3 consenting orgs have reported outcomes.
This same taxonomy is what regulators have written into the EU AI Act high-risk decision-support obligations effective Aug 2026.See the mapping
Want to see this on your own memo?

Run your next strategic memo through the same taxonomy.

Upload takes 60 seconds. Your data stays yours. Anonymized aggregation is opt-in.

Audit your memo See the full taxonomy
Seed snapshot · 2026-05-30·Live updates begin when ≥ 3 customer orgs have opted into anonymized outcome sharing.