The Science Behind AI4EBITDA

An MRI for organizations.
Not a mirror.

Surveys hold a mirror up and ask people to describe themselves. AI4EBITDA reads what people already wrote, across 410 million leadership profiles, fifteen years of validated data, and a 200-year-old science of language.

410M+leadership profiles
analysed to date
15yrsof validated
performance data
p=0.019statistical significance
vs. financial outcomes
0surveys, interviews,
or self-reports required
Sean Languedoc, CEO & Co-founder LinkedIn ↗
Nicole Whittle, COO & Co-founder LinkedIn ↗
Built on Stealth Dog Labs research, validated against ACSI
How We Use Public Data, Ethically

No surveys. No interviews. No private data.

We use publicly available data to decode the real conditions shaping an organization. From those signals we generate operational specifications that let leadership teams define goals, benchmark progress, and execute their mandates with confidence. Nothing about the method requires access to private systems, individual consent at the subject level, or anyone's time.

What we use
Public language only
Profiles, websites, filings, transcripts, press releases, public job descriptions, published thought leadership. Material the subject already chose to make public.
What we never use
Private signals
No internal email. No Slack. No surveys. No interviews. No IT access. No private documents. Nothing the subject would expect to be confidential.
How we report
Patterns, not portraits
We surface patterns at the team or company level so leadership can act on them. Individual profiles are reserved for the executives named in the engagement scope.
The simple version

The same public footprint a journalist, analyst, or competitor could read. We just read it at scale, against a validated framework, and report what it says about the organization's capacity to deliver.

What We Actually Read

The average person uses about 45,000 words. We focus on 1,500.

You speak and write with a working vocabulary of roughly forty-five thousand words. Most of those are interchangeable. Synonyms, fillers, technical terms specific to your job. The pattern that reveals how you think sits inside a much smaller set: about fifteen hundred high-signal words you reach for again and again, and the strongest signal in that set is not the content words you would expect.

It is the boring little words. Pronouns, articles, prepositions, conjunctions. They are invisible to the speaker, invisible to anyone trying to perform a different self, and they predict personality, decision style, and risk posture better than any noun or verb. Content words tell you what someone is thinking. Function words tell you who they are.

~45,000
Working vocabulary
The full range of words an average professional uses across all writing and speech.
~1,500
High-signal words
The narrow set whose density reveals how you reason, decide, and operate.
pronouns articles prepositions conjunctions negations tense
100%
Subconscious
You do not pick these words deliberately. That is why the signal is honest, and why a survey cannot surface it.
Why density, not content

We are not reading what you said. We are reading how often you reach for I versus we, for the versus a, for but, because, although. Function-word density is involuntary. Content is performed.

A 200-Year-Old Science

Psycholinguistic profiling. Older than the personality test.

Reading psychology from language patterns is not a new field. It is older than Myers-Briggs, older than DISC, older than every survey-based assessment in use today. The peer-reviewed literature contains thousands of citations across linguistics, psychology, political science, and forensic analysis.

What is new is the data and the compute. Until recently, you could only score a handful of subjects: political leaders, criminal defendants, public figures whose writing was already collected. AI4EBITDA reads four hundred and ten million profiles continuously, from the open record alone.

200yrs
Established field, modern instrument. The science traces to early-19th-century philology and matured through 20th-century forensic linguistics, computational stylistics, and corpus psychology. AI4EBITDA did not invent the field. We built the instrument that makes it operational at organizational scale.
Pronouns
I/we ratio reveals ownership, status posture, and accountability orientation.
Articles
The vs. a separates concrete, categorical thinkers from abstract, exploratory ones.
Prepositions
Density of in, through, above tracks analytical complexity and decision depth.
Conjunctions
But, because, although expose causal reasoning, hedging, and integrative thought.
From Words to Financial Outcome

One pipeline. Six steps.

Every score we publish, for an individual, a team, or a company, is the product of the same six-step path. Words become signals. Signals become mindsets. Mindsets become work-styles. Work-styles roll up into a company score that predicts financial performance.

01 · Collect
Public
Public language
Profiles, filings, transcripts, published communications. No surveys. No private data.
02 · Measure
1,500
Word density
Score the high-signal words: verbs, adjectives, adverbs, pronouns.
03 · Compute
657
Algorithms
Six hundred and fifty-seven psycholinguistic variables, scored in parallel.
04 · Roll up
6
Mindsets
Variables collapse into six business mindsets, each scored on a continuous scale.
05 · Classify
4
Work styles
Mindsets place a person into one of four work styles, weighted by authority.
06 · Score
1–100
Company score
Authority-weighted roll-up. Predicts financial performance at p=0.019.
↓ Public data in Continuous scale, not binary Financial signal out ↑
After scoring: the development loop

For engaged companies, AI4EBITDA writes a per-leader development program tied directly to their lowest-scoring mindsets. We monitor whether the practices are followed, and we re-score continuously. Every point gained increases the probability of delivering the mandate.

The Six Mindsets · Step 04

Six mindsets. One composite score.

Each mindset is named for the thinker who first formalized the underlying body of work. The science is rigorous; the idea behind each one is simple. We score each mindset on a continuous scale, never binary, and roll the six into one composite for every person.

MindsetWhat it measuresWhy it matters
IRC. Iterative Reasoning Think in steps

The ability to reason in steps, hold uncertainty, and course-correct when the first answer does not hold up.

The single strongest predictor of productive AI use. High-IRC people treat AI output as a starting hypothesis, not a final answer.

LeCun-Skinner. Adaptive Intelligence See downstream

Systems-level thinking, downstream-consequence awareness, and the capacity to integrate new information into existing models.

Essential for directing AI on complex problems. Low scorers accept outputs at face value without probing assumptions.

Clayton. Strategy & Innovation Recognise real change

Whether someone recognises when a new method represents a genuine shift, not just a faster version of the old way.

Predicts willingness to build AI-native workflows rather than layering AI onto legacy processes.

Taguchi. Execution Discipline Learn from failure

Whether structured improvement and learning from failure are built into how work gets done.

High Taguchi scorers naturally run the verify-adjust loop that makes AI use safe and effective.

Deming. Quality Orientation Quality before speed

Orientation toward process quality over speed and output volume.

High Deming scorers ask "is this right?" before "is this fast?" A critical check on AI overreliance.

Ackoff. Systems Thinking See the whole

The ability to see the whole system and design for it, not just optimise individual parts.

AI produces narrow answers to narrow questions. Ackoff-oriented thinkers know to ask bigger ones.

A note on naming

The frameworks behind these mindsets, Deming on quality, Taguchi on disciplined improvement, Ackoff on systems, are well-established business theory. We use them as scientific anchors, not as the headline. What matters in practice is the simple version on the badge: think in steps, see downstream, learn from failure, quality before speed.

Secondary Roll-up · Step 05

Four work-styles. How the mindsets show up at work.

The six mindsets roll up, drawing on Torbert's developmental action-logic framework, into four work-styles. This is a useful shorthand for distribution, planning, and program design. It is not the whole story; the underlying mindset score is what AI4EBITDA tracks and develops.

Thought Leader
Independent · sets direction
Reasons in novel territory without a model to follow. High IRC, high Ackoff. Drives the work others learn from.
Fast Follower
Adaptive · scales the model
Moves quickly once a model exists. Strong on Clayton and Taguchi. Recognises real change, then operationalises it cleanly.
Guided Learner
Developing · needs structure
Capable, given clear examples and structured support. Mindset profile is forming; the development program is most leveraged here.
Foundation Builder
Stable · holds the floor
Builds the floor others stand on. Low risk, narrow autonomy. Movement up the stack starts with one repeatable habit.
Pattern Detection

Dysfunctions and expert opportunities. Read in the same data.

The same word-density signal that scores mindset also surfaces patterns that block, or amplify, execution. We do not diagnose individuals. We flag patterns at the population level so leadership can act on them before they show up in the financials.

Blockers · Dysfunction patterns

Blocker
Burnout signal
Pronoun shift, declining adjective range, rising hedging. Visible in language six to twelve months before attrition.
Blocker
Narcissism pattern
Pronoun inflation, attribution asymmetry, low-uncertainty verbs. Predicts capture risk in decision authority.
Blocker
Authoritarian posture
High imperative density, low question rate, narrow modal range. Suppresses Taguchi and Ackoff signal in the team below.
Blocker
Performative ownership
Active-voice claim language without follow-through verbs. Common in stalled transformations.
Blocker
Authority gap
High capacity, low decision-rights language. The team can do the work; the structure will not let them.
Blocker
Resistance density
Negation rate, hedging, framing-shift vocabulary. Tracks where transformation will encounter ground friction.

Amplifiers · Expert opportunity patterns

Amplifier
Latent expert
High mindset score under low-authority language. Promote, sponsor, or release into a Thought-Leader role.
Amplifier
Coachable Fast Follower
High Clayton, mid-IRC. Small, well-aimed program produces the largest tier movement.
Amplifier
Quality anchor
High Deming and Taguchi co-occurrence. Anchors safe AI deployment; disproportionately important on AI rollouts.
Human and AI · One Framework

Human and AI intelligence, scored on the same scale.

The cognitive framework that measures how a person reasons, decides, and learns also measures the AI systems your team is deploying. Traits. Mindset. Dysfunction. Computable for both. That makes one of the hardest questions in mid-market AI overwhelm answerable: are the leaders we have, and the AI systems we are buying, cognitively compatible?

01
Measure both, the same way
Quantify the cognitive makeup of leaders, teams, and AI systems on a unified set of variables. The first time psychological insight has been made comparable across biological and artificial intelligence.
02
Detect mismatch before rollout
Find the gaps in decision logic, trust posture, and interaction friction that quietly stall AI deployments. Read in language before they show up in adoption rates or written-off pilots.
03
Match cognition to task
Identify which workflows should be led by AI, which by people, and which are safe to automate end-to-end. Fewer pilots, fewer reversals, less budget burned proving the obvious.
Why this matters for AI4EBITDA

The cost of an AI rollout is not the licence. It is the dozens of leaders who have to direct, trust, and verify it. If their cognitive architecture and the system's are mismatched, the rollout stalls and the EBITDA hit lands twelve to eighteen months later. We score the match before you sign the contract.

Why It is Different

An MRI. Not a mirror.

Survey-based assessments (Myers-Briggs, DISC, Hogan) ask people to describe themselves. The output is filtered by self-image, by how the subject wants to be seen, and by what they know about the test. AI4EBITDA reads what people already produced, before they knew anyone was looking.

The mirror
Survey-based assessments
Myers-Briggs · DISC · Hogan · 360s
Source
Self-report. The subject answers questions about themselves.
Filter
Self-image, social desirability, fatigue, gameability.
Frequency
Annual or one-off. Stale by the time it is read.
Coverage
Whoever sat the assessment. Misses everyone else.
Validation
Construct validity against other surveys.
The MRI
AI4EBITDA
Public language · 657 variables · continuous
Source
Public language the subject already produced. Unfiltered, unprompted.
Filter
None. Word-density choice is subconscious. It cannot be performed.
Frequency
Continuous. Re-scored as new public language appears.
Coverage
Every person with a public footprint. Whole organisations, ecosystems.
Validation
Predictive validity against ACSI and 15 years of financial outcomes (p=0.019).
The shorthand

A mirror tells you what someone wants to see. An MRI tells you what is actually there. AI4EBITDA is the second one.

External Validation

The numbers, with their source tags.

Two distinct datasets sit behind the AI4EBITDA framework. Mixing them weakens the proof. Here they are kept separate, with the relationship between them shown last.

Our methodology dataset The population AI4EBITDA measures and benchmarks against
410M+
Leadership profiles
scored across the database
1.4M
Companies benchmarked
against the framework
657
Psycholinguistic variables
across six named dimensions
ACSI external validation cohort Independent third-party data. American Customer Satisfaction Index, not collected by AI4EBITDA
350
ACSI-tracked companies
in the validation cohort
2,288%
Cumulative ACSI Leaders return
since 2006, vs. 646% S&P 500
3.5×
Outperformance vs. S&P 500
over 15 years (ACSI top percentile)
Validated relationship How AI4EBITDA's measurement maps to ACSI's third-party performance record
Read carefully

The 1.4M number is our methodology dataset. The 350 number is ACSI's independent validation cohort. They are not the same thing and they are not interchangeable. The proof of the framework comes from how cleanly our scores predict ACSI's outcomes (p=0.019), not from the size of either number on its own.

The Causality Chain

Language today. Financial outcome later.

The causal chain runs in one direction: decision-quality language predicts mandate credibility, mandate credibility predicts execution, execution predicts customer satisfaction, and customer satisfaction predicts long-run shareholder return. Each layer takes twelve to thirty-six months to manifest. AI4EBITDA scores the leading layer.

T+0
Score
AI4EBITDA composite
1–100, authority-weighted. Available now.
T+6–18mo
CSat
Customer satisfaction
Score correlates with ACSI movement at p=0.019.
T+12–36mo
Rev
Revenue trajectory
CSat shifts compound into top- and bottom-line trajectory.
T+15yr
2,288%
Shareholder return
ACSI Leaders portfolio since 2006, vs. 646% S&P 500.
What the chain means

By the time a board sees a financial problem, the leadership decision that caused it was made two to three years earlier. AI4EBITDA scores the layer where that decision is still being made, and where intervention still moves the outcome.

Common Questions

What executives actually ask about the science.

We work with mid-market and enterprise companies, typically $100M to $1B+ in revenue, where board pressure on EBITDA and transformation delivery is real. The methodology scales. What matters is whether the gap between your mandate and your delivery is a problem worth solving right now. If you are smaller than $100M and growing fast, and you have a board or investors pushing for results, you may want to reach out to us for AI consulting work so we can scale your people, processes, and tech that align to your budget.
Myers-Briggs and DISC measure what people say about themselves. This measures what people actually demonstrate in their natural language. The thinking patterns that show up in public data before they show up in outcomes. It has been validated against 350 ACSI-tracked companies (independent third-party data) with statistical significance at p = 0.019. Our own methodology dataset covers 1.4M companies and 410M individual profiles. Nobody knows they have been assessed. And the accuracy on a sample of 1,000+ words is p = 0.001, one in a thousand chance of being wrong. The underlying psycholinguistic instrument is in use by 7 of the Fortune 50. Not because it is interesting, but because it is right.

1. People lie. Usually to themselves.

When you fill out a Myers-Briggs or DISC survey, you answer based on who you think you are, or who you want your boss to think you are. Psychologists call this social desirability bias. It is not dishonesty. It is human nature. People answer in ways that make them look good, capable, and fit for the role. The result? The assessment tells you what the person believes about themselves. Not what they actually do under pressure, at scale, in complexity.

2. Self-awareness is the thing we are trying to measure, not a tool we can rely on.

Here is the irony of every survey-based leadership tool: the people who need assessment the most have the lowest self-awareness. A narcissist does not know they are a narcissist. An obstructionist does not score themselves as difficult. A leader who is cognitively overloaded will rate themselves as highly adaptable. You cannot measure a blind spot by asking the person where they cannot see.

3. They produce personality types, not performance predictions.

Myers-Briggs gives you INTJ or ENFP. DISC gives you D, I, S, or C. What does that tell you about whether an engineer will produce 30% more barrel-equivalent output than the person next to them? Nothing. These tools describe personality style. We measure cognitive capacity, decision architecture, burnout trajectory, and execution readiness. Those are the things that actually connect to financial outcomes.

The proof is not a testimonial. It is mathematics.

We have validated our methodology against 350 companies using the American Customer Satisfaction Index, an independent third-party measure of actual business performance.

The result: Pearson p = 0.019. In plain English, there is less than a 2% chance the relationship between our scores and real business outcomes is random.

Cohen's d = 0.83 on rich data profiles. In academic research, anything above 0.8 is classified as a large effect.

This is not a sales claim. It is measured against the same statistical standard used in pharmaceutical trials.

We use only publicly available material that the subject already chose to publish. Profiles, websites, filings, press releases, public talks, published writing. The same footprint a journalist, analyst, or competitor could read.

No internal email. No Slack. No surveys. No interviews. No IT access. No private documents.

We report patterns at the team or company level rather than portraits of named individuals, with the exception of the executives explicitly scoped into the engagement. The methodology is published. The validation against ACSI at p = 0.019 is independent. Customers see exactly what we measured and how.

If a question on data ethics has not been answered for you here, raise it in the discovery call. We will walk you through what gets collected, what does not, and how it is governed.

Yes, on the cognitive variables the framework measures. We score outputs, decision patterns, and interaction language for an AI system the same way we score them for a human leader. What that gives you is comparability: a single scale where you can see whether a given leader and a given AI system are cognitively compatible for the workflow you want to run.

It does not mean AI is conscious or has personality. It means cognition is measurable, and we measure it consistently across both.

For mid-market companies under board pressure on AI investment, this is the difference between buying a tool and knowing whether the people you have can actually run it.

Ready to see what your language already says?

See a scrubbed sample report. Or get a benchmark on your own organisation. No surveys, no IT access, no disruption.