How the AI
score works

Three questions, in sequence

A company with highly automatable work is not automatically an attractive acquisition. If competitors can reproduce the same savings, that value may pass to customers through lower prices. The score therefore separates improvement potential from durability.

A and B measure how much the business could improve. C determines how much of that opportunity survives the durability scenario.

Evidence moves a prior

Each axis begins at a prior score. The current priors are neutral defaults of 50; industry- and revenue-specific priors have not yet been calibrated from investment outcomes. Listing evidence supplies an integer delta rather than a new absolute score. The server ignores model-supplied totals and recomputes every axis, cap and composite.

Evidence limits

Listing evidence	Maximum movement	What it means
Thin	±10	Short teaser or material fields missing
Moderate	±18	Enough facts to support a direction
Rich	±25	Detailed operating and financial evidence

Evidence density caps individual axis movement. Confidence is separate: it discounts the combined A/B upside later. The application recalculates the fixed arithmetic; the model interprets evidence and proposes deltas.

Axis A · Compression upside

Axis A estimates the economically addressable labour pool. It distinguishes work that software can absorb from physical execution and accountable judgement that must remain.

Workforce assumptions

Work type	Examples	Compression rate
Transactional	Orders, invoicing, forms, data entry	80%
Repetitive judgement	Quoting, dispatch, shifts, inspection	60%
Tacit expertise	Craft judgement, on-site diagnosis	30%
Accountable judgement	Negotiation, supervision, licensed sign-off	20%
Non-knowledge work	Physical execution or required presence	0%

The target zone

If no non-compressible core is identified—such as a licence, physical asset, local density, certification, installed capacity, or embodied service—Axis A is capped at 40 regardless of the ratio.

Compression is not immediate EBITDA. In Japan, savings generally emerge through attrition, avoided replacement, or productive redeployment. The conversion horizon is a diligence signal, not part of the formula.

Axis B · One credible growth engine

Four revenue plays are scored independently. Axis B takes the maximum because a deal needs one strong growth engine, not four average ones.

A speculative play can dominate the result. Evidence limits and the confidence discount restrain that risk; the winning play still requires diligence.

Axis C · Durability

If intelligence is abundant and cheap around 2030, is this company still worth owning?

The thesis is that information processing becomes less scarce while licences, installed assets, local density, physical capacity, and hard-to-hire labour remain scarce for longer.

Durability design

Signal	Design weight	More durable	Less durable
Constraint type	25%	Licence, assets, local density	Matching or unprotected relationships
Occupation tightness	25%	Difficult to hire	Surplus applicants
Intermediation	20%	Earns from doing work	Earns from transaction friction
Demography	15%	Benefits from ageing	Depends on population growth
Competitor adoption	10%	Fragmented, slow peers	Well-capitalised peers
Robotics surface	5%	Automation adds capacity	No useful deployment surface

These weights describe the investment design, but Axis C is not calculated as a literal weighted spreadsheet. The model considers the signals and evidence, supplies a delta, and the application calculates prior plus delta.

Gates and disposition

Four structural gates are assessed after scoring. Each is pass, fail, or unknown. Missing broker information does not count as a failure.

A numerical score is retained for validated rejected records. That preserves the distinction between an economically interesting business and a structurally eligible acquisition. Responses with unverifiable evidence, invalid deltas, missing fields or schema errors are retried and cannot replace a trusted score.

How to use the output

Read the base composite and final score together. A large gap indicates attractive improvement potential with weak durability. A small gap means more of the opportunity is supported by durable scarcity.

Current limitations

1. 01

   Neutral priors remain provisional and need calibration from human comparisons and investment outcomes.

2. 02

   Marketplace descriptions are incomplete and uneven; the score can partly reflect broker disclosure quality.

3. 03

   Workforce composition and wages are screening proxies, not company-level payroll diligence.

4. 04

   Conversion horizon and risk flags inform review but do not enter the numeric formula.

5. 05

   Axis C is a scenario weight, not a prediction that abundant intelligence will arrive on a fixed schedule.

The goal is not to remove judgement. It is to make the judgement consistent, inspectable, and easier to challenge.

Return to browser →