Methodology v1.2
The founding rule
The Index scores a company's public footprint, not the company. A firm can be strong internally and score low here; the score describes only what is publicly visible.
Everything below is the whole of it. The same document is served as JSON at https://www.nikksi.com/api/v1/methodology, read from the same source, so the version an agent gets and the version you are reading are the same words.
The five dimensions, and what each score means
Each dimension is anchored at four points. A score sits at an anchor when the footprint matches its description, or between two when it sits between them. The anchors describe an organisation's published material — not the organisation.
Prompting
Evidence window 24 monthsHow the organisation communicates with AI systems — the context, structure and constraints it gives them.
What typically moves thisThe step from 45 to 65 is naming AI skills or tools in job adverts, or publishing guidance for staff on using AI. Reaching 85 takes specific AI requirements across more than one role, alongside evidence of internal training or published usage standards.
- 25Footprint examined, no AI-skill signal found anywhere in it, including live job adverts.
- 45Generic claims about AI capability, with no specifics behind them.
- 65Job adverts name AI skills or tools, or the organisation has published staff guidance on using AI.
- 85Multiple roles carry specific AI requirements, alongside evidence of internal training or published usage standards.
Tooling
Evidence window 24 monthsWhich AI tools the organisation uses and how deliberately it selects them.
What typically moves thisThe step from 45 to 65 is naming the specific tools in use and tying them to the functions that use them, rather than mentioning AI tooling in general. Reaching 85 takes named tools across more than one function, with integrations or builds referenced.
- 25No tools named anywhere in an examined footprint.
- 45One generic mention of AI tooling, with nothing named.
- 65Specific tools named and tied to the functions that use them.
- 85Multiple named tools across more than one function, with integrations or builds referenced.
Workflow Integration
Evidence window 24 monthsHow far AI is embedded in routine work rather than used ad hoc.
What typically moves thisThe step from 45 to 65 is describing at least one concrete process or offering with AI in the loop, rather than claiming AI enhancement without a mechanism. Reaching 85 takes AI embedded across several described workflows.
- 25No AI appears in any described process.
- 45"AI-enhanced" claims with no mechanism described.
- 65At least one concrete described process or offering with AI in the loop.
- 85AI embedded across several described workflows.
Critical Judgement
Evidence window 24 monthsHow the organisation evaluates AI output — checking accuracy, handling error, and deciding when not to use AI.
Not assessable where the footprint makes no AI claims at all: this dimension returns null rather than 25, because there is nothing claimed to govern or sustain.
What typically moves thisThe step from 45 to 65 is publishing an AI or data policy, disclosing where AI is used, or stating a human-review practice, rather than carrying a boilerplate responsible-AI line. Reaching 85 takes a substantive policy alongside disclosure and honest statements about limits.
- 25AI claims are made but no policy, disclosure or oversight statement appears anywhere, or visibly unedited AI-generated content has been published.
- 45A boilerplate responsible-AI line and nothing further.
- 65A published AI or data policy, or disclosure of AI use, or a stated human-review practice.
- 85A substantive policy, plus disclosure, plus honest statements about limits.
Confidence & Momentum
Evidence window 12 monthsThe organisation's stated confidence with AI and the direction of travel over time.
Not assessable where the footprint makes no AI claims at all: this dimension returns null rather than 25, because there is nothing claimed to govern or sustain.
What typically moves thisThe step from 45 to 65 is a regular pattern of AI-related output across the last twelve months, rather than sporadic mentions. Reaching 85 takes that cadence plus a stated direction — leadership statements, successive hires, an evolving offering.
- 25AI mentions exist, but nothing new in twelve months or more.
- 45Sporadic AI-related output, with no pattern to it.
- 65Regular AI-related output over the last twelve months.
- 85A sustained cadence plus a direction — leadership statements, successive hires, an evolving offering.
The rules a score has to satisfy
Evidence windows
Evidence for Prompting, Tooling, Workflow Integration and Critical Judgement must have been published or updated within 24 months of the date it was read. Confidence & Momentum reads a 12-month window, because it is about direction of travel rather than what exists.
Examined and silent is a finding
A footprint that was examined and found to say nothing on Prompting, Tooling and Workflow Integration scores 25 on that dimension, and the URLs that were examined are cited as the evidence. Silence that was actually looked for is a finding, not a gap.
A thin footprint is not a low score
A footprint too thin or too dormant to examine is different. That dimension is published as null and labelled "insufficient public signal". It is never given a low score, because there was nothing to read rather than nothing to find.
No AI claims means two dimensions are not assessable
Where an examined footprint makes no AI claims of any kind, Critical Judgement and Confidence & Momentum are not assessable and return null — a company is not scored on governing or sustaining activity it does not claim to have. "Examined and silent" scoring at 25 applies only to Prompting, Tooling and Workflow Integration.
Two independent items, minimum
A dimension carries a score only when at least 2 independent evidence items support it. One item is an observation; it is not a score.
Self-published evidence caps a dimension at 65
Evidence published by the organisation about itself can take a dimension no higher than 65. Above that, at least one third-party source is required — a job board, a partner listing, a press or industry write-up, a public repository.
How the overall score is formed
The overall score is the mean of the dimensions that carry a score. Dimensions published as null are left out of the mean rather than counted as zero, and no overall score is published at all unless at least 3 dimensions are scored.
Every score is dated and versioned
Each record carries the date it was scored and the methodology version used. A re-score updates that date and republishes the evidence behind it. Scores are not edited silently.
Every score carries a review-due date
Every published score carries a review-due date six months after scored_at. Scores past review are marked "review overdue" until rescored. Evidence links may age; the dates disclose this.
Rescores are free and unconditional
Rescores are free, on request, for any company, and are never conditional on purchasing anything from Nikksi. A company that improves independently is rescored on identical terms to a client. The evidence standard never varies.
Bands on the overall score
- 0–59ExploringPublic signal shows early or occasional AI use, without a settled pattern.
- 60–79ApplyingPublic signal shows AI used in repeatable ways across more than one area of work.
- 80–100AdvancedPublic signal shows AI embedded in routine work, with stated review practice.
A dimension published as null is labelled “insufficient public signal”. It is not a zero, and it is left out of the overall mean rather than counted as one.
Review cadence
Every published score carries a review-due date 6 months after scored_at, derived from that date rather than stored. A record past its review date is flagged "review overdue" and keeps its last published score until it is re-scored. Evidence links may age between reviews; the dates on a record disclose how old the reading is.
The review date is derived from the date a record was scored rather than stored alongside it, so it cannot drift from the reading it describes. Both dates appear on every record, on the page and in the JSON.
For AI agents and developers
This methodology as JSON
The anchors, the rules and the founding rule, machine-readable and versioned. Attribution is the only condition: CC BY 4.0 — attribution with a link to nikksi.com required
Cite as https://www.nikksi.com/readiness-index/methodology. A change to how the Index scores is a new version, not an edit to this one.
Your own score
Get your personal AI score
This methodology reads organisations from the outside. The assessment scores you from the inside — twelve questions, about five minutes, the same five dimensions.