Methodology note · Version 1.0
How the FinAI Data Readiness Index is scored
FADRI measures whether a company's finance and business data foundation is strong enough to support trustworthy AI-enabled decision-making. This note sets out everything that turns 30 answers into a score: the pillars and their weights, the two formulas, the safeguard caps, the interpretation bands, how peer benchmarking works — and what the index cannot tell you.
- Readiness pillars
- 6
- Scored questions
- 30
- AI use cases
- 5
- Readiness scale
- 0–100
What the assessment asks
The assessment takes 15–20 minutes. It opens with 6 profile questions, which are never scored, and then asks 30 scored questions — 5 in each of the 6 pillars.
The profile answers — role, stage, revenue, sector and organisational complexity — do no work in the score at all. They exist to decide which companies a respondent is fairly compared with, which is the subject of the benchmarking section below.
Every scored question is answered on the same 0–4 maturity scale. Each question puts that scale into language specific to what it asks about, but the underlying rubric is this one:
| Score | Label | What it means |
|---|---|---|
| 0 | Not in place | No consistent practice or process exists |
| 1 | Mostly ad hoc | Occasional or inconsistent practice |
| 2 | Partly working | Practice exists but is incomplete or unreliable |
| 3 | Mostly reliable | Practice is consistent with occasional gaps |
| 4 | Strong and repeatable | Robust, consistent, and well-established practice |
The 30 scored questions and the wording of their answer options are not published here. Everything that turns those answers into a score is — the weights, the formulas, the caps and the bands — but an assessment whose answer anchors can be read in advance measures how well somebody read them. Participants see each question once, in the assessment.
The six pillars and their weights
The 6 pillars do not count equally. Each carries a fixed share of the overall score, and the 6 shares add up to the whole of it.
1. Business Visibility & Data Access
20%
5 questions
AI cannot help management if the business cannot reliably pull together the information needed to understand sales, margin, cash, customers, and operations. Without adequate data assembly, any AI output is built on incomplete foundations.
2. Decision-Grade Data Quality
25%
5 questions
AI does not just need data. It needs data that management would trust for a real decision. If the data is incomplete, contradictory, or routinely inaccurate, AI will make confident-sounding but unreliable recommendations, scaling errors instead of insights.
3. Operating Cadence & Responsiveness
20%
5 questions
AI is most useful when the business can refresh its view quickly and regularly enough to act. If reports arrive weeks late or forecasts are updated once per quarter, AI-assisted decisions will be based on stale information that no longer reflects reality.
4. Shared Business Definitions
15%
5 questions
AI becomes dangerous when teams use the same words for different things. If revenue, margin, churn, or burn rate mean different things to different people, AI-generated analysis will produce outputs that cannot be trusted or explained.
5. Trust, Checks & Exception Management
15%
5 questions
AI can generate answers quickly, but without checks, the business may act on the wrong answer faster. Strong exception management ensures that anomalies are investigated and that the AI-assisted numbers seen by management are validated before influencing decisions.
6. Accountability & Safe AI Use
5%
5 questions
Even lightweight AI readiness requires clear accountability: who owns the data, who owns the metrics, and what is safe to use with AI tools. Without this, organisations risk uncontrolled AI use, conflicting versions, and exposure to data privacy or compliance risks.
How a score is computed
Two steps, and no rounding until the result is shown to you.
First, each pillar. The 5 answers in a pillar are added up — giving a raw total between 0 and 20 — and that total is rescaled to run from 0 to 100.
Pillar score = (sum of 5 answers ÷ 20) × 100
Then, the overall score. Each pillar score is multiplied by that pillar's weight from the table above, and the 6 results are added together.
Overall FADRI score = sum of (pillar score × pillar weight)
Nothing is rounded along the way: the figures you see on a report are rounded only at the moment they are printed. A partly finished assessment is never scored — an unanswered question is not treated as a zero, because a low score and a missing answer are not the same fact.
The five AI use cases
Alongside the overall score, 5 specific uses of AI in finance are scored separately. Each is a different weighting of the same 6 pillar scores — no extra questions are asked — with one safeguard rule on top.
| Use case | Business Visibility & Data Access | Decision-Grade Data Quality | Operating Cadence & Responsiveness | Shared Business Definitions | Trust, Checks & Exception Management | Accountability & Safe AI Use |
|---|---|---|---|---|---|---|
| AI Forecasting & Planning | 20% | 25% | 25% | 10% | 15% | 5% |
| AI Pricing & Margin Decisions | 15% | 25% | 10% | 30% | 10% | 10% |
| AI Management Insight | 15% | 20% | 25% | 15% | 20% | 5% |
| AI Board & Investor Reporting | 10% | 20% | 20% | 20% | 25% | 5% |
| AI Business Performance Analysis | 20% | 25% | 15% | 20% | 15% | 5% |
1. AI Forecasting & Planning
Forecasting requires reliable source data (Pillar 1), high quality inputs (Pillar 2), and a regular cadence of refreshed views (Pillar 3). Controls are also important to prevent confident but wrong forecasts.
Safeguard cap. Capped at 49 if Decision-Grade Data Quality < 40 or Operating Cadence & Responsiveness < 40
2. AI Pricing & Margin Decisions
Pricing and margin analysis depends critically on consistent product, customer, and channel classifications (Pillar 4) and high-quality cost and revenue data (Pillar 2). Governance of who owns and can change pricing logic is also important.
Safeguard cap. Capped at 49 if Shared Business Definitions < 40 or Decision-Grade Data Quality < 40
3. AI Management Insight
Real-time management insight depends on cadence and responsiveness (Pillar 3) and strong controls to prevent bad data reaching senior decisions (Pillar 5).
Safeguard cap. Capped at 49 if Operating Cadence & Responsiveness < 40 or Trust, Checks & Exception Management < 40
4. AI Board & Investor Reporting
Board and investor reporting requires that definitions are stable and agreed (Pillar 4), that controls are strong (Pillar 5), and that numbers are delivered in a timely and consistent manner (Pillar 3).
Safeguard cap. Capped at 49 if Shared Business Definitions < 40 or Trust, Checks & Exception Management < 40
5. AI Business Performance Analysis
Commercial performance analysis depends on having accessible, structured data (Pillar 1), high quality inputs (Pillar 2), and consistent classifications for products, customers, and channels (Pillar 4).
Safeguard cap. Capped at 49 if Business Visibility & Data Access < 40 or Decision-Grade Data Quality < 40
The safeguard caps prevent a use-case score from indicating readiness when a critical prerequisite pillar is critically weak (below 40 out of 100). These caps reflect a practical judgment that some data quality conditions are non-negotiable for a given use case: for example, a company cannot safely use AI for forecasting if its input data quality or reporting cadence is below a minimum threshold, regardless of how well it scores on other pillars.
What a score means
A score on the 0–100 scale falls into one of five bands. The same five bands are applied to the overall score, to each pillar and to each use case.
- 80–100Strong
- The organisation has a well-developed finance and business data foundation. Reporting is reliable, data ownership is clear, and control discipline is strong. AI-enabled decision-making in finance is likely to deliver trustworthy and actionable outputs. Some optimisation opportunities may still exist.
- 65–79Emerging
- The organisation has meaningful strengths but noticeable gaps in one or more pillars. AI adoption in finance is feasible for selected, well-bounded use cases. Targeted improvements in weaker pillars will significantly increase the reliability and ROI of AI investments.
- 50–64Limited
- The organisation has a partial foundation. Some practices are in place but inconsistencies, ownership gaps, or reporting weaknesses create meaningful risk for AI-enabled decisions. An investment in data and reporting process improvement is recommended before broad AI adoption.
- 35–49Fragile
- The organisation's finance and data foundations are at an early or fragile stage. Manual workarounds, inconsistent reporting, unclear ownership, and limited reconciliation create high risk for AI outputs. AI adoption is likely to amplify existing data weaknesses rather than resolve them.
- 0–34At Risk
- The organisation's data foundation is not currently adequate for trustworthy AI-enabled decision-making. Significant foundational work is required before AI tools can be introduced safely. The priority is stabilising core finance processes, data ownership, and reporting discipline.
How peer benchmarking works
A score on its own is a number. A score against comparable companies is a finding — so every report says which group it compared you with, and how many companies were in it.
Companies are grouped by stage of growth, into Startups, Scaleups, Established SMEs, Multi-location / Franchise SMEs. From there the comparison is made as narrow as the available data honestly allows: the narrowest group below that has enough companies in it is the one used, and if none of them does, no benchmark is shown at all.
| Compared on | Companies needed |
|---|---|
| Stage and industry | 30 |
| Stage and revenue band | 15 |
| Stage | 15 |
The floors are the point. Below 15 companies a percentile is arithmetic rather than evidence, and adding industry to the comparison costs more because industry cuts the group down hardest. A report whose cohort has not filled up yet says so, rather than quoting a position against a handful of companies.
Where a benchmark is shown, the percentile is the share of the cohort scoring strictly below the respondent, and the respondent is not counted among their own peers. Only completed assessments count.
What this index does not tell you
Every diagnostic has an edge, and an instrument that does not publish its own is asking to be over-read. These are FADRI's.
- It is self-reported, and nothing is audited
- Every answer is the respondent's own assessment of their business. No system is inspected, no ledger is reconciled and no figure is verified against anything. The index measures what a company believes about its data, which is a useful thing to know and is not the same as what is true of it.
- One person answers for the whole organisation
- There is one respondent per assessment. A CFO, a financial controller and a head of operations may read the same business differently, and the score reflects whoever sat down to answer it rather than a consensus across the finance function.
- The arithmetic is exact; its inputs are judgments
- Every score is computed the same way for everybody, and nothing is rounded before it is presented. But the 0–4 scale each question is answered on is a set of qualitative descriptions, so a precise number is being produced from an approximate reading. Treat a score as a position, not as a measurement.
- A safeguard cap is a judgment, not a computation
- Where a use case depends on a pillar that is critically weak, its score is held at a ceiling regardless of how well everything else scores. That is a deliberate editorial rule about what is non-negotiable for that use case, and a capped score is not the weighted result the other numbers would suggest.
- The peer group is other participants, not a market
- A benchmark compares a respondent with other companies that have completed this assessment. It is not a representative sample of any industry or any economy. No cohort is published below 15 peers, and where the narrowest peer group is too small the comparison widens to a broader one — the report always names which group it used and how many companies were in it.
- A score belongs to the day it was taken
- Scores are computed once at completion and stored, so a report says the same thing next year as it did on the day it was produced. It is stamped with the methodology version that produced it — currently 1.0 — and scores from different versions are not compared with each other.