How the AI score is built
The inputs, dimensions and weights behind the AI score, and what a missing input does to it.
Updated 1 October 2026
On this page
The AI score is a single 0 to 10 number built from five weighted dimensions, each scored from a company's own reported numbers. It is a compact summary of the evidence, not a recommendation: it never reaches for a confident score it cannot support, and it says so plainly when it cannot.
The five dimensions and their weights
| Dimension | Weight | What it looks at |
|---|---|---|
| Quality | 25% | Returns on capital (ROE, ROCE), margins, Piotroski score, cash conversion, and how many of the last several years were profitable |
| Valuation | 20% | P/E against the sector median, P/B against what the company's ROE would warrant |
| Growth | 20% | Multi-year revenue, profit and EPS growth, and whether profit is keeping pace with revenue rather than lagging behind it |
| Risk | 20% | Debt to equity, Altman Z-score, promoter pledging, market cap and traded value, and the stability of earnings and cash flow |
| Momentum | 15% | RSI, MACD, Supertrend, position against the 50 and 200-day moving averages, and the 52-week range |
Each dimension is itself built from several signals grouped by what they measure. Correlated signals, such as ROE and ROCE, are averaged within one group first and only then weighted against the other groups, so one underlying strength does not count twice.
What a missing input does
A missing signal is dropped, not guessed at or treated as an average. If a whole dimension has no usable input, for example a company with no multi-year financial history for Growth, that dimension is excluded and the other weights are rescaled to fill the gap, rather than the dimension being scored as a middling 5.
Below three of the five dimensions having data, there is no headline score at all: the page reports "Insufficient data" rather than a number built on too little.
A dimension resting on a single signal group also cannot score above 7.5, and one built on two groups cannot exceed 9.0. This ceiling never pulls a poor score up; it only stops a great score from being awarded on thin evidence.
When a dimension is critically weak
A weighted average alone would let one very poor dimension hide behind three good ones. So a dimension that scores below 2.0 out of 10 applies a direct deduction to the overall score, growing sharply the closer that dimension sits to zero. Risk carries the largest possible deduction, since a company that may not be able to service its debt is not redeemed by looking cheap; Quality and Growth carry smaller ones. The total deduction across all dimensions is capped, so no single weakness can collapse the score to zero on its own.
Mixed signals
When one dimension scores strongly (7.5 or above) while another scores weakly (3.0 or below), the verdict reads "Mixed" instead of the band the number would otherwise suggest, and names both sides. A high score and a live contradiction are not shown as if they agree.
Confidence
Alongside the number, DocStoX shows how much to trust it: confidence is set by the weakest dimension behind the score, not the average of all of them, and drops further when two or more dimensions had to rely on low-confidence evidence, or when the underlying data is stale.
See The AI score and verdicts for how the score is shown on a stock page.
Related
Was this page helpful?
Something wrong or missing on this page? Email [email protected]