The method.

Two sources. Large UK companies are legally required to publish how quickly they pay their suppliers — we import every filing that’s in the plausible range (median between 1 and 365 days) and store the full history per company. Every invoice sent through Kalnov, when a payment is recorded against it, contributes its days-to-pay to the company’s network score.

A public-filing score is the average time to pay from a company’s most recent filing. A network score is computed from Kalnov invoices with the formula below. When both exist for a company, we show both.

Every score starts at 100. Four things can bring it down. Each is capped, and the final score is clamped to the 0–100 range.

The biggest driver. We take every paid invoice for the company inside the recency window, compute how many days past due each one was, and take the median of that list. A value of 0 means the median invoice was paid on the due date; a value of 30 means the median invoice landed a month late.

The deduction is banded:

• 0 days or fewer−0
• 1 to 7 days−5
• 8 to 14 days−12
• 15 to 30 days−25
• 31 to 60 days−40
• 61 to 90 days−55
• over 90 days−65

We use the median rather than the mean so a single catastrophic invoice can’t drag the score down on its own. If ten invoices all paid on time and one paid six months late, the median is 0 and the deduction is 0 — the outlier is described separately by the late-frequency component below.

The percentage of paid invoices that arrived after their due date, multiplied by 0.15, rounded, capped at 15.

A company that pays half its invoices late loses 8 points here (50 × 0.15 = 7.5 → 8). A company that pays every invoice late loses the full 15. This is the component that surfaces companies whose median is acceptable but whose consistency isn’t — the kind that pay four out of five invoices in a week and then one invoice in three months.

Two deductions for invoices that are past their due date and still unpaid at the moment we compute the score.

• Over 60 days overdue: each such invoice adds 10 to the deduction (contributor weighting applied — see below). Capped at 20.
• Over 120 days overdue: each such invoice adds 15 to the deduction (also weighted). Capped at 30.

These are cumulative — an invoice that’s 180 days overdue counts for both bands, and the two caps apply independently. A company sitting on a pile of deeply overdue invoices loses up to 50 points from this alone, before any of the other components have fired.

The biggest risk to a network score is one heavy contributor swinging it — a single freelancer with thirty invoices to one company can’t be allowed to define its record. Every contributor is capped at a share of the total, based on how many distinct contributors the company has:

• 2 to 5 contributors60% max
• 6 to 10 contributors45% max
• 11 or more30% max

A contributor over their cap has their invoices proportionally down-weighted so their total contribution matches the cap. The tighter the network gets, the tighter each individual’s influence.

We also compute the score twice: once with the outlier cap applied, once without. If the capped score would otherwise be more than 20 points higher than the uncapped version, we take the uncapped score plus 20. This prevents the cap from turning a genuinely bad-payer signal into a good-payer score just because one heavy contributor happens to be a victim of them.

Only invoices issued in the last 36 months are counted. A company that used to pay slowly but has cleaned up their payment behaviour over the last three years shouldn’t be haunted by invoices from a decade ago; a company that’s always paid slowly will still show it in the recent window.

The public-filing side does the same thing through the filing cadence — the latest filing covers the most recent six-month reporting period, and the payment history chart on a company’s page shows every historical filing so you can see the trend.

A network score becomes visible on the site only when both of these are true:

• At least two distinct contributors have invoiced the company through Kalnov.
• At least three invoices exist for the company in total.

Below either threshold the company still has a score computed, but it’s marked internally as invisible and doesn’t surface. This is the point below which we don’t have enough data to publish a score with any confidence. Public-filing scores have their own thresholds at the source: only companies above the UK government’s size cutoff are required to file, so anything smaller only appears through the network side.

A company that thinks a score is wrong can dispute it. We look at the underlying invoices and correct the record if the evidence supports it — not because a company doesn’t like the answer, but because a specific payment event was misdated or misattributed. The score stays visible during a review, with a badge saying it’s under review.

Read the full disputes policy →

No opinions. There are no stars, no comments, no reputational ratings anywhere on Kalnov. If a client paid on time, the record says they paid on time, regardless of whether the freelancer thought the client was pleasant to work with.

We also don’t count draft or cancelled invoices, only invoices that were sent. And voided payment events (payments that were later reversed or reclassified) drop out of the calculation on the next recompute.

If we change how the score is calculated, this page gets updated first and every affected score gets recomputed on the next scheduled sweep. The current formula and its constants are the same ones in the source code that runs the calculation — there’s no separate published-vs-actual version.