How we review and score South African providers

Most comparison sites will not tell you how their rankings are produced, because the honest answer is often that the highest bidder is at the top.

That is the thing this page exists to rule out. Below is the actual method: the weights, the data sources, the statistical correction we apply to review scores, what happens when we cannot score something, and what a sponsor does and does not get for their money.

If any of it does not match what you see on a page, that is a defect and we want to hear about it. There is a correction route at the bottom, and it is free whether you pay us or not.

The five dimensions, and what each one weighs

One rubric, applied to every provider in a category. No provider is graded on a curve.

Scores are published to one decimal place. A score above 7.0 has to cite verifiable evidence on the page carrying it, not an assertion.

Where the numbers come from

Every published price, contract term and feature flag is a row in a fact layer, and every row carries the URL it came from and the date it was read.

There are currently 745 publishable facts behind the comparison and review pages, most recently sampled 11 October 2026. If we cannot source a figure, we do not print it. That rule is why some cells on our comparison tables say nothing rather than guessing.

The sources are the provider’s own published material, official newsrooms, and regulators such as ICASA. We do not take figures from third-party aggregators, directory scrapes or search-result snippets.

That last one is not a stylistic preference. We have repeatedly measured the same provider reading higher in a search-result snippet than on its own live profile, on the same day. Anyone harvesting from snippets is publishing numbers that flatter everybody.

How the customer-review dimension is calculated

A raw star average rewards obscurity. We correct for that, and the correction changed real scores.

We read Google and HelloPeter directly from their profile APIs rather than from search results. HelloPeter’s TrustIndex is scored out of 10 and is not a rescaled star average – it weights recency, volume and response rate – so the two are never mixed under one label.

The figures are then pulled toward the cohort average using Bayesian shrinkage, and weighted by the log of review volume. In plain terms: a provider with ten glowing reviews does not outrank one with tens of thousands of mixed ones simply for being small.

This matters because the uncorrected version produced exactly that. Before the fix, a provider with ten reviews outranked a major ISP with tens of thousands of them, purely on the arithmetic of a small sample. The corrected method moved scores across the corpus, mostly downward, and we published the new numbers.

What happens when a dimension cannot be scored

Some providers have no South African public review presence. We do not invent one.

Where the customer-review dimension cannot be assessed, its 20% is redistributed proportionally across the remaining four, and the page says so in visible text rather than in a footnote.

That visibility is deliberate. Quietly re-weighting a large incumbent while smaller competitors carry the full five dimensions would read as favouritism even where it is not. If a score was computed on four dimensions instead of five, the page you are reading it on will tell you, and it will tell you why.

What sponsorship buys, and what it cannot

Sponsors get disclosed brand presence and first sight of matching leads. They do not get a better score, a higher position, or a softer verdict.

Directory listings are free and uniform for every provider in South Africa. No badges, no boosted positions, no featured slots. A provider that has never paid us a cent is listed on the same template as one that pays monthly.

The proof is not the promise, it is the record. When we rebuilt the scoring method in August 2026 and republished, the scores moved – and they moved against paying partners as readily as anyone else. The highest-scoring provider on the site at the time was a sponsor, and its score went down. So did another sponsor’s. The largest single drop in the corpus went to a provider that pays us nothing, because the old method had been flattering its review standing.

A scoring system that never costs a paying customer anything is not a scoring system. Ours has, publicly, and the old numbers are in the record.

Corrections, disputes and re-verification

Every review page carries a dated verification stamp. If the date looks old to you, it probably is, and we would rather you said so.

If you are a provider and something on your page is wrong, send us the correction with a source and we will verify it and update the fact, the score if the fact feeds one, and every page that renders it. This is free and it is not conditional on any commercial relationship.

If you are a buyer and a published price does not match what you were quoted, tell us that too. Providers change rates and our sample date tells you when we last looked, but a gap between our figure and your quote is something we want to chase.

We correct visibly. Where a published score or fact changes materially, the change is recorded rather than quietly overwritten.

Related