How we rank businesses

What a score is built from

Every score on every directory we publish is built from public customer reviews. Nothing else feeds it. Not payment, not advertising spend, not whether we have ever spoken to the business. The same engine runs behind all the independent Singapore service directories published here.

Four things are read out of those reviews.

Volume. How many people have written about this business. A five-star average from three reviews and a 4.6 average from four hundred are not the same claim, and the score treats them differently.

Recency. When those reviews were written. Trades change hands, technicians leave, and a business that was excellent in 2022 may not be excellent now. Recent reviews carry more weight than old ones, which means a score can fall without a single new complaint simply because the good reviews aged.

Distribution. The shape of the ratings, not just the average. Twenty five-star reviews and five one-star reviews average out respectably, but that spread describes a business that is inconsistent, and inconsistency is exactly what a householder is trying to avoid. A flat 4.4 with almost no one-star reviews is a stronger signal than a 4.6 with a visible tail.

Content. What reviewers actually say about the parts of the job that matter for that particular trade. This is the part that differs between directories, and it is worth explaining properly.

Why the criteria differ by trade

The engine is the same everywhere. What it reads most closely is not.

Take the Singapore Locksmith Guide. A locksmith is almost always an emergency purchase, made by somebody standing outside their own front door. What matters is how long the wait actually was, and whether the fee quoted on the phone survived the visit. Reviews describing those two things are weighted up, because they describe the failure modes that actually ruin a locksmith job.

Now take aircon. Nobody has an aircon emergency at 1am, but almost everybody buys aircon servicing repeatedly for years. So the useful signal is consistency: whether the fourth visit went as well as the first, and whether a routine servicing quote turned into a chemical overhaul once the technician was on the ladder. A locksmith’s weighting would be the wrong instrument for that trade entirely.

Plumbing splits the difference, and the weighting follows: emergency availability for the unplanned half of the work, and whether a quote held once someone opened up a wall for the rest.

None of this is a secret formula. It is the ordinary observation that different trades fail in different ways, applied consistently and written down so a reader can disagree with it.

What the score never uses

Three inputs are excluded absolutely, and the exclusion is structural rather than a promise of good behaviour.

Payment. No business can pay for a better position. There is nothing to buy today, and the ranking engine is built so that it cannot read payment status even if there were.

Advertising spend. Whether a business advertises anywhere, including with us if that ever exists, has no bearing on where it ranks.

Commercial relationship. We hold no ownership stake in any listed business and no listed business is a client. The full independence position is set out in our editorial standards.

This is what “featured-blind” means. It is not a claim that we resist temptation. It is that the ranking process is not given the information it would need to be corrupted by it.

How the four inputs combine

People reasonably ask whether there is a formula, and whether we will publish it to the decimal place. The honest answer is that publishing exact coefficients would be less useful than it sounds, because it invites optimisation against the number rather than the thing the number is trying to measure. What we will do is describe the shape of it, which is what actually lets you audit a ranking.

Volume acts as a confidence gate rather than a straight multiplier. Below a threshold of reviews, a business simply cannot reach the top of a list, however good its average, because there is not enough evidence to justify the position. Above that threshold, extra volume adds progressively less: the difference between fifty reviews and five hundred matters far less than the difference between three and fifty.

Recency is a decay, not a cliff. A review from four years ago still counts, it just counts for less than one from last quarter. This is why a business can drift down a list during a quiet period without receiving a single complaint.

Distribution acts as a penalty on inconsistency. A visible tail of one-star reviews pulls a score down harder than the raw average would suggest, because a directory exists to help someone avoid a bad outcome rather than to find the theoretical best.

Content is the trade-specific weighting described above, and it is the part that most affects the order of the top few entries, where averages are usually close enough to be indistinguishable.

The monthly re-check

A directory that is compiled once and left alone starts lying within a year. Businesses close, phone numbers get reassigned, companies drop a trade or add one.

So the whole listing set is re-checked every month. Three things happen at each re-check. Closed businesses are removed automatically, without anyone having to report them. Validity checks are re-run, so a listing that no longer passes comes out. And scores are recalculated against the current review set, which is why a business can move without doing anything differently: its competitors gained reviews, or its own aged.

Every listing shows the date it was last re-checked. If that date looks stale, the score is stale, and you should read it accordingly. Publishing the date is deliberate: it is the only way a reader can tell the difference between a directory that is maintained and one that was compiled once and abandoned, and the two look identical otherwise.

What this method cannot tell you

A score is evidence, not a guarantee. It is built from what previous customers chose to write in public, which leaves real gaps: price competitiveness is largely invisible in reviews, work that nobody reviewed is invisible entirely, and a genuinely good new business with eleven reviews will score below a mediocre one with four hundred.

We set those limits out at length in what our scores do and do not measure, and we would rather a reader used a score as one input among several than treated it as a verdict.

If something on a directory looks wrong to you, or you are a business owner who disagrees with a score, the process for saying so is at corrections, disputes and removals. It is free, and removal requests are honoured.