What our scores do and do not measure

What the score can actually see

A score on any of our independent Singapore service directories is built from public customer reviews and nothing else. The inputs are set out in full in how we rank businesses: volume, recency, distribution, and what reviewers say about the parts of the job that matter in that trade.

Read across hundreds of reviews and years, that turns out to capture a few things well.

Patterns, rather than incidents. One angry review tells you almost nothing. Forty reviews over three years, of which six describe the same failure in the same words, tell you something real. Aggregation is the whole value here.

Consistency. The distribution matters as much as the average. A business with a 4.4 and almost no one-star reviews is a safer bet than one with a 4.6 and a visible bad tail, because most people are trying to avoid a disaster rather than optimise for a triumph.

Direction of travel. Recency weighting means a business that has slipped in the last eighteen months shows it, even if its lifetime average still looks respectable.

What it cannot see

Three blind spots are structural. They are not bugs we intend to fix, because no review-based method can fix them.

Price. Reviews say whether a price felt fair, which is not the same as whether it was competitive. Somebody delighted by a fast response rarely knows they paid thirty per cent above the going rate. A score on the Singapore Aircon Guide will tell you whether a company’s quotes tend to hold; it will not tell you whether they are cheap. Get your own quotes.

Work nobody reviewed. Most jobs never generate a review at all. Commercial contracts, maintenance retainers, and quiet repeat work are largely invisible, so a company doing excellent unglamorous work for the same twenty clients for a decade may look thin next to a consumer-facing competitor with an active review habit.

A good new business. This is the one that bothers us most. Volume is an input, so a genuinely excellent company that opened last year, with eleven reviews, will rank below a mediocre one with four hundred. That is a real cost of the method and there is no honest way around it: we cannot credit quality that nobody has yet written about. If you are shopping and a newer name looks promising, the score is not the tool for evaluating it.

Two things reviews systematically distort

Beyond the blind spots, there are two biases worth knowing about, because they affect how you should read any review-based number, ours included.

Reviews cluster at the extremes. People write when something went unusually well or unusually badly, and stay silent when a job was simply fine. That means a review corpus over-represents the tails of a business’s performance and under-represents its ordinary output, which is most of it. Distribution weighting helps, but nothing removes the effect entirely.

Who asks, gets. A company with a habit of asking every satisfied customer for a review will accumulate a better profile than an equally good one that never asks, regardless of the work. That is not manipulation, it is diligence, and it is invisible from the outside. It is also a real reason a directory should not be read as a ranked list of competence.

Both are arguments for treating a score as a filter rather than a verdict.

Evidence, not a guarantee

A high score means a lot of people, over a sustained period, described this business doing the job properly. That is genuinely useful. It is also not a promise about your particular job.

Any business can send its weakest technician on the day you call. Any business can be having a bad month that has not reached the reviews yet. The score describes the past behaviour of a company, in aggregate, as observed by strangers.

The sensible way to use it is as one filter among several. Use it to build a shortlist, then do the things a score cannot do for you: describe your actual job, get a quote in writing, ask what happens if the work needs revisiting, and notice how the conversation goes.

What a low score does and does not tell you

Symmetry matters here, so it is worth saying the reverse of the above.

A low score is not proof that a business is bad. It can mean a thin review record, a young company, a trade served mostly by word of mouth, or one visible bad patch that the recency weighting has not yet let go of. Read the count alongside the score before drawing conclusions.

What a low score does tell you is that the public evidence to reassure you is not there. That is a reason to ask more questions, get the quote in writing, and check what happens if the work needs revisiting. It is not a reason to assume the worst.

Where it goes wrong

Scores can be wrong for reasons that are our fault rather than the method’s: reviews attributed to the wrong branch, a business placed in a category that flatters or damages it unfairly, a validity check that misfired. Those are correctable, and the route is corrections, disputes and removals.

Scores can also be distorted deliberately, which is a different problem with its own honest limits. That is covered in how we handle fake and incentivised reviews.