Two suppliers. Both look competent. Neither has a rating anywhere, or if they do it is a handful of reviews that tell you nothing. You have to pick one, and the usual method — see which one has more stars — is simply unavailable.
For the broader context behind this business guide, read How Business Directories Actually Make Money: The Five Models.
This is a normal situation rather than an unlucky one, and there is a method for it. It produces a table rather than a verdict, and that is deliberate. The point of the exercise is to move the decision from "which one feels better" to "which risks am I actually taking, and which supplier has more of the things that would reduce them."
What this is, and who it is for
This is for anyone choosing between two small businesses with no useful public feedback: a fabricator, a freelance designer, a small software product, a workshop, a supplier of something you will have to reorder.
It works best when the two are genuinely comparable and you have a real decision to make. It works badly as a general ranking exercise, and it is not designed to produce one. By the end you will have a filled-in table and the decision will still be yours.
Why the absence of ratings is normal for small businesses
A rating is a statistical instrument, and statistical instruments need volume. A small business serving a few dozen customers a year, most of whom are perfectly satisfied and none of whom think about leaving feedback, will never accumulate enough to say anything.
Consider a five-star average built from eleven reviews. Ask what would have to change for it to read differently. One dissatisfied customer moves it noticeably. One friend moves it back. The people who left them are not a sample of the customers — they are a sample of the customers who write reviews, which is a different and much stranger population, skewed toward the delighted and the aggrieved with very little in between. The average is arithmetically correct and informationally close to empty.
Worse, the absence of reviews carries no signal either. A supplier with none might be new, might be small, might work entirely through referral, or might simply never have asked. You cannot distinguish these from outside, so you cannot treat "no reviews" as a mark against anyone.
Where reviews do exist in volume, they can be read as evidence — but as evidence about specific recurring problems, not as a score. How to read a review section when there is one covers that skill separately. This post assumes you do not have that option.
Step 1: write down what would actually make this go wrong
Before comparing anything, spend three minutes on the failure you are actually trying to avoid. Be specific to this purchase. The generic answer ("they might be bad") is useless; the specific answer generates everything else in the method.
Typical honest answers look like:
- It arrives late and I miss the date it was needed for.
- It arrives and is not what I described, and I have to start again.
- They stop responding halfway through and I cannot reach anyone.
- It works now and there is nobody to fix it in a year.
- The price moves after I have committed.
- They cannot handle the volume I will actually need in month three.
Pick the two or three that would genuinely hurt. Most people, when they do this honestly, find that their real fear is not quality at all — it is being left unable to reach anyone once money has changed hands.
Step 2: turn each risk into something you can check
Now convert each risk into an attribute you can verify from outside, today, without asking either supplier a favour. The conversion is the whole trick, and the rule is that the attribute has to be observable rather than inferred.
| The risk | An attribute you can actually check |
|---|---|
| It arrives late | Whether lead times are stated anywhere, and how specific they are |
| It is not what I described | Whether there is a described process for agreeing the spec before work starts |
| They stop responding | How many independent routes exist to reach a person, and whether one is named |
| Nobody to fix it later | Whether anything on the site indicates continuity — dated updates, a support route, a stated policy on older work |
| The price moves | Whether pricing, or a pricing basis, is stated before contact |
| They cannot handle the volume | Whether the site describes the size of work they normally take |
Notice what these have in common. Every one is a question about what the supplier has written down, not about how good they are. Written-down things are checkable. Goodness is not, which is why the ratings industry exists and why it does not work well at this scale.
Two more attributes belong on every table regardless of the risks you listed, because they determine what happens when anything goes wrong:
Reachability. How many distinct ways can you contact a person, and is any of them attached to a name? A single contact form is one route with no fallback.
Responsiveness. Send both suppliers the same short, specific, easy question. Not "tell me about your services" — something answerable in two lines that requires them to have read what you wrote. Note how long each takes and whether the answer engages with the question. This is the only attribute here you generate yourself, and it is usually the most informative one on the table.
Step 3: build the table
Attributes down the left, suppliers across the top. Fill each cell with what you found, in words. Not a score, not a tick, not a rating out of five.
The instruction to use words rather than marks is not fussiness. The moment you convert "states a two to three week lead time" and "no lead time stated anywhere" into 2 and 0, you have built a scoring rubric, and a rubric resolves to a single number that pretends the attributes are commensurable. They are not. A missing lead time and a missing contact route are not two units of the same substance.
Three cell values are legitimate, and the third one matters:
- What you found, stated plainly.
- Not stated — you looked and it is not there.
- Unknown — you have not checked, or could not.
Keep "not stated" and "unknown" distinct. The first is information about the supplier. The second is information about your research. Collapsing them is the most common way this method goes wrong, and it is the same discipline we apply to our own records, where an unknown value is left explicitly empty rather than filled with something plausible.
Step 4: the tie-breaker that isn't price
Frequently the table comes out close. When it does, the instinct is to fall back on price, and price is usually a poor tie-breaker between two small suppliers because the difference is small relative to the cost of the thing going wrong.
Better tie-breakers, in rough order of usefulness:
- Which one answered your question better? Not faster — better. Did they engage with what you actually asked, or send something prepared?
- Which one told you something you did not want to hear? A supplier who says "that timescale is not realistic" or "we are not the right fit for that" is displaying the behaviour you will need most later.
- Which one would be easier to leave? If this goes badly in month four, which situation is easier to unwind? Ask about handover, file formats, access, and who holds what.
- Which one has more slack? Not capacity in a boastful sense, but whether there is any evident margin for something going wrong.
And then, if it is still close: choose on preference. You are allowed to. Once the checkable attributes have genuinely tied, the remaining difference is taste, rapport, or the fact that one of them is twenty minutes down the road. Those are real inputs and pretending otherwise just extends the deliberation without improving the outcome.
Worked example with two anonymous suppliers
Two small suppliers for the same piece of work. Call them Supplier A and Supplier B. Neither is a real business and no real business is being described.
The buyer's stated risks: it arrives late, and there is nobody to fix it afterwards.
| Attribute | Supplier A | Supplier B |
|---|---|---|
| Says plainly what they do | Yes, on the homepage, one sentence | Yes, but you have to open a services page to find it |
| Lead time stated | "Currently four to six weeks" with a note that it is updated monthly | Not stated |
| Process for agreeing the spec | Described in three steps, including a sign-off point | Not stated |
| Contact routes | Email address and phone number, no individual named | Named individual, email address, and a form |
| Pricing basis stated | Not stated; quote on request | Day rate published, with a note that project quotes vary |
| Continuity signals | Dated project notes, most recent within the last few months | Undated portfolio, no other dated content |
| Size of work normally taken | Stated: a range, with the low end named | Not stated |
| Response to the test question | Two working days, answered the question directly, asked one clarifying question back | Same day, answered partially, did not address the second half |
| Ease of leaving | Not stated | Stated: files handed over at completion |
Read the table rather than totalling it. A few things are visible immediately.
Supplier A has written more down. Lead time, process, scope, and dated evidence of recent activity all exist, and the response engaged properly. What is missing is any pricing signal at all, and there is no named person on the contact page.
Supplier B is reachable in more ways, names an individual, publishes a rate, and has stated what happens at the end. What is missing is almost everything about when — no lead time, no process, no dated content — and the reply was quick but incomplete.
Against the buyer's two stated risks, this is not a tie and it is not a clean answer either. On lateness, A is more legible. On being left stranded afterwards, B is more legible. Which of those two risks is worse depends entirely on what the work is for, and only the buyer knows that.
So this example ends here, without a winner. That is not a failure of the method and it is not a coy refusal. There is no supplier that is better in general, only a supplier that is better against a particular set of risks, and the risks are yours. Handing the table back is the honest end of the exercise.
What this method cannot do for you
It cannot tell you whether the work will be good. Every attribute above is about what a supplier has written down and how they responded to one question, and none of that is a sample of their actual output.
It cannot detect a supplier who is competent but unlucky, or one who is reachable and pleasant and produces poor work. Both exist and neither shows up in the table.
It cannot substitute for the basic reality checks. If either supplier is unfamiliar to you, run what to check before you buy first and then come back to this. Comparison assumes both parties are real and trading, and that assumption is worth two minutes of verification.
And it cannot be outsourced to us. We apply the same refusal to ourselves that this post asks you to apply to star ratings: the register orders nothing. Entries are in or out against a published bar, presented in no meaningful sequence, with no score, tier, badge or default sort that implies preference. If two entries here both do what you need, this site has deliberately given you no basis for choosing between them, and this method is what we would suggest instead.
The one thing that inclusion does mean is narrow and worth stating exactly: on a stated date, a named person found the business, and it was recognisably what we described. It is not an assessment of quality, safety or solvency, and it is not a recommendation. The reasons we refuse to publish one are in why we publish no ratings, and the wider list of things this site will not publish is at what we refuse to publish.
In Plain English
When two small suppliers have no ratings between them:
- Write down what would actually go wrong for this specific purchase. Be concrete.
- Turn each risk into something checkable — what has the supplier written down about it?
- Build a table, attributes down and suppliers across, filled with words. Keep "not stated" separate from "unknown". Never convert it to a score.
- Add reachability and responsiveness to every table. Send both the same short specific question and note what comes back.
- Tie-break on the answer quality, on who told you something inconvenient, and on who would be easier to leave — not on price.
- If it is genuinely tied, choose on preference. That is allowed.
A five-star average from a handful of reviews is arithmetically fine and tells you almost nothing, because the people who write reviews are not a sample of the customers. The absence of reviews tells you nothing either.
Definitions for the terms used here are in the glossary.
See why we don't rate anyone → /blog/why-we-dont-rate-anyone