Agent or do it yourself

Separating them is most of the value of this page, because they have different costs, different learning curves and different failure modes.

Finding. Cheap, fast and easy to do yourself: directories, markets, shows. The output is a list of names, which is the least valuable part of the chain and the part people spend the most time on.

Verifying. Expensive to do properly from abroad and the single largest source of loss when it is skipped: is this a maker or a reseller, does the invoicing entity match, is the site real. This is the job an agent is genuinely selling — everything else is logistics.

Negotiating and specifying. Doable yourself with discipline, and it is where the money is: a specification that names what may not be changed without approval is worth more than a price concession. It also needs somebody who can read the answer, which is a language and a domain problem at once.

Quality control. Almost universally outsourced already, to a third-party inspector — which is worth noticing, because it means the argument is not really about outsourcing in principle.

Export and logistics. Likewise: a forwarder is an intermediary, and nobody calls using one "not doing it yourself".

You are close to the product. Nobody understands a specification like the person who designed it, and every layer between you and the factory is a layer where a nuance can be dropped.

You keep the relationship. A supplier relationship held by an intermediary is a relationship you do not own, and that matters the day you change intermediary or the day the intermediary changes its priorities.

You learn. Sourcing knowledge is cumulative and it compounds; the second product is much cheaper to source than the first, and buying that learning as a service means you never accumulate it.

And the cost is genuinely lower once your volume justifies the time. No percentage is published here for that crossover, and it is not modesty: it depends on your margin, your product complexity and what your own hours are worth — and a figure would be somebody else's business, not yours.

The honest cost of DIY, stated as plainly: your time, in blocks, unpredictably; a slower first cycle; and a class of mistake that is only visible after it has been made. If you are doing it yourself, spend the saved fee on the verification step — that is where the losses live.

What it actually buys is presence: somebody in the same time zone as the factory who can go and look, who has seen this failure before, and who carries part of the risk. For a buyer without the volume to justify travelling, that is frequently the difference between a supply chain and a series of experiments.

And it converts a variable into a fixed cost, which is worth more than it sounds when you are small: a known fee against an unknown number of hours and an unknown probability of a bad batch.

When it is the wrong answer. When your product is genuinely simple and you have time — the agent is then a tax on a job you could do. When you need deep technical iteration with the factory, because every layer costs a nuance. When your volume is large enough to justify your own person in China, at which point you are buying the capability rather than renting it. And when the intermediary is only a mark-up — has not visited, has not checked, carries no risk. That is a directory with a commission attached, and it is worse than a directory because it feels like verification.

So interrogate an intermediary exactly as you would a factory. What did you check on this supplier, when, and can I see it? Who is the contracting party, and who is the exporter of record? What happens, specifically, if a batch is bad? Vague answers to those three are the whole signal.

Split it. The all-or-nothing framing is imported from the way agents sell, not from the way the work divides.

Do the finding yourself. It is cheap, it teaches you your category, and nobody can do it in your place because only you know what you are trying to build.

Buy verification for the shortlist. It is the highest-value paid step and the one you cannot do from abroad — and paying for three verifications is far cheaper than one wrong supplier.

Do the specifying yourself, in writing, always. Whoever writes the specification controls the product. Never delegate that entirely, however good the intermediary.

Buy inspection. You already do; make it deliberate, and move it earlier in the production window than feels comfortable.

Buy freight. Also already true.

And review the split every year. What you should hold changes as your volume grows and your knowledge accumulates — which is the strongest argument for doing some of it yourself from the beginning, even when buying all of it would be easier.