Importing smart watches from China: what to check

A watch with a low-cost optical sensor can read the pulse within a reasonable deviation at rest and drift noticeably when moving — exactly when the user checks it. The minimum order in this category starts where the minimum for accessories ends: between 500 and 1,000 units per reference, with 25 to 45 days of lead time if there is customisation. Before committing that volume, three things no photograph shows must be settled — the real accuracy of the sensor, the wording of the health functions and who maintains the app — because each turns into returns or a regulatory problem once the product is sold.

Two watches that look identical on the outside are not the same product

Two watches built on the same case and the same screen can measure the pulse with very different results, because the measurement is decided by the sensor plus the algorithm, and neither appears in a photograph. The sensor is quoted; the algorithm that filters the signal and discards motion noise is not, and that is what separates one product from another.

Photoplethysmography — PPG, the technique used by practically every wrist wearable — measures changes in light absorption in the tissue to estimate blood volume per beat. It is a weak signal. At rest, with the wrist still and the sensor well seated, almost any module delivers a usable number. When walking or moving the arm, the sensor shifts over the skin, ambient light enters and the signal is contaminated; there the filtering algorithm and the factory calibration decide whether the result is usable or an invention.

That pattern — acceptable at rest, bad in motion — is the worst possible commercially, because the user discovers the defect in the first week of real use, not in the box, and it is the hardest to detect in a visual inspection, which only checks that the screen turns on and the interface responds.

How a pulse sensor is verified before buying the lot

Verification is done by comparing, not by reading specifications: a sample is taken from the lot and checked against a validated reference device, with the subject at rest and in motion, recording the deviation in each condition. No universal acceptable error figure exists — it depends on the intended use and on the reference chosen — but the acceptance criterion is agreed in writing before the test. The minimum procedure has five steps:

1. Random samples from the lot you are going to ship, not the unit the supplier picks for the demo.
2. Subject seated at rest, five minutes, comparing the watch reading against the reference at the same time.
3. Subject walking on a treadmill or outdoors, the stretch that matters and the one almost nobody tests.
4. Recording the deviation under each condition, separately. An aggregate number hides the datum you need.
5. Two or three units from the lot, because the spread between units says more than the result of one.

A device that is accurate at rest and drifts when moving gives a bad experience, because movement is the normal condition of use. It is also worth comparing step cadence against a manual count over a measured stretch: it is the function the user checks daily and the one that gets inflated most.

The regulatory trap: the sentence that changes the product’s category

The moment you describe the product as capable of measuring blood pressure, detecting arrhythmias or diagnosing anything, it stops being a consumer accessory and becomes a medical device. That is not a marketing question: it brings health registration prior to commercialisation, quality requirements on the process, and legal liability if the device gives a wrong reading and someone acts on it.

The practical distinction is in the verb. “Activity tracking”, “indicative heart-rate estimation” and “wellness logging” describe wellness functions; “measures”, “detects” and “diagnoses” describe clinical functions. It is the same electronics and two different legal products. The sentence on the product listing, on the box and in the manual is what fixes the category, which is why it should be drafted before ordering the packaging, not after.

If your product genuinely needs to be a medical device, there is a legitimate path, but it is longer and more expensive. The decision is taken at the start of the project, not when the goods are at the port.

SpO2 and blood pressure: why those two claims are the most expensive

Many low-cost watches declare that they measure blood oxygen (SpO2) and blood pressure with no clinical validation behind them, and selling them with that claim in Mexico, Argentina or Brazil is a real problem, not a theoretical one. Neither function is resolved by the optical sensor that estimates the pulse: blood pressure by optical method requires individual calibration against a cuff and an algorithm validated against a recognised protocol, and SpO2 without validation only makes sense as a wellness trend.

The pattern in the market is well known: the supplier’s listing declares the function, the box prints it, and the importer repeats it on the marketplace. At that moment the importer becomes the one asserting a diagnostic capability, and the holder of the liability is the importer, not the factory.

The recommendation is consistent and has no nuances: describe wellness functions and leave out diagnostic claims, including blood pressure even if the supplier insists that their firmware “already calculates it”. A watch that declares blood pressure and does not measure it with clinical validity is not a product with an extra function: it is a product with a problem.

Battery life: the number on the box and the method for measuring it

Declared and real battery life differ so much in wearables because they depend on three variables the manufacturer does not publish: screen brightness, measurement frequency and whether the app keeps a constant connection to the phone. A watch that declares fourteen days can last five in normal use, and both figures can be honest if they describe different conditions.

There is no single standard that fixes how the battery life of a wearable is measured, so the manufacturer defines it and almost never publishes the test condition. That is why what you ask for is not a number, it is the method. The minimum protocol:

• Fixed screen brightness, because it is the variable that consumes the most and the easiest to lower in order to stretch the figure.
• Declared measurement frequency, continuous or at intervals: with the pulse sensor on continuous, battery life falls markedly.
• Defined connection to the app: active notifications and live sync consume radio steadily.
• Explicit cut-off criterion, and notice of whether the figure includes low-power mode.
• Three units from the lot, to see the spread between watches of the same order.

Battery life in low-power mode is not a useful datum for the end buyer, because it disables the functions they bought the watch for. Ask for the figure with the functions on.

The app: the risk you do not see in the sample

Many Shenzhen wearables depend on a third-party app shared between several factories, and if that app stops being maintained the user is left with a device that no longer pairs. It is not a hardware failure and it is not visible in inspection: the watch turns on, the screen works, and the problem appears months later, when the phone’s system updates and the app does not.

The pattern is easy to recognise. The same case model can be sold with two different apps depending on the chipset the factory managed to get that month, and an app used by twenty factories has nobody with an incentive to keep it alive. Three verifiable questions before committing the order:

1. Who develops the app and under which account. Not “we have it in-house”: the name of the developer that appears on the store listing.
2. Whether it is published in both stores — Google Play and App Store — with recent updates.
3. Whether you can publish your own build under your brand, with a written commitment that the factory will deliver the binary and will not change platform without telling you.

Without a written answer to the third question, your product depends on the continuity of a third party that has no contract with you. The app is what ages most in a wearable, more than the battery.

Mexico, Argentina and Brazil: what each destination adds

On top of the common documents — Bluetooth SIG for the radio, UN38.3 and the MSDS for the battery — each market adds its own layer, and the procedure is settled before shipping because the tests are performed on the finished product. The full matrix is in certifications; this is what applies to a watch.

The declaration before Bluetooth SIG is mandatory in all three markets and is made by the manufacturer under its own account, tied to a specific product identifier. It is checked in the public qualified-products database before paying the deposit, and it has to match the exact model of the box you are going to import. If you sell under your own brand and your own model, the manufacturer’s listing does not cover you on its own.

In Mexico, CRT homologation applies to every device that transmits by radio, and on top of that there is the NOM on commercial information, which requires labelling and instructions in Spanish. In Brazil, ANATEL and Inmetro are separate bodies with separate processes, and a watch with a battery may need both.

MOQ: why this category starts higher than accessories

In accessories the minimum starts at 100 units; in wearables it sits between 500 and 1,000 per reference, with 25 to 45 days of lead time if there is customisation. It is not an arbitrary commercial decision: the case carries tooling, the screen and the module are bought in matched lots, and the firmware with your identification and the printed packaging are set-up jobs that only pay for themselves above a certain volume.

If you accept the manufacturer’s brand and packaging, the minimum falls and the lead time shortens. What changes is not only the price: the product exists under another name, with no exclusivity, and you control neither the firmware nor the continuity of the model. For a first market test that is usually the right call; for building a brand, it is not. A reasonable middle route is a mixed order of references with customisation limited to the packaging, the cheapest line item and the one that moves perceived price the most.

What to ask for before paying the deposit

• Brand, model and lot of the optical sensor, with a commitment not to substitute it without telling you.
• The UN38.3 report naming the cell and the pack of the exact model you are going to receive, together with the MSDS. The report is for the model, not for the brand.
• The product identifier declared before Bluetooth SIG, verified in the public database, with the commercial name and the model of the box.
• The method for measuring battery life in writing, with brightness, measurement frequency and cut-off criterion.
• The name of the app’s developer, the two stores where it is published and whether you can publish your own build.
• The exact wording of the health functions that will go on the box and on the product listing, with no diagnostic claims.

The sample is validated with two or three units, not one: the spread between units of the same lot is the datum that says whether the order will repeat. MeliPrep receives the goods at its Shenzhen warehouse, verifies them and coordinates inspection before consolidating.