Nearshore data annotation, calibrated on your hours.

A nearshore annotation team works from a neighboring country, during your business day, inside the labeling platform you already use. For US companies the two strongest locations are Mexico and Colombia, and the clock is a big part of why.

Mexico’s time-zone law, published in October 2022, ended daylight saving time for most of the country. Mexico City, Guadalajara and Monterrey stay on UTC-6 all year, which matches US Central in winter and Mountain in summer. Colombia stays on UTC-5 all year, matching US Eastern in winter and Central in summer.

That overlap changes how annotation work moves. A question about an ambiguous image or an odd utterance goes to your ML engineer in the morning and the answer is in the guideline by lunch, not the next day.

Sources: Mexico, Ley de los Husos Horarios (Cámara de Diputados) · Colombia legal time, Instituto Nacional de Metrología

Any annotation task can run nearshore, but some get much more out of shared hours and native Spanish. These are the strongest fits:

Spanish is not one dataset. The same word can be neutral in one country and rude in another, and a model trained on labels from someone who doesn’t know the difference learns the wrong thing.

Annotators in Mexico and Colombia read Latin American Spanish as native speakers and work in English with your team. They catch sarcasm, regional slang and the switch from Spanish to English mid-sentence that a translated guideline never mentions. When they find a pattern your guideline doesn’t cover, they can explain it to your team the same day, in English.

Good annotation is a loop, not a handoff. Here’s the sequence we use, with the nearshore advantage at each step being speed of iteration.

Your data should never leave your control. The safest setup has annotators working inside your own annotation platform or a vendor workspace you administer, not on copies of your files. Put these rules in writing before anyone logs in:

Some annotation work means looking at violent, sexual or hateful material, whether for content moderation or for training a safety classifier. People who do that work need protection built into the project, not offered as an extra.

We agree these measures with the partner center before launch: annotators know what the queue contains before they accept it, exposure time per shift is limited, people rotate between sensitive and ordinary queues, tools blur or gray out images by default where the task allows, and confidential counseling is available. Anyone can step off a sensitive queue without penalty.

Many ML teams use more than one model for different parts of the data pipeline. Here’s how the three compare for annotation work.

Nearshore isn’t the answer for every dataset. Offshore usually wins when the task is stable, the guideline is settled and the volume is large: bounding boxes on hundreds of thousands of images, frame-by-frame video tracking, English transcription or moderation queues that run around the clock.

If nobody on your side needs to answer questions while the work is being done, the shared hours you pay for in nearshore matter less, and a team that works while you sleep gives you labeled data every morning.

Many teams split the work: nearshore to design and calibrate the task, and to handle Spanish data, then offshore to run the stable queue at volume.

Shared hours are the starting point. These questions show how a provider will actually run your data.

OTRO runs annotation teams in Mexico and Colombia through partner centers we select and manage, under one contract and one set of reports. Two countries give you two hiring markets with nearly the same hours and a wider range of Spanish. For scope, process and how a project starts, see the data annotation outsourcing page.