Manual audit and review

For Premium Classify users, Zonos will manually audit, review, and correct a subset of your Zonos Classify requests in an effort to constantly and consistently improve the accuracy of your classification results.

Audit and review process 

Classify is built on a multimodal AI system that is powered by billions of training records and advanced algorithms. This system allows Classify to immediately provide accurate and reliable results for the majority of customers.

While Classify provides sufficient, reasonable results from the get-go for all users, we understand that some users may require even higher accuracy rates due to low-risk business models or low-quality input data.

Here are some examples of overly-detailed or overly-broad descriptions that would need manual effort:

• Unique marketing terms and/or too much detail: "Medieval Socketed spearhead 350mm. - Bronze, iron bronze - 14th century" or "Japanese Made Seiko NH35 Automatic Movement - 24 Jewels, 21,600 bph"
• Vague descriptions: "flat case" or "crafts" or "launcher"

Both overly detailed and excessively brief descriptions can be improved with a manual review. This process, ideally coupled with Premium support to collaborate and help us better understand your data. This is especially crucial when dealing with unique, complex, or concise data sets, where deciphering product descriptions can be challenging.

For these reasons and others, we offer this Premium service that includes a monthly manual audit and review process from our in-house classification experts. This process allows us to continuously refine and improve results by manually classifying a subset of data and ingesting this known, labeled data to train our AI model.

We start our manual review process by selecting a random subset of classification requests each month, as well as a subset of classifications with low confidence scores. This helps us identify areas where our AI system may be struggling and improve our accuracy rates for future requests. Our Classify team reviews this subset of classifications and manually corrects them if necessary. The corrected classifications are then incorporated into our AI model to retrain and improve future results. Through this process, we continually enhance our accuracy and deliver the most precise classifications to our premium users.