Many CRM workflows depend on images — product photographs attached to inventory records, property images linked to deals, ID documents uploaded during onboarding, or branded assets associated with accounts. Without automated validation, ensuring those images are correct, relevant, and appropriately categorised falls entirely on your team, creating bottlenecks and inconsistencies as record volumes grow.

Zia Vision Image Validation uses machine learning to train on your existing image library and then apply that learning to images attached to CRM records going forward. Rather than manually reviewing each upload, Zia validates images against trained models — confirming whether they match expected products, contain detectable objects, or satisfy both criteria at once — and surfaces confidence-scored results directly in the record.

Three Validation Rule Types

Zia Vision supports three distinct approaches to image validation, each suited to different use cases:

How the Model Trains

Zia Vision's matching capability is based on a model you build from your own image library. The model learns from examples you provide — a minimum of 5 images and a maximum of 300 images are required to train a single rule. The images must represent what a valid match should look like, and Zoho recommends providing varied, representative examples to improve accuracy across real-world conditions.

Once trained, Zia evaluates a confidence score for each validated image. A threshold of 80% accuracy is required for the validation result to register as a match — images that score below this threshold are flagged accordingly, giving your team a clear signal that manual review is needed. Each image field can have one Vision rule attached to it.

Configuring a Vision Rule

Navigate to Setup > Zia > Vision > Image Validation to manage rules.

1. Create a new rule and select the target module and image field. Choose the module where the validation should apply (e.g., Products, Leads, Contacts) and the specific image field on that module. One rule per image field.
2. Select the rule type. Choose Match Only, Detect Only, or Match and Detect based on your validation requirement. Note that Detect-based options are US DC only.
3. Upload training images (Match and Match and Detect only). Provide between 5 and 300 training images in JPG, JPEG, PNG, GIF, BMP, or TIFF format. Upload varied, representative examples to maximise model accuracy.
4. Train the model and review accuracy. Zia processes the training set and reports a confidence accuracy figure. An 80% threshold is required before the rule can be considered production-ready. If accuracy falls short, add more training images or refine the existing set.
5. Activate the rule. Once the accuracy threshold is met, activate the rule. Zia will validate images uploaded to that field on all subsequent records.

Supported Formats and Limits

Model Retraining

As your product catalogue or image library evolves, the trained model may need to be updated to reflect those changes. Zoho recommends retraining the model when your dataset has changed by 40% or more — either through the addition of new image categories or the replacement of a significant portion of the existing training set. Retraining follows the same process as the initial training and requires meeting the 80% accuracy threshold before reactivation.

Availability