FAQs: Predict Picklist Field Values Using Field Predictions | Zoho Desk

FAQs: Predict Picklist Field Values Using Field Predictions

What is Zia's Field Prediction?
Support agents handle a wide range of activities as part of their routine, and when they receive high volumes of requests from various channels, ensuring every detail is accurately entered during ticket creation becomes challenging. However, these details are critical—they determine ticket segmentation, severity, service cost, and SLA implementation.

Zia's Field Prediction feature addresses this by analyzing historical ticket data, identifying patterns, and predicting the correct values for picklist fields such as ticket category, priority, issue type, and even the ticket owner. When Zia's prediction meets the configured accuracy score, it can automatically update the predicted value in the respective field. This reduces the manual effort required from agents, ensures consistency in how tickets are categorized, and accelerates the time it takes to route tickets to the right team for resolution.
What types of fields can Zia predict?
Zia can predict picklist field values only—both system picklist fields and custom picklist fields. In addition to picklist fields, Zia can also predict the ticket owner. The prediction is based on Zia's analysis of existing tickets that contain similar information.

An important prerequisite is that the picklist fields must be added to the default layout, because prediction can only be executed in the default layout.
What is the minimum number of tickets required for Zia to train on field predictions?
A department must have at least 500 tickets for Zia to begin training. Additionally, it is recommended to have at least 500 tickets for each picklist value that you want Zia to predict effectively. This means that if you want Zia to predict across five different picklist values, ideally each value should have 500 tickets associated with it in the training data.
Does Zia support field prediction in multiple languages?
Yes. Zia supports field prediction in all available languages. This means organizations operating in multilingual environments can leverage field predictions regardless of the language used in ticket conversations.
How does Zia train itself to predict field values?
Once field prediction is enabled, Zia begins training using the existing tickets in the Desk account. During configuration, you select the field(s) and the picklist values you want Zia to predict. Zia learns and creates a pattern using approximately 80% of the total tickets and then evaluates its prediction abilities on the remaining 20%. Based on the results, it calculates a probability and estimates an average accuracy score. The higher the score, the more accurate the prediction will be.

For example, if Zia correctly predicted "Sam" as the ticket owner for 484 records, it might give a probability of 97%, and the average accuracy score would be estimated as 100%.
What is the difference between training Zia using specific tickets versus all tickets?
The choice depends on the type of prediction you want Zia to make:
  1. Training using specific tickets – Recommended when you want explicit, focused predictions. For example, if you want Zia to classify issues as "bug fix," "feature request," "incident," or "data loss," training on specific tickets that fall into these categories allows Zia to analyze relevant data and interpret values based on a clear pattern.
  2. Training using all tickets – Recommended when you want Zia to predict something broader, such as the ticket owner. Using all tickets provides a wider scope of data for Zia to learn, analyze, interpret, and predict from, which is beneficial when the prediction depends on a wide variety of ticket characteristics.

How does Zia determine and use the accuracy score?
Based on its training and evaluation, Zia calculates a probability and auto-generates an average accuracy score. The higher the accuracy score, the more precise the prediction result. Administrators set a minimum accuracy score threshold during configuration, and Zia will only auto-update the predicted value if it meets or exceeds that threshold.

For example, if you set the accuracy score to 70% and a prediction comes back at only 65% accuracy, Zia will not auto-update the value. The article recommends setting the accuracy score to 70 and above for reliable results.
What are the best practices for achieving a better accuracy score?
The article provides several recommendations:
  1. Set criteria to train Zia only on specific tickets when you want explicit predictions—for example, predicting the problem type only for tickets from the IT sector.
  2. Give a wide range of training data to enhance Zia's learning ability—for example, predicting the product type benefits from diverse training data.
  3. Retrain Zia using the latest tickets to keep predictions current and append new patterns.
  4. Set the criteria as created time to train on specific data—for example, predicting issue type based on tickets received between January and June.

What are the two modes of updating predicted field values, and how do they differ?
Administrators can choose between two field update modes:
  1. Auto-update predicted value – Zia automatically fills in the predicted value without agent intervention. For example, if Zia predicts the "Issue Type" as "Bug Fix," it can also identify the corresponding "Support Tier" as L2 and auto-update both fields. This is ideal for high-volume environments where speed and automation are priorities.
  2. Let me confirm predicted value manually – The agent reviews the ticket and verifies whether the predicted value is correct before it is applied. For example, if Zia suggests "Support Tier is L2" for a "Bug Fix" issue but the agent believes it is more complicated and requires technical assistance, the agent can manually change it to L3. This gives agents an edge in decision-making for special or edge cases.
In both modes, Zia will only update the value if the prediction meets the configured accuracy score threshold.
When can field predictions be executed—during ticket creation or customer reply?
Administrators can define when Zia should execute predictions based on business requirements:
  1. When the ticket is created – Useful for automating ticket profiling at the point of creation. For example, predicting "Issue Type," "Service Type," and "Ticket Owner" so the ticket can be immediately routed to the right agent.
  2. When a customer response is received on a ticket – Useful when a customer replies to an existing ticket and you want to predict the "problem" to reassign it to the appropriate agent.
  3. Both – You can configure prediction to execute on both ticket creation and customer reply.
How does field prediction integrate with workflow rules for ticket assignment?
When Zia predicts a value and it meets the accuracy score, the predicted value is updated in the respective field. If a workflow rule is configured for that field, the update automatically triggers the workflow, which can assign the ticket to the respective team or individual. This saves time and rules out inappropriate handling of tickets, ensuring timely action and quick closure.

An important condition: the workflow will be triggered only if the criteria is set as Create and/or Customer reply, matching the execution timing of the field prediction.
What happens if a new ticket doesn't match any of the selected picklist values?
When you configure field prediction, you select the specific picklist values that Zia should predict. If a new ticket's content does not match any of those selected values, the field will remain un-predicted. For example, if you choose "Bug fix," "Code error," and "Data Loss" as the values Zia should predict, and a new ticket is about a "New Feature" request, Zia will not suggest a value for that ticket since "New Feature" was not selected.
Does Zia automatically learn from new tickets received after configuration?
No. Zia predicts based on the tickets that are available at the time of configuration. Tickets received after this period must be manually added to the prediction in order to retrain Zia. This is why the article recommends periodically retraining Zia using the latest tickets to keep predictions current and accurate as your business evolves.
Can I control which fields Zia uses for training?
Yes. Businesses can enable focused training by selecting the specific fields that Zia uses for learning and training purposes. For example, if you want Zia to predict only tickets that require service from third-party vendors, you can set training criteria such as:
  1. "Issue" contains "Replacement," "Technical issue," or "Faulty product"
  2. "Service charges" is "Yes"
  3. "Ticket category" contains "Service needed"
  4. "Service cost" is "Greater than or equal to $10"
This allows Zia to train on a targeted subset of tickets that are relevant to the prediction you want to make, rather than analyzing all tickets indiscriminately.
How do I configure field predictions?
To configure field predictions:
  1. Go to Setup > Zia > Field Predictions.
  2. Click Create New Field Prediction.
  3. In the Prediction Configuration section:
    1. Select a field from the Field to predict drop-down
    2. Based on the field selected, select a value in the Values for Zia to predict field.
    3. Select the criteria to be used for training purposes.
  4. In the Inference Configuration section:
    1. Select Ticket creation, Customer reply, or both from the Execute on drop-down.
  5. Configure the field update mode (auto-update or manual confirmation).
  6. Set the accuracy score threshold.
  7. Select the fields to be used for training, if applicable.
Info