Zylker processed food manufacturing company manages their orders using an order management application that is built using Zoho Creator. They supply their products to many buyers in bulk and send out auto-generated sales orders as acknowledgement. All sales orders are maintained in a form along with other essential information like order ID and buyer name.
Although maintaining records of sales orders is an efficient process, doing it manually is monotonous and inefficient. In this scenario, the custom OCR model can be trained to work exclusively on the sales order layout that Zylker issues. Upon upload of Zylker's sales order, layout-specific information like order ID and buyer name can be automatically extracted and maintained in the same form, saving time and manual labor.
7. Custom object detection: Auto-tag damage level
An automobile resale company, Zylker buys and sells used vehicles online. They embedded a Zoho Creator form in their website for the sellers to provide information about the vehicle that needs to be sold including its image. Before a proper inspection is done, a custom object detection AI model can be used to tag and categorize the proposals based on the damage level of the vehicle. For example, a vehicle with too many dents can be tagged as 'Severe damage', vehicles with a few dents can be tagged as 'Moderate damage', and so on.
Setting up an AI model
No set up is required to deploy ready-to-use models in your application. However, for custom models , the following procedure needs to be followed before it can be added to your applications as fields in forms.
In the Microservices section, choose the required custom model type to be implemented in your application. A straightforward builder is offered to create the selected AI model with custom intelligence catering to your business needs.
The next step is to upload training data and train the model. Training data is the initial dataset that must be fed to teach the model to identify patterns so it can make predictions or perform a required task. The principle behind a good AI model is collecting sufficient, relevant, and high-quality data to train it. Keep in mind one of the oldest and most used phrases in data science: "garbage in, garbage out". Low quality training data will compromise the accuracy of the model's outcome. Some factors that can ensure the quality of training data are:
- Relevance - Narrow down the dataset to include only values that are required.
- Validation - Make sure that the values you input your dataset are correct.
- Consistency - Values in your dataset must consistently follow the same format.
- Comprehensiveness - The dataset must be large enough and have the proper scope and range to encompass all of the model’s desired use cases.
- Data cleaning - Remove or modify data that is incorrect, incomplete, or duplicated. Ensure that no missing values are present in your dataset. You can substitute the missing values with the most frequent values and in the case of numerical values, mean figures.
Once the dataset is ready, upload it and train the model. Training takes a while based on factors like model size, queued models, etc., however you can navigate away and come back later once the model is trained.
After successful training, the model is practically ready to be deployed. However, it is important to evaluate the model's performance and quality. You can test the model by giving sample inputs; if the test result is not satisfactory, it could be because your model is either underfitting or overfitting.
Your model is underfitting if it performs poorly on the training data. This is because the model is unable to understand the relationship between the input and output variables. Your model is overfitting if it performs well on the training data but produces incorrect outputs on the test data. This is because the model has memorized the data it had seen and is unable to generalize to unseen examples.
In other words, underfitting models learn the variations in the training data rather than the patterns, whereas the overfitting models learn the training data by heart that it wouldn't perform well on any other inputs.
Both overfitting and underfitting are problems that eventually cause poor results on new data and the key to fixing this is to regularize the training data, i.e., the training data should neither have too much variations nor have values that are too close to the expected outcome.
See how you can improve the accuracy of each custom model type:
4. Publish model
The next and final step to set up an AI model is to publish it. Until this point, the model is accessible only by the admins. If you are satisfied with your model, make it available for all users by publishing it. This means the model is ready to be implemented in your Zoho Creator applications.
Managing an AI model
A custom AI model , once configured, can always be edited, renamed, retrained, or deleted. A good AI model must be built, fed with quality data, and deployed. However, it doesn't end there. Retaining the quality of an AI model is a continuous process. Over time, as the business environment and data change, it is inevitable that AI models experience model drift . In other words, the accuracy of model will begin to degrade. For example, a company's sales before covid and after covid are different. An AI model that is built to forecast sales must be trained considering the covid factor too.
For this reason, it's crucial to test and retrain the AI models with updated new data periodically or whenever an evident change in trend occurs. Options to edit, rename, retrain , and delete the model are available in the model details page. This page also shows other details like versions and applications list where the model is deployed.
Using an AI Model
To implement an AI model in your Zoho Creator applications, use one of the following methods:
- Use the required model from the AI modeler page. You will be redirected to the application builder where the model needs to be implemented.
- Add the required AI fields to the form. When you pick a custom model, all published models will be listed while adding the AI field.
Note: For the applications that have environments enabled, you can use custom AI model only if the application has at least one version published in the production environment.
AI calls are calculated for your account based on the API calls made while accessing the AI services in Creator.
1 AI Call = 1 API call made i.e., every time an AI model is deployed in your Creator application, an AI call is made and the required output is fetched.
Users in Ultimate/Professional
plan can use upto 1000 AI calls
per month. The remaining no. of requests per account can be monitored anytime from the Billing
Points to note
- As the admin, you can create and use custom AI models, whereas your users can only consume the models you've created. Any user with permission to edit your applications can indirectly utilize AI models by adding AI fields to forms.
- As a low-code platform, Creator doesn't require you and your users to have prior coding and machine-learning skills to create, utilize, or access AI models.
- You should know enough about all your business requirements to determine and gather data that'll be used for training your custom AI models.
- It is preferred that you do an exploratory data analysis (EDA) while collecting training data for custom AI models.
- AI Modeler is available for Professional or Ultimate plans. Refer to our pricing page .
- You must be using Zoho Creator 6 (C6) to be able to create custom AI models from the Microservices section whereas, deploying the AI models as fields are supported in both C6 and C5.
- View guidelines for each model type:
Get started and build your custom AI models using the sample data we've provided in the attachment.