Four business use cases you cannot miss with Vertex AI Action in GenAI x Looker
In my previous blog, I already listed out all steps needed to install Vertex AI Action on your Looker instance. After setting it all up, now you should know how to utilize this feature in Looker to gain. With the era of AI-everything, this action in Looker can provide you with the power of generative AI into data analytics.
How to use the action?
- Step 1: Choose necessary measures and dimensions in an explore as you normally do when exploring data in Looker
- Step 2: Click on the setting button next to “Run” and choose “Send” then click “Vertex AI”
- Step 3: Put necessary inputs or prompts into the form. You can even customize the AI model by choosing “No” for “Default Parameters”. After that, you can choose between 2 different model types, which are Code Bison and Text Bison. Depending on the model type you choose, you can customize model’s temperature, max output tokens, top-k and top-p.
- Step 4: Click “Send” and Vertex AI will process your data, apply algorithms and generate prediction or whatever your model is designed to do
- Step 5: Report will be sent to your email.
Whom is this for and in which scenarios?
This action is helpful for a wide range of business users with different background in data literacy. You can extract insights from data without any prior experience with SQL or data analytics because you can just use natural language to do this. The rest of the job will be in Looker’s hand. Or even when analyzing data is your daily task, Vertex AI Action can be a handy tool for you to enrich data and automate task, which saves you a lot of time and effort.
- Data summarization: When there is a lot of data and you are not sure how to interpret it, this tool can be helpful to summarize and highlight important insights. This is very helpful for business users who are not experienced with data. An example use case is to summarize inventory warehouse status as below. You can achieve this by just a simple prompt such as “Can you summarize this dataset in 10 bullet points?”
- Personalized content generation: You can generate emails and messages based on the data returned. For example, you work for customer service department, and you have a list of delayed orders and want to send emails to these customers to inform of this and apologize for the delay. This tool can generate personalized emails with customer’s name and order ID for each order. You can adjust temperature parameter for more friendly and human-like responses so that your customers don’t know you use AI to write them emails. Here is an example of email generated by Vertex AI for the situation above:
Dear Joseph Benson,
We are writing to apologize for the delay in processing your order #380608. We understand that this is frustrating and we are working hard to improve our processing times.
Your order is currently in the processing stage and we expect it to be shipped out within the next 24 hours. We will send you a tracking number as soon as it is available.
We appreciate your patience and understanding.
Sincerely,
The Customer Service Team
- Sentiment analysis: You are a product owner or customer service team and want to know what customer’s feedbacks about your product are. But you are too busy with the development process and don’t have time to go through each of the feedbacks. This is where Vertex AI Action comes in. You can throw a lot of texts at it and it can tell you if they have positive and negative tone.
- General prediction and forecasting: Vertex AI Action use PaLM natural language model and it’s not an actual prediction or forecasting model. However, it still can tell you general prediction based on some metrics that it thinks is crucial to the prediction. For example, you are an inventory manager and want to know when a certain item needs refilling. This tool can tell you some items that should be refilled soon and the reason for its prediction. You can also schedule this to run weekly so you can keep up to date with your warehouse situation.
Sone nuances
In order for the action to perform its best, there are some criteria you need to pay attention to. First, you need to provide only enough and related data for your prompt. If you provide too much data, the model might fail to run because there is a limit on the number of input tokens. If you don’t providing enough data, the prompt might make up some assumptions, which can lead to false answer. Another aspect that you want to have an eye on is your LookML readiness. Because Vertex AI Action is based on a PaLM language model, the more detailed text it can read from your input data, the better the result is. Your LookML project should have clear descriptions and tags for each dimension and measure. Also, naming conventions should be consistent thorughout your project. Therefore, ensuring LookML quality is a priority before you integrate AI into your data analytics in Looker.
My take on Vertex AI Action
Vertex AI Action is a helpful tool for users in your organization with different background in data knowledge, varying from non-data to experienced people. It can save you time and effort in tasks such as data summarization, personalized text generation and sentiment analysis. It also performs general prediction as well. This tool is a new way of enriching data analytics with AI. Since this is an open-source feature in Looker, you can also replace PaLM model with your model of choice such as Gemini to fit your organization’s use case.
However, there are still some drawbacks of this tool. It might be difficult to deploy on your Looker instance because of its open-source nature. It requires some technical knowledge to do so, which is not very user-friendly compared to one-click installation. For a small data team, they might not have enough resource to implement this tool or they should seek help from data service companies. Also, this tool sometimes can give some inconsistent answers. so I don’t recommend using this as an only tool for you to make a decision. You should also consult with other data sources as well before doing so. And the tool is still in continuing development, it might be glitchy sometimes such as not able to schedule the report. So you should keep in mind of this when deploying the action in your Looker instance.
If Vertex AI Action interests you or you are not sure if your LookML project is ready for this, please leave your contact here and we will reach out to you. We also offer other GenAI x Looker packages, including Explore Assistant and Dashboard Summary, on which we also have some blogs in our Medium. To know more about who we are and what we do, please visit our website here.
