AI Training
Last updated
Last updated
AI training bridges the understanding between your data, your team, and the AI by by specifying terminology in your data tables and columns, explaining the output you are seeking when asking a question, and defining terms to clarify any ambiguity.
To access it, toggle to the data tab. On the top right, click on "AI Setup" button which will bring you to the AI training page. From there, click "Add Example" to add training pair.
The AI can be trained using plain English or SQL. Toggle between the tabs to apply either feature.
** Note: To enhance the AI and provide accurate training pairs, refer to the tables and column names to define them as much as possible.
Natural Language Training (Plain English) Guide the AI by defining ambiguous or specific terms and questions with plain English through question/term-definition pairs.
Establish Key Terms: Add any corporate jargon or ambiguous KPIs that may be asked in your project. You can also input questions or commands in the same way you would when speaking to a data analyst.
Define Key Terms: Explain by describing how you want the AI to execute the command or question by connecting with the tables and column names in your data. Make sure you are as clear as possible and indicate criteria for ambiguous terms. This includes defining what you mean by "top", "best", "target", etc.
Examples:
"Top Customers" can be defined by highest value for "Closed Won" column since that is the metric we care the most about for our customers
"Best Sales Season" can be defined by "Highest Profit. Profit = revenue - manufacturing cost"
"What is the Best Personal Care Product?" can be defined by "Personal care products include lotion, sunscreen, shampoo, conditioner, and body wash. Best means highest quantity sold"
SQL Training Guide the AI by defining ambiguous or specific terms and questions with SQL training by writing specific commands and formula with SQL programming.
Establish Questions: For SQL training, instead of defining a key term, you will choose a difficult but representative question for your data.
Define Results: Write the SQL output you would expect the AI to deliver for the question, and run it to make sure it is correct.
Examples
If you're training the AI to recognize "target 1s," your SQL query might identify these as opportunities with a probability greater than 0.7 and with "DO_NOT_CALL" set to false.
