How to Teach AskTable Your Business Knowledge: A Semantic Configuration Guide for Operations, Sales, and Marketing Teams

The screenshot uses synthetic demo data. The important point is not the specific numbers. It shows that AskTable can explain the calculation method first, then return a table and key observations based on the business definitions you configured.

After connecting data to AskTable, many teams ask: if the data is already connected, why does AI still need configuration?

The reason is simple: data systems usually store field names and numbers, but the real business meaning often lives in people’s heads.

For example:

• •Does account_name mean an ad account, a customer account, or a store account?
• •Does an “effective customer” include every lead, or should duplicate applications be excluded?
• •Does a “high-quality customer” include three-star and four-star customers, or only four-star customers?
• •Should cost be calculated by leads, effective customers, or another denominator?
• •Should the answer start with the conclusion or with the detailed process?

AskTable needs you to tell it these things. The clearer you are, the more it can analyze according to your company’s business habits instead of guessing from field names.

This guide explains how non-technical users can configure semantic information so AskTable understands the business better.

The Short Version

Think of AskTable semantic configuration as five kinds of business instructions:

1. •Table comments: explain what a whole dataset represents.
2. •Field comments: explain what each column means.
3. •Field value index: help AI recognize common names, statuses, and categories in a field.
4. •Business documents: store complex business rules, metric definitions, and calculation logic.
5. •Preferences: tell the data agent what background knowledge and answer style it should remember long term.

They are not only for technical users. They are for people who understand the business.

How to Choose the Right Configuration

Simple rule of thumb:

• •Explain a whole dataset in a table comment.
• •Explain one column in a field comment.
• •If users often mention a store, channel, account, or level by name, configure a field value index for that field.
• •If the rule is long, crosses datasets, or involves formulas, write a business document.
• •If you want the agent to answer in a stable style every time, use preferences.

In datasource field settings, you can add table comments and field comments. For suitable fields such as account name, channel, and customer level, you can enable field value indexing so AskTable can better recognize business names and categories.

1. Table Comments: Explain What the Dataset Is For

A table comment answers the question: what does this dataset represent?

For business users, a “table” can be understood as one block of business data. It may come from Excel, ERP, CRM, an ad platform, or an order system.

A good table comment should answer four questions:

• •What business does this dataset record?
• •What does one row represent?
• •What questions is it mainly used to analyze?
• •What is clearly not included?

Example:

Daily ad account spending

Records the daily advertising performance of each account, including spend, impressions, clicks, click-through rate, and appointments. Each row represents one account on one day. This dataset includes account performance only; it does not include customer lead levels or duplicate-application status.

This helps AskTable avoid misunderstanding the data.

If a user asks “show yesterday’s top self-operated accounts by effective cost,” AskTable can know that spend is in this dataset, while “effective customers” may need to come from another lead dataset.

2. Field Comments: Explain Each Column in Business Language

Field comments explain what a column really means.

Many field names are friendly to systems but not clear enough for analysis, such as:

• •cost
• •gmv
• •valid_cnt
• •account_name
• •status
• •source

Write field comments in business language.

Clear field comments help AI choose the right fields and explain results.

They are especially useful for:

• •Business meaning;
• •Units for amounts, counts, and percentages;
• •Status value definitions;
• •Whether a field can be used to join another dataset;
• •Whether null means unknown, not happened, or not applicable.

3. Field Value Index: Help AI Recognize Business Names

A field value index lets AskTable remember common business values in a field.

Users may ask:

• •“How is the Shenzhen MixC store doing recently?”
• •“Show the spend from Tencent Ads.”
• •“What is the effective cost for three-star customers?”
• •“Which self-operated accounts were in the TOP 10 yesterday?”

These questions include concrete names: stores, channels, customer levels, and accounts. AskTable needs to know how those names appear in the data.

Good fields for value indexing include:

• •Store name;
• •City or region;
• •Channel or media;
• •Product category;
• •Customer level;
• •Order status;
• •Advertising account name;
• •Salesperson or team name.

Fields that usually should not be indexed:

• •Amounts, spend, sales, clicks;
• •Dates and times;
• •Phone numbers, order IDs, ID card numbers;
• •Unique IDs that are almost different in every row;
• •Long comments or descriptions.

The reason is simple: value indexing is for a limited set of business names. If a field is numeric, date-like, or nearly unique per row, it is not a good fit.

4. Business Documents: Write Your Company Rules for AI

Business documents are for knowledge that cannot be explained in one or two sentences.

For example:

• •What is an effective customer?
• •What is a high-quality customer?
• •How should cost be calculated?
• •How should multiple datasets be joined?
• •Which data should be excluded?
• •What does an industry-specific term mean?

A business document usually has four parts:

A business document is the right place for complete metric definitions, keywords, and calculation rules. When users ask about “effective cost” or “high-quality customers,” AskTable can use these instructions to understand the question.

5. Preferences: Make the Agent Work in Your Style

Preferences are configured on the data agent. They usually include two parts:

Domain Knowledge

Domain knowledge is stable business background the agent should remember.

Example:

This agent mainly serves the advertising operations team. It focuses on account spend, effective customers, high-quality customers, appointments, media channels, and account cost. When analyzing, distinguish self-operated accounts from partner accounts.

Analysis Preference

Analysis preference explains how you want AI to answer.

Example:

For cost-related questions, first explain the calculation definition, then provide a TOP table. Keep the conclusion concise. If data may be incomplete, clearly mention the risk.

Preferences are best for short and stable habits. Long business definitions should be maintained in business documents.

A Complete Example: “Self-Operated Account Effective Cost”

Suppose an operations lead asks:

List yesterday’s top self-operated accounts by effective cost and show them in a table.

To make AskTable answer reliably, configure the semantics in this order:

1. •Add table comments for the daily account spending dataset and the customer lead dataset.
2. •Add field comments for account_name, spend, media, customer_level, and duplicate_status.
3. •Enable value indexing for fields users mention directly, such as account_name, media, and customer_level.
4. •Create a business document named “Customer Leads and Advertising Cost Definitions.”
5. •Add preferences that tell the agent to explain the formula first, then return a TOP table and risk notes.

With these settings, AskTable can understand that:

• •“Yesterday” refers to a date filter;
• •“Self-operated account” is a business scope;
• •“Effective cost” is not a raw field, but a formula;
• •Spend comes from the account spending dataset;
• •Effective customers come from the lead dataset;
• •The two datasets should be joined by account name;
• •The answer should start with the definition, then provide the table.

Common Mistakes

1. •

   Only writing field names, not business meaning.
   Bad: spend: cost. Better: spend: daily ad spend of the account, in yuan, used to calculate acquisition cost.

2. •

   Putting long rules into field comments.
   If a rule needs several sentences or multiple datasets, use a business document.

3. •

   Enabling value indexing for every field.
   Only index names, categories, and statuses that users will say directly.

4. •

   Making preferences too long.
   Preferences are for stable habits. Complex definitions belong in business documents.

5. •

   Writing business documents without descriptions or keywords.
   Descriptions and keywords help AskTable find the right document when users ask questions.

Recommended Maintenance Flow

Start with the most important datasets and add table comments first. Then add field comments to key fields. Enable value indexing for stores, channels, accounts, categories, and customer levels. Put metric definitions, jargon, and cross-data relationships into business documents. Finally, define the agent’s business background and answer style in preferences.

After launch, use five to ten real business questions to test the setup. Every time AI misunderstands something, turn the reason into configuration:

• •Wrong dataset: improve the table comment.
• •Wrong field: improve the field comment.
• •Cannot recognize a store, channel, or account name: check value indexing.
• •Wrong metric definition: update the business document.
• •Wrong answer style: update analysis preference.

AskTable semantic configuration is not about making business users learn technology.

It is about putting the business knowledge you already understand into the right place, so AI can move from “querying data” to “understanding business.”