Chapter 3 · Office Work, Solved — One Task at a Time

3.3 Data Analysis & Insights

You have sales data, inventory data, and supplier cost data, scattered across different Base. Field names don't match, formats differ, languages differ. This lesson uses multiple tables in Base: check field meanings and join rules first, then ask the agent to clean, merge, and analyze the data.

Resources Used

TypeResourceWhat it does here
AppBaseHolds the sales, inventory, and supplier source tables
SkillData cleaningCleans messy data into analysis-ready tables
SkillAnalysis methodologyProvides reusable steps for retail analysis

Prepare source tables with Tables and fields and Record operations. The Base task exercise shows querying and writing data in a conversation. See Views for sorting and grouping results.

Steps

  1. Make sure data sources are in Base (sales table, inventory table, supplier table, etc.)

  2. Tell the Agent what to analyze and which tables to pull from

  3. The Agent auto-queries multiple tables and identifies field mappings

  4. Merges data and runs analysis (trends, rankings, anomalies, margin, etc.)

  5. Write results to the specified Base and generate the requested report, then check record counts and calculations

Exercise: Calculate Italy sell-through

Calculate sell-through for Italy from 2026-10-01 through 2026-10-07 in three steps:

  1. Read and match the source tables. Select Base and read Sales and Inventory. Their product-code (SKU) fields are named ProductCode and Codice; matching codes refer to the same product.
  2. Calculate rates and separate exceptions. Use units sold divided by units received (Sold / Received). Of the five products, three have usable receipt quantities. One has zero receipts and one is missing the receipt quantity; list these separately.
  3. Save the analysis. Write the three ranked products and two exception records to the Sales Analysis table in Base, and generate Italy Sales Analysis.md. The next lesson uses this report for both an HTML brief and a PowerPoint presentation.

Sample Prompt

Read Sales and Inventory from Base and identify the product-code, units-sold, and units-received fields.
Use Italy data from 2026-10-01 through 2026-10-07 and calculate sell-through as units sold divided by units received. List products with zero or missing receipt quantities separately.
Sort valid rates from lowest to highest and save them with the exceptions to Sales Analysis.
Generate Italy Sales Analysis.md, identify the product with the lowest sell-through, and list the additional data needed to investigate the cause.

Expected Results

  • Sell-through analysis written to the Base, sorted ascending

  • The slowest-selling SKUs in Italy are clearly listed

  • This example lists three ranked products and two exceptions; the report identifies A101 as having the lowest sell-through and lists follow-up investigations

  • Later, you can ask the Agent to drill into any SKU's details

Advanced: Multi-Source Data Cleaning

If you have multiple data sources from different systems — SAP sales CSV, Magento e-commerce Excel, supplier quotes in Italian — field names don't match, formats differ, languages differ. Align the fields and record definitions before analysis.

With AOS Work, you hand all the data to the Agent. AI auto-identifies field mappings, cleans the data, merges it into one unified table, and gives you a clean data foundation for analysis.

Steps

  1. Upload each dataset to a Base

  2. Tell the Agent what each dataset is and ask it to auto-identify field mappings

  3. The AI shows you the field mappings (e.g. SAP's "Product Code" = supplier's "Codice"), you confirm

  4. The AI runs data cleaning: deduplication, date format unification, currency normalization, missing value annotation, anomaly detection

  5. The cleaned, merged data is written back to the Base, ready for analysis

Sample Prompt

Here are three datasets: SAP store sales (fields in English: Store_ID, Product Code, Revenue, Date), Magento online orders (fields in English: Product Code, Total order, Discount, Revenue), and supplier product costs (fields in Italian: Codice, Prezzo, Categoria). Auto-identify field mappings between these datasets — fields that mean the same thing but have different names — and show me the mapping for confirmation. Also list what one row represents in each table, which fields match records, how to combine them, and the expected merged row count to prevent double counting from many-to-many joins. After confirmation, clean the merged data:
1) Define what one row represents and which fields uniquely identify it in each table, list possible duplicates, and deduplicate after confirmation
2) Unify dates to YYYY-MM-DD and retain original currencies. Before converting to EUR, list the required exchange rates, dates, and sources; convert after I provide and confirm them
3) Mark missing values and suggest how to handle them
4) Detect anomalies and flag them. Write the cleaned merged data back to a Base.

Expected Results

  • Field mapping inventory: AI auto-identifies field mappings across the three datasets

  • Cleaned merged table: deduplicated, format-normalized, missing values annotated, anomalies flagged

  • Data quality report: number of duplicate records, missing value distribution, anomaly list

  • Merged data written back to the Base — ready for analysis without re-cleaning

Data cleaning comes before analysis

Confirm field mappings and cleaning rules, then check row counts, duplicates, and calculations. Want to know what to do with the cleaned data? → See 4.8 "Merchandise Planning: Product Operations Analysis & Inventory Risk Alerts".