Problem

• Business answers were spread across sales, finance and inventory modules
• Cross-department questions needed analysts and several reports
• Stock risk depends on demand, vendor delivery times and even weather near the warehouse
• Outside factors like weather were never part of the picture
• AI on business data must never take actions on its own
• Getting a simple answer from the ERP took far too long.

Solution Exploration

• Building more static reports and dashboards
• A single general-purpose AI chatbot over all the data
• A lead assistant with a specialist for each business area, backed by strict safety rules

How Dezoko Helped

• A team of specialists: A lead assistant routes each question to the right specialist for sales, finance, procurement and inventory, or supply-chain risk.
• Cross-department answers: One question can draw on several specialists at once.
• Weather-aware supply chain:
  • The assistant checks the live weather forecast near the warehouse.
  • It combines the forecast with stock levels, sales trends and vendor performance.
  • It flags whether a reorder needs attention.
• Safe by design: The assistant declines to run transactions, and reorder recommendations still need a person to act.
• Honest about gaps: When the data needed for an answer is missing, the assistant says so.

Safety and Control

• Transaction requests are always declined
• Reorder decisions stay with a person
• Missing data is disclosed rather than guessed
• Off-topic and ambiguous questions are handled safely

Key Outcomes

• Most questions are answered in about 5–20 seconds
• Complex cross-department questions are answered in 1–2 minutes
• The assistant never places orders or runs transactions by itself
• Accuracy issues found in testing were fixed and rechecked
• Ready to connect to a real SAP environment as the next step

How We Built It

We built the SAP Intelligence Agent on Google Cloud. Here is how it works:

A Team of Specialist Assistants

• A lead assistant reads each question and routes it to the right specialist.
• Four specialists cover sales, finance, procurement and inventory, and supply-chain risk.
• Every assistant is powered by Google Gemini.

Answers Straight From SAP Data

• The specialists look up answers in SAP sales, finance and inventory data.
• The data is organised using Google Cloud's ready-made foundation for SAP data.
• This proof of concept uses Google Cloud's public synthetic SAP sample data.

Weather-Aware Stock Decisions

• The supply-chain specialist checks the live weather forecast.
• It weighs the weather against stock levels, sales trends and vendor performance to flag reorder risk.

What We Learned Along the Way

• We tested the assistant with missing data, ambiguous questions, off-topic requests and attempts to place orders.
• Full end-to-end testing surfaced several accuracy issues, which we fixed and rechecked.
• The sample data has no record of when customers pay, so instead of guessing, the assistant clearly says it can't compare vendor payments with customer collections.

Built With

• AI: Google Gemini
• Business Data: Google BigQuery
• SAP Data Foundation: Google Cloud Cortex Framework
• Hosting: Google Cloud

Business Impact

• Answers in 5–20 seconds for most questions
• 1–2 minutes for complex cross-department questions
• Specialists keep each answer focused and accurate
• The whole setup can be recreated reliably and consistently