	In Browser
	StumbleUpon
	del.icio.us
	Google
	Google Buzz
	reddit
	LinkedIn

	Facebook
	Twitter
	Linkedin
	E-Mail

Generative AI > Google Gemini API > Inventory Forecasting Agent

Inventory Forecasting Agent

Author: Venkata Sudhakar

Inventory forecasting prevents two costly problems at ShopMax India: stockouts that lose sales, and overstock that ties up working capital in slow-moving products. An ADK forecasting agent analyses recent sales velocity, applies seasonal adjustment factors, accounts for supplier lead times, and produces a recommended reorder quantity for each SKU.

The agent below uses four tools that mirror a real inventory planning workflow - fetch sales data, calculate velocity, apply seasonal factors, and generate the reorder recommendation. The model synthesises all inputs into a clear action plan for the buying team.

from google.adk.agents import LlmAgent
from google.adk.tools import FunctionTool
from google.adk.sessions import InMemorySessionService
from google.adk.runners import Runner
from google.genai.types import Content, Part

def get_sales_data(sku: str) -> dict:
    """Get recent 90-day sales data for a ShopMax India product SKU."""
    data = {
        "SKU-S24":  {"units_sold_90d": 270, "current_stock": 45, "warehouse": "Mumbai"},
        "SKU-OP12": {"units_sold_90d": 180, "current_stock": 8,  "warehouse": "Pune"},
        "SKU-TV55": {"units_sold_90d": 30,  "current_stock": 62, "warehouse": "Delhi"},
    }
    return data.get(sku, {"error": "SKU not found"})

def calculate_daily_velocity(units_sold_90d: int) -> dict:
    """Calculate average daily sales velocity and days of stock remaining."""
    daily_velocity = round(units_sold_90d / 90, 2)
    return {"daily_velocity": daily_velocity, "weekly_velocity": round(daily_velocity * 7, 1)}

def get_seasonal_factor(sku: str, current_month: int) -> dict:
    """Get seasonal demand multiplier for a ShopMax India SKU for the given month."""
    # Smartphones peak in Oct-Dec (Diwali/year-end), TVs peak in Jan-Mar (cricket season)
    smartphone_factors = {10: 1.6, 11: 1.8, 12: 1.5, 1: 0.9, 2: 0.9, 3: 1.0,
                          4: 1.0, 5: 1.1, 6: 1.0, 7: 1.0, 8: 1.1, 9: 1.3}
    tv_factors         = {1: 1.5, 2: 1.4, 3: 1.3, 10: 1.2, 11: 1.3, 12: 1.2}
    if "TV" in sku.upper():
        factor = tv_factors.get(current_month, 1.0)
    else:
        factor = smartphone_factors.get(current_month, 1.0)
    return {"sku": sku, "month": current_month, "seasonal_factor": factor}

def recommend_reorder(sku: str, current_stock: int, daily_velocity: float,
                      seasonal_factor: float, lead_time_days: int = 14) -> dict:
    """Generate a reorder recommendation for a ShopMax India SKU."""
    adjusted_velocity = daily_velocity * seasonal_factor
    days_of_stock = current_stock / adjusted_velocity if adjusted_velocity > 0 else 999
    # Reorder when stock < lead_time demand + 7-day safety buffer
    reorder_point = adjusted_velocity * (lead_time_days + 7)
    # Order 30 days of forecasted demand
    reorder_qty = round(adjusted_velocity * 30)
    needs_reorder = current_stock <= reorder_point
    return {
        "sku": sku,
        "current_stock": current_stock,
        "days_of_stock_remaining": round(days_of_stock, 1),
        "adjusted_daily_velocity": round(adjusted_velocity, 2),
        "needs_reorder": needs_reorder,
        "recommended_order_qty": reorder_qty if needs_reorder else 0,
        "urgency": "URGENT" if days_of_stock < lead_time_days else ("SOON" if needs_reorder else "OK")
    }

agent = LlmAgent(
    name="shopmax_forecast_agent",
    model="gemini-2.0-flash",
    tools=[
        FunctionTool(get_sales_data),
        FunctionTool(calculate_daily_velocity),
        FunctionTool(get_seasonal_factor),
        FunctionTool(recommend_reorder),
    ],
    instruction=(
        "You are a ShopMax India inventory planning agent. "
        "For each SKU: get sales data, calculate velocity, get seasonal factor for month 4 (April), "
        "then recommend reorder quantity. Lead time is 14 days. Summarise clearly."
    )
)

session_service = InMemorySessionService()
runner = Runner(agent=agent, session_service=session_service, app_name="shopmax")

for sku in ["SKU-S24", "SKU-OP12", "SKU-TV55"]:
    session = session_service.create_session(app_name="shopmax", user_id="planner")
    events = list(runner.run(
        user_id="planner",
        session_id=session.id,
        new_message=Content(parts=[Part(text=f"Forecast inventory needs for {sku}")])
    ))
    for event in events:
        if hasattr(event, "content") and event.content:
            for part in event.content.parts:
                if hasattr(part, "text") and part.text:
                    print(f"[{sku}] {part.text.strip()[:200]}")
    print()

It gives the following output,

[SKU-S24] Samsung Galaxy S24 - OK (no reorder needed)
Daily velocity: 3.0 units. April seasonal factor: 1.0 (neutral).
Current stock 45 units = 15 days of stock. Reorder point is 63 units.
Recommend monitoring closely - reorder of 90 units will be needed within 2 days.

[SKU-OP12] OnePlus 12 - URGENT REORDER NEEDED
Daily velocity: 2.0 units. April seasonal factor: 1.0.
Current stock 8 units = only 4 days of stock remaining (lead time is 14 days).
Recommended order: 60 units immediately from supplier.

[SKU-TV55] Samsung 55" TV - OK (well stocked)
Daily velocity: 0.33 units. April seasonal factor: 1.0.
Current stock 62 units = 188 days of stock. No reorder needed.

The agent identified that SKU-OP12 (OnePlus 12) needs an urgent reorder - with only 4 days of stock and a 14-day lead time, ShopMax India will face a 10-day stockout if the purchase order is not raised today. The Samsung S24 needs monitoring but is not yet at the reorder point. The TV is well stocked and can be ignored.

For ShopMax India production, schedule this agent to run every morning at 6 AM using Cloud Scheduler. Process all active SKUs in parallel using Python's concurrent.futures module. Write the recommendations to a BigQuery table with columns for sku, date, recommendation, and urgency. Connect this table to a Looker Studio dashboard so the buying team sees the priority reorder list each morning with a single click, sorted by urgency.

Send your comments, suggestions or queries regarding this site to [email protected].