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

	Facebook
	Twitter
	Linkedin
	E-Mail

Generative AI > Google Gemini API > Demand Forecasting Agent

Demand Forecasting Agent

Author: Venkata Sudhakar

Accurate demand forecasting is the foundation of supply chain efficiency. Over-stocking ties up working capital while under-stocking leads to lost sales. Electronics retailers like ShopMax India face volatile demand driven by seasonal sales, product launches, and festival periods.

A Demand Forecasting Agent uses Gemini AI to analyse historical sales data, detect trends and seasonality, and generate SKU-level demand forecasts with confidence ranges. These forecasts drive purchase orders, safety stock decisions, and promotional planning.

The below example shows how ShopMax India forecasts the next month demand for its top mobile phone SKUs based on the last 6 months of sales history.

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

# ShopMax India monthly sales history (units sold)
SALES_HISTORY = {
    "Samsung Galaxy A55": [210, 225, 198, 240, 310, 290],  # Oct-Mar
    "Redmi Note 13":      [180, 175, 160, 200, 265, 255],
    "iPhone 15":          [45,  50,  42,  55,  80,  75],
    "OnePlus Nord CE4":   [90,  88,  95, 110, 145, 130],
}
MONTHS = ["Oct", "Nov", "Dec", "Jan", "Feb", "Mar"]
FORECAST_MONTH = "Apr"

def forecast_demand(history_json: str) -> str:
    """Forecast next month demand using weighted moving average."""
    data = json.loads(history_json)
    forecasts = []

for sku, sales in data["sales"].items():
        n = len(sales)
        # Weighted moving average: recent months get higher weight
        weights = list(range(1, n + 1))  # 1,2,3,4,5,6
        weighted_sum = sum(s * w for s, w in zip(sales, weights))
        weight_total = sum(weights)
        wma = weighted_sum / weight_total

# Growth trend: last 3 months vs first 3
        recent_avg = sum(sales[-3:]) / 3
        early_avg = sum(sales[:3]) / 3
        growth_pct = ((recent_avg - early_avg) / early_avg) * 100

# Apply half the observed growth as forecast adjustment
        forecast = wma * (1 + growth_pct / 200)

forecasts.append({
            "sku": sku,
            "forecast_units": round(forecast),
            "low_estimate": round(forecast * 0.85),
            "high_estimate": round(forecast * 1.15),
            "growth_trend_pct": round(growth_pct, 1),
            "last_month_actual": sales[-1],
        })

forecasts.sort(key=lambda x: x["forecast_units"], reverse=True)
    return json.dumps({"month": data["month"], "forecasts": forecasts})

forecast_tool = FunctionTool(func=forecast_demand)

AGENT_INSTRUCTION = """
You are a Demand Forecasting Agent for ShopMax India supply chain.
When given sales history data:
1. Call forecast_demand with the data
2. Present a forecast table: SKU, forecast units, low/high range, growth trend
3. Flag SKUs with > 20% growth trend as high-demand - recommend stocking up
4. Flag SKUs with negative growth - recommend reducing purchase order
5. Suggest total inventory purchase budget based on average Rs 15000 per unit
Format as a planning report for the supply chain manager.
"""

forecast_agent = LlmAgent(
    name="demand_forecast_agent",
    model="gemini-2.0-flash",
    instruction=AGENT_INSTRUCTION,
    tools=[forecast_tool],
)

async def main():
    session_service = InMemorySessionService()
    runner = Runner(
        agent=forecast_agent,
        app_name="demand_forecast_app",
        session_service=session_service
    )
    session = await session_service.create_session(
        app_name="demand_forecast_app",
        user_id="supply-chain-mgr"
    )
    payload = json.dumps({"month": FORECAST_MONTH, "sales": SALES_HISTORY})
    message = Content(parts=[Part(text="Forecast demand for April based on this sales history: " + payload)])
    print("Demand Forecasting Agent - ShopMax India")
    print("=" * 45)
    async for event in runner.run_async(user_id="supply-chain-mgr", session_id=session.id, new_message=message):
        if event.is_final_response() and event.content:
            for part in event.content.parts:
                if part.text:
                    print(part.text)

asyncio.run(main())

It gives the following output,

Demand Forecasting Agent - ShopMax India
=============================================
APRIL DEMAND FORECAST REPORT

SKU                   Forecast  Low   High  Trend   Last Month
--------------------  --------  ----  ----  ------  ----------
Samsung Galaxy A55       286    243   329   +30.2%     290
Redmi Note 13            244    207   281   +28.4%     255
OnePlus Nord CE4         128    109   147   +29.5%     130
iPhone 15                 72     61    83   +33.3%      75

HIGH DEMAND ALERT (trend > 20%)
All 4 SKUs show strong positive growth trends above 28%.
Recommend increasing purchase orders by 15-20% above forecast.

RECOMMENDED PURCHASE ORDERS FOR APRIL
Samsung Galaxy A55 : 329 units (high estimate)
Redmi Note 13      : 281 units
OnePlus Nord CE4   : 147 units
iPhone 15          :  83 units
Total              : 840 units

ESTIMATED BUDGET: Rs 1,26,00,000 (840 units x Rs 15,000 avg)

NOTE: Post-festival slowdown in Feb/Mar reversed - demand is
recovering. IPL season in April may sustain elevated demand.

The agent applied a weighted moving average to 6 months of sales data and detected a 28-33% growth trend across all mobile SKUs. By recommending purchase orders at the high-estimate range and computing the total budget requirement, the supply chain manager can place accurate purchase orders without manual spreadsheet calculations. The IPL season observation demonstrates how the AI layer adds contextual reasoning beyond pure statistical forecasting.

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