ChainCast — Industrial AI Supply Chain Strategy

Step 1

Pick your data

Upload your own CSV file or try one of our built-in samples

Drop your CSV hereor click to browse

Maximum 10 MB · CSV format · any column names

Dataset Requirements

1. Time Column: A column with dates (e.g., 2024-01-01). Avoid using IDs or Names here.
2. Value Column: A numeric column to forecast (Sales, Demand, Quantity).
3. Data Density: At least 5 unique dates are required to detect a pattern.

Data sources we recommend:

Kaggle FRED World Bank Data.gov Our World in Data

Popular Kaggle datasets: Superstore Sales, Supply Chain Analysis, Avocado Prices, Delhi Climate

Loading sample datasets...

Step 2

Map your columns

Tell us which column contains dates and which contains the values to forecast

Data preview (first 5 rows)

Date column * Column containing timestamps or dates

Value column * Column to forecast (demand, sales, units, etc.)

Series ID column (optional) For multiple products/SKUs – leave empty for a single series

Step 2

Set your targets

Tell the AI how many days to predict and how much buffer stock you need

Season length—

Horizon—

CV windows5

Models compared4

Manual Mode — Parameter Reference Guide

Not sure what values to enter? Here's a plain-English guide to every parameter below.

Horizon

How many future steps to predict.
Safe range: 7–90. Start with 14 (2 weeks of daily data).

Season Length

The repeating cycle in your data.
Daily data = 7 (weekly) · Monthly data = 12 (yearly) · Hourly = 24

CV Windows

How many times the model self-tests its accuracy before forecasting.
Keep at 5. Increase to 10 for more reliable error metrics.

p — AR Order

How many past time steps the model looks at to make its next prediction (its "memory").
Recommended: 1–3. Setting to 0 disables this component.

d — Differencing

How many times the data is "de-trended" to make it stable for modelling.
Recommended: 1 (most data has a trend). Use 0 if data is already flat/stable.

q — MA Order

How much the model learns from its own past prediction errors.
Recommended: 1–2. Setting all three (p, d, q) to 0 will cause a CV error.

P, D, Q (Seasonal)

Same as p, d, q but applied to the seasonal cycle you set above.
Defaults (P=1, D=1, Q=1) work for most seasonal datasets.

Lead Time

How many days it takes for a new order to arrive at your warehouse after you place it.
Example: if your supplier ships in 5 days, enter 5.

Service Level

How often you want to guarantee you have stock available (probability).
95% is the industry standard. 99% means very little risk but more safety stock.

Setting p, d, and q all to 0 will cause the CV engine to fail and produce no reliability score. Always keep at least one non-zero.

Forecast settings

Horizon ?

Season length ?

CV windows ?

ARIMA (p, d, q)

Max p (AR order)

Max d (differencing)

Max q (MA order)

Prediction length (days)

max="5" />

Max D

Max Q

Supply chain parameters

Lead time (days)

Service level

Holding cost per unit per day

Stockout cost per unit

Supply chain inputs

Lead time (days)

Service level

Holding cost per unit per day

Stockout cost per unit

Running forecast pipeline

Checking if your data is stable...

Looking for patterns and 'beats' in history...

Testing the Smart Pattern-Matcher AI (SARIMA)...

Measuring accuracy for different models...

Generating predictions for the future...

Calculating your inventory buffers...

Finalising your industrial report...

Forecast results

🧠 DATA CHARACTER

Analyzing your data's heartbeat...

🛡️ RISK LEVEL

Measuring historical volatility...

💡 EXPERT ADVICE

Generating strategic recommendations...

Demand forecast dashboard

Simplified View: History (solid) vs. AI Prediction (dotted).

Historical baseline profile

Pure Data: This is your raw historical performance without AI.

Model comparison (MAE)

What this shows: Mean Absolute Error (MAE) measures the strict average mathematical distance between the model's predictions and actual historical observations. A lower MAE indicates a superior model with tighter, more accurate forecasting precision.

Residual diagnostics

Ljung‑Box test will appear after forecast.

What this shows: Residuals display the pure mathematical "misses" of the model. A perfect model leaves trailing white noise centered exactly on the zero axis. Significant clustering or patterns suggest underlying seasonality that the model failed to capture.

Performance metrics

How it works – 7 steps

Understanding the methodology behind your forecast

Demand Forecasting
Turbo Edition (ARIMA / SARIMA)

Future Demand

Inventory Risk

Model Selection

Pick your data

Dataset Requirements

Map your columns

Data preview (first 5 rows)

Set your targets

Forecast settings

ARIMA (p, d, q)

Supply chain parameters

Supply chain inputs

Select series to forecast (max 5)

Running forecast pipeline

Forecast results

How it works – 7 steps

Demand ForecastingTurbo Edition (ARIMA / SARIMA)

Future Demand

Inventory Risk

Model Selection

Pick your data

Dataset Requirements

Map your columns

Data preview (first 5 rows)

Set your targets

Forecast settings

ARIMA (p, d, q)

Supply chain parameters

Supply chain inputs

Select series to forecast (max 5)

Running forecast pipeline

Forecast results

How it works – 7 steps

Data Intelligence Report

Demand Forecasting
Turbo Edition (ARIMA / SARIMA)