Kahin: see demand before the stock runs out
Kahin is the demand forecasting engine that runs inside HaROS. It analyzes your sales, historical data, live stock and site traffic together. The result is a per-product forecast and campaign suggestion.
Respond to changing demand in time
Demand shifts quickly with campaigns, seasons and traffic. Kahin reads this change from live HaROS data and responds in three layers.
Starts with live data
Kahin takes sales, past orders, cost, seasonality, live stock and traffic data directly from HaROS. You do not need to prepare files. The data is correlated; forecast quality starts here.
data analysis · correlationForecasting with a neural network
Kahin uses the LSTM (long short-term memory) model, which evaluates multiple inputs together. The model keeps past patterns in memory and updates with new data.
ANN LSTMFinds products bought together
Kahin extracts the products bought together in orders. It tracks the effects of the period, customer segment and shopping habits, and turns the result into a campaign suggestion.
Apriori · FP GrowthThree steps from data to decision
Kahin reads the data in HaROS, runs the models and carries the result to the planning screens.
Data arrives from HaROS automatically
Sales, order, stock and traffic data is read continuously from HaROS. You do not need to upload files or prepare columns.
automatic data flowModels run
You choose the channel (domestic, international), product type and date range. Kahin runs ABC analysis, moving average, regression and LSTM models; it runs association analysis on the same data as well.
daily, automaticThe result returns to the plan
Per-product forecasts, accuracy (MAPE), the Gold Group products that carry revenue and association rules are listed. The forecast becomes input to replenishment and allocation; you can export the results or turn them into campaign suggestions.
MAPE · exportAnalyses from the order panel to the neural network
All analyses run with the same data selection and date range. You choose the method; Kahin lists the result per product.
Order tracking and monitoring panel
You monitor products sold, orders, customers and revenue weekly in one panel. You start the analysis with a domestic/international and date range filter.
weekly panelABC analysis (by revenue)
Separates the Gold Group products that carry 80% of revenue from products with low revenue impact. You see which products are right to stock. Each product's share, order quantity, customer count and revenue are reported.
Gold Group · 80% / 20%Factors affecting revenue
Analyzes by product type, category, customer and brand. Shows what contributes to sales growth and in what proportion.
product type · category · customer · brandMoving & weighted moving average
You define the date range, test duration and number of periods. In the weighted method you can give more weight to the latest period (for example 0.1 · 0.3 · 0.5). The result lists the MAPE table for the best-selling products.
period · weight · MAPEMultiple linear regression
Step-Wise regression is used. Parameters whose effect falls below the threshold (0.5) are eliminated. A forecast equation is produced for each product with the remaining parameters.
Step-Wise · threshold 0.5Neural network (LSTM) analysis
The LSTM model runs for the best-selling products in the date range you select. 80% of the data is set aside for training and 20% for testing. Accuracy (MAPE) is reported for each product.
training 80% · test 20%Compare the forecast with actual sales
The chart shows the LSTM forecast for a single product next to actual sales. The model applies what it learned to the test period; the deviation is measured with MAPE.
- Forecast accuracy is measured with MAPE
- You see the results of different methods for the same product side by side
- You choose the date range, channel and product scope
Illustrative LSTM analysis view
Which product is bought together with which?
Kahin extracts the products bought together in orders. It lists the most frequent product sets and "buyers of this also buy that" rules.
Most frequently co-occurring product sets
| Product set | Support |
|---|---|
| T-shirt | 0.451 |
| Shorts | 0.288 |
| Dress | 0.166 |
| T-shirt + Shorts | 0.153 |
| Tank top | 0.141 |
| Shirt | 0.135 |
| Trousers | 0.133 |
| Jeans | 0.115 |
Association rules
| Antecedent → Consequent | Support | Confidence | Lift |
|---|---|---|---|
| T-shirt → Shorts | 0.153 | 0.34 | 1.178 |
| Shorts → T-shirt | 0.153 | 0.531 | 1.178 |
| Shorts → Tank top | 0.059 | 0.205 | 1.45 |
| Tank top → Shorts | 0.059 | 0.418 | 1.45 |
| Jeans → T-shirt | 0.057 | 0.499 | 1.105 |
| T-shirt → Jeans | 0.057 | 0.127 | 1.105 |
Apriori
Scans orders, eliminates rarely seen products and examines the associations of the remaining ones. The process repeats until no meaningful product set remains.
FP Growth
Scans the data once and keeps it in a tree structure. Because it does not rescan, it runs faster than Apriori.
What data Kahin works with and what it produces
Forecast quality depends on the richness of the data. Kahin takes data from HaROS and lists the result of the chosen method per product.
Forecasts turn into plans and campaigns
Position stock according to demand
The forecast feeds directly into replenishment and allocation. The right product is in the right warehouse or store.
Clear excess stock with campaigns
Combo suggestions come from products bought together. End-of-season excess stock clears through campaign sales.
See the busy period in advance
Traffic and campaign effects are reflected in the forecast. You plan stock and capacity before the campaign.
Planning Family
Frequently asked questions
What is Kahin and what does it do?
Kahin is the demand forecasting engine that runs inside HaROS. It analyzes sales, historical data, live stock and traffic; it produces per-product demand forecasts and campaign suggestions.
Which forecasting methods does it use?
On the time-series side, moving average, weighted moving average and Step-Wise regression are used. On the AI side, LSTM runs. Products bought together are found with Apriori and FP Growth; ABC analysis by revenue is also performed.
Why is a neural network (LSTM) used?
LSTM evaluates multiple inputs together and keeps past patterns in memory. It does not need rigid assumptions like classical statistical methods. That is why it gives strong results for demand with seasonality and campaign effects.
How is forecast accuracy measured?
In every analysis, the MAPE (mean absolute percentage error) value is listed per product. Data is split into 80% training and 20% test. The model applies what it learned to the test period; the deviation is reported with MAPE.
What does ABC analysis tell me?
It shows the Gold Group products that carry 80% of revenue and the products with low revenue impact. You see which products are right to stock and each product's share. You can also get the list by product type or brand.
How does association analysis turn into a campaign?
Products bought together are listed with support, confidence and lift values. For example, if T-shirt buyers are likely to buy shorts, at season end you can combine excess T-shirts with shorts and trousers and put them on campaign.
What is the difference between Apriori and FP Growth?
Apriori rescans the data in every loop and eliminates rarely seen products. FP Growth scans the data once and keeps it in a tree structure; that is why it runs faster.
Do I need to upload data separately?
No. Kahin runs inside HaROS; it reads sales, order, stock and traffic data directly. You do not need to prepare or upload files.
See demand ahead of time with Kahin
Both stockouts and excess stock are forecasting errors. Kahin turns your sales, stock and traffic data into demand forecasts and campaign suggestions.
Request a demoRelated pages: Supply Chain Planning → · Replenishment →