HANA Demand Forecasting Component Overview

The HANA Demand Forecasting component runs an algorithm on HANA to produce sales predictions for a set period in the future. The component functionality is a subset of Unified Demand Forecast (UDF), a module in SAP Customer Activity Repository (CAR). A primary focus of the component is to forecast Consumer Demand. As well as providing forecast and forecast interval information, the algorithm also provides data on price elasticity for all products in the workflow. Described below are the configurable properties, results grid, and algorithm summary of the HANA Demand Forecasting component.

Note For a guide to configuring the component properties, see Configuring the HANA Demand Forecasting Component.
Component Properties Settings
Table 1: HANA Demand Forecasting Wizard - Properties Tabbed Page Settings

Section Name

Property Descriptions

Forecast Horizon

Set the prediction period Start and End dates.

Variables

Set the following Variable properties:

Product ID: Select the string-only column from the input table that contains the product identifier code, which can be up to 60 characters long.

Location ID: Select the string-only column from the input table that contains the location identifier code, which can be up to 60 characters long.

Transaction Timestamp: Select the column from the input table that contains the transaction timestamp, which must be in date or timestamp format.

Unit Sales: Select the numeric-only column from the input table that contains the unit sales figure.

Revenue: Select the numeric-only column from the input table that contains the revenue figure.

Holidays (Optional)

Contains information about public holidays in particular locations. Set the following Holidays properties:

Schema: Select the schema for the input table from the list in the HANA database.

Tables: Select a table from the schema.

Views: Select a view from the schema.

Time Stream ID: Select the string-only column from the input table that contains the time stream identifier code, which can be up to 10 characters long.

Public Holiday Key: Select the string-only column from the input table that contains the public holiday key, which can be up to 3 characters long.

Operational Status: Select the integer-only column from the input table that contains the operational status.

Timestamp: Select the column from the input table that contains the transaction timestamp, which must be in date or timestamp format.

Locations to Holiday Mapping (Optional)

Set the following Locations to Holiday Mapping properties:

Schema: Select the schema that contains the table with information about mapping locations to public holidays.

Tables: Select a table from the schema.

Views: Select a view from the schema.

Location ID: Select the string-only column from the input table that contains the location identifier code, which can be up to 60 characters long.

Holiday ID: Select the string-only column from the input table that contains the holiday identifier code, which can be up to 10 characters long.

Demand Influencing Factors (Optional)

Set the following Demand Influencing Factors properties:

Schema: Select the schema that contains the table with information about Demand Influencing Factors.

Tables: Select a table from the schema.

Views: Select a view from the schema.

Product ID: Select the string-only column from the input table that contains the product identifier code.

Location ID: Select the string-only column from the input table that contains the location identifier code.

DIF Attribute: Select the string-only column from the input table that contains the Demand Influencing Factor (DIF) attribute, which can be up to 32 characters long.

Timestamp From: Select the date-only column from the input table that contains the date that the timestamp begins.

Timestamp To: Select the date-only column from the input table that contains the date that that the timestamp ends.

DIF Value: Select the numeric-only column from the input table that contains the Demand Influencing Factor (DIF) value.

Expected Future Prices (Optional)

Set the following Expected Future Prices properties:

Schema: Select the schema that contains the table with information about Expected Future Prices.

Tables: Select a table from the schema.

Views: Select a view from the schema.

Product ID: Select the string-only column from the input table that contains the product identifier code.

Location ID: Select the string-only column from the input table that contains the location identifier code.

Timestamp From: Select the date-only column from the input table that contains the date that the timestamp begins.

Price: Select the numeric-only column from the input table that contains the price.

Component Advanced Settings (Optional)
Table 2: HANA Demand Forecasting Wizard Properties - Advanced Tabbed Page Settings

Section Name

Property Descriptions

Configuration Parameters (Optional)

In the Configuration Parameters section, you have the option to configure the following settings:

Damping factor (FC_TREND_DAMP): Defines the damping factor for the trend regressor. The range is value >= 0.00000.

Total regress mass (MOD_HDM_NEAR_HOLIDAY_DENSITY): Sets the proportion of total regressor mass to the right (POST: left) of the half-way date. Note that there are two groups of HDM-regressors, SYS:CAL:YR:HDM:PRE:* and SYS:CAL:YR:HDM:POST:*. The PRE-regressors define the ramp-up before the holiday, the POST-regressors the ramp-down after the holiday. The range is 0.50000 <=value < 1.00000.

Time Delay Effect (Optional)

In the Time Delay Effect section, configure the following settings:

Observation weight (MOD_TIME_WEIGHT): Sets the weight of a one-year-old observation in modelling, compared to an observation taken today. This way, the variable helps to decide whether to give equal importance (or weight) to all the records irrespective of their timestamps. For example, when building the model, setting the parameter value to 1 gives equal importance to all the records irrespective of the time that they were recorded. Whereas, setting a value to the parameter less than 1 enables the user to give less importance to the records that have older timestamps, as compared to the records that have recent timestamps. The range is 0.50000 <=value < 1.00000.

Lower Boundary on weight (MOD_TIME_WEIGHT_MIN): Defines a lower boundary below which the weight will not fall. The range is 0.00001 <=value <= 1.00000.

Out of Stock Detection (Optional)

In the Out of Stock Detection section, configure the following settings:

Zero sales period(MOD_OOSD_MIN_LEN): Sets the minimum length of the continuous zero sales to be considered for an out-of-stock period evaluation. The range is value >= 1.00000.

Probability threshold (MOD_OOSD_THRSHLD): Sets the threshold for the probability score to determine whether an item is out-of-stock. The probability score for each item is derived based on the occurrence of zero sales for a period greater than the values specified in the parameter, MOD_OOSD_MIN_LEN. The range is value >= 1.00000.

Time Series Decomposition (Optional)

Select the appropriate Time Series Decomposition checkboxes to decompose and clearly see the influence on your results of the following factors:

Seasonality: Seasonality can impact the results at different times of the year, such as the start or end of the year, or every alternate month.

User Promotion: A user promotion can impact the results with sudden spike in sales.

Holidays: National holidays such as Christmas Day and Thanksgiving can impact sales.

De-select the checkboxes if you do not want to consider the impact of these factors in your results.

Outlier Detection (Optional)

In the Outlier Detection section, configure the following settings:

Configure outlier detection: Checkbox to switch on or off the outlier detection.

Acceptable distance from mean (MOD_OUTLIER_MEAN_FACTOR): Defines the outlier detection factor to determine how far away from the mean is acceptable.

Min. non-zero observations (MOD_OUTLIER_STD_DEV_FACTOR): Defines the outlier detection minimum number of non-zero observations, regular and promotional, counted before 0-filling.

Standard deviation factor(MOD_OUTLIER_STD_DEV_FACTOR): Defines the outlier detection factor to determine how many deviations away from the mean is acceptable.

Component General Settings (Optional)
Table 3: HANA Demand Forecasting Wizard Properties - General Tabbed Page Settings

Section Name

Property Descriptions

Basic (Optional)

Component Name: Not configurable.

In the Outlier Detection section, you can configure the following settings:

Alias Name: An alias for the component name.

Description: The purpose of the component.

Component Results Grid
Table 4: HANA Demand Forecasting Component - Result Grid Columns

Column

Description

PROD_ID

Product ID

LOC_ID

Location ID

TSTMP_FR

Timestamp From

TSTMP_TO

Timestamp To

ACTUAL_UNIT_SALES

Actual Unit Sales

FC_CONF_INDEX

Forecast Confidence Index (FCI)

FC_UNIT_SALES

Forecasted Unit Sales

INTERCEPT

Intercept of the time series decomposition component.

TREND

Trend of the time series decomposition component.

SEASONALITY

Seasonality of the time series decomposition component.

DAY_OF_WEEK

Day-of-week of the time series decomposition component.

HOLIDAY

Holiday of the time series decomposition component.

SALES_PROMOTION

Sales Promotion of the time series decomposition component.

PRICE

Product-location specific future price on a daily basis. Historical price calculated based on sales and unit price.

PRICE_ELASTICITY

Price Elasticity. Measures the responsiveness of the quantity demanded of a good or service to a change in its price.

FORECAST_INFO_MSG

Provides additional information that explains the Forecast Confidence Index (FCI).

FORECAST_INFO_DIF_DESC

Provides additional information that explains the demand influencing factor for that impacts the forecast.

Algorithm Summary
Table 5: Descriptions of the algorithms featured in the component

Category

Number of Product Location Combinations

Criteria

Descriptions

Perfectly Inelastic Demand

0

E = 0

The category is an extreme case because the quantity demanded is unaffected by any price change. The quantity is fixed and modifications to the price have no effect on the result.

Inelastic

0

-1 < E < 0

UDF constrains price elasticity to be less than zero and above -10 in the default parameterization. Therefore, a price elasticity above -1 is called an inelastic demand. This means that changes in price have a relatively small effect on the quantity of the good or service demanded. In contrast, if price elasticity is below -1, changes in price have a relatively large effect on the quantity of the good or service demanded, which is an elastic demand.

Unit Elastic

0

E = -1

Any change in price induces an equal relative change in quantity. For example, a 20% change in price includes a 20% change in quantity demanded. Unit Elastic is the dividing line between the elastic and inelastic ranges.

Relatively Elastic Demand

1

E < -1

The quantity demanded is extremely responsive to price because relatively small changes in price cause relatively large changes in quantity. For example, a 2% change in price leads to more than a 20% change in quantity demanded (perhaps more than 40%).

HANA Demand Forecasting (UDF)

The HANA Demand Forecasting component is a subset of Unified Demand Forecast (UDF) for SAP Retail applications on SAP HANA. It is part of the SAP Customer Activity Repository (CAR). UDF uses all necessary near-real time input data automatically out of this platform. Thus UDF relies on the Demand Data Foundation component of CAR. This component can provide for tasks such as NW-based job scheduling, batch job parallelization framework, exception workbench and configuration IMG-screen.

The unified forecasting engine represents a combination of the scientific forecasting expertise and methodologies from several sources. These include the SAP acquisitions SAF AG (SAP Forecasting and Replenishment) and Khimetrics (Demand Management Foundation).

UDF empowers business analysts with the knowledge of the impact that each Demand Influencing Factor (DIF) had on consumer demand in the past. For example, DIFs can be price changes, promotions, seasonality or a trend. The decomposed values can be used to forecast future demand to support consuming applications in Retail and Consumer Products. What's more, UDF learns from new demand data. This means that it automatically adapts the demand model to strengthen the forecast as more data is introduced.

The capabilities of the in-memory database technology are leveraged as much as possible. Therefore you can take full advantage of the opportunities provided by Big Data. As a result, you can model and forecast large amounts of data to enable new business scenarios. This enables you to support a high volume of data, to perform near real-time processing and to push detailed granular insight into demand data.

Note For further information, see the SAP Help page, Unified Demand Forecast (UDF).