Practical Framework for Demand Forecasting in India
Achieving high accuracy through a combination of process, technology and people
Demand forecasting is used by many organizations globally to predict the requirement of the right product at the right place and the right time. The objective is very simple, to reduce excess inventory and to reduce shadow demand.
In developed economies, the sales data is available at the retail shelf level even in smaller stores. Added to that the certainties’ of lead times, demand forecasting and replenishment planning is a predictable and a repeatable process. Demand forecasting tools like Demantra, APO and JDA, give a very good forecast accuracy even at the shelf level.
However, it has been noticed that in India, there are many factors which make it inappropriate to just fit a forecasting software to generate a sales forecast. It has been observed that the forecast by doing so, is very poor in accuracy even at the depot level.
This paper presents the output of four years of research, practice and innovation, in India, in Indian conditions, to achieve a forecast accuracy of as high as 85% for an SKU at the retail level. The accuracy at the depot level has been up to 93% in a FMCG scenario. We share the framework to do so in the following pages.
Our framework is based on deploying our solution in practice in the following types of Indian organizations
|Segment||Size (INR Crore)||SKU’s||Depots||Distributors||Pre solution forecast accuracy|
|F&B||8000||1200||800||More than 100K retail outlets||75%|
|Auto (OEM & Aftermarket)||750||400||30||550||50%|
All the organizations used some sort of sales forecasting methodology. However there was no collaborative planning process in place.
The period of the study is from 2009 to 2013, and hence it has seen the spurt of growth between 2009 and 2011 and then the recessionary downtrend subsequently.
All these organizations were not very forthcoming in disclosing their existing forecast accuracy i.e. before deploying our framework.
Data at the retail outlet level was available for only one organization, where the information is released by the government.
Though all the organizations, did have a very mature and visionary head, who understood the benefits due to these tools, there lacked a people organization to run and execute the process.
The Indian Scenario
First of all, in India, 95% of the organizations do not forecast demand of a product. The forecast the sales. In fact, many do not even understand the difference.
If a particular market segment is being services by 5 brands of 3 different organizations, the total demand is the sum of what all the three companies together fulfill. Whereas, if only sales were being forecasted, a company will tend to predict only its market share. So for strategic and market growth purposes, it is better to forecast demand. However, when you do not have the data of not only your sales figures at the retail level, how do you expect to get the data of your competitors. Forecasting demand rather than sales also predicts much better, what can be the shadow demand, and hence the safety stock required to service that demand.
Availability of data
While almost all the companies do have timely reliable data on secondary sales, there is still a paucity of data at the tertiary level (distributor to retailer) and at the end consumer level. This leads to forecasting being reliable only at the distributor level . This is in sharp contrast to the developed markets where sales data from a particular shelf is available. Data of any past promotions, schemes or events is totally unavailable in a manner that it can be used to further predict.
There are two pan India significant business processes that lead to unreliable distributor level sales data. Using this data in Toto for forecasting will lead to inaccurate forecasts.
The first is the push by company sales persons to push the distributor to load stock in the last week of the month so that the sales targets are achieved. The distributor invariably returns the stock in the first 2 days on the next month.
The other is the credit cycle of the distributor itself. By the time he realizes money from his debtors, it is the 20th of the month and only then he places orders on the company. This leads to a skewed sales towards the end of the month. So using a weekly forecasting system will again give skewed results. The pictorial below shows the picture of how sales in the first week and last week compare
Fig: Week 1 of the month
Fig: Last week of the month
Non Availability of collaborative planning processes
It may seem surprising, but even very mature and large organizations, do not have an established collaborative planning process. The concept of S&OP is existing but it is not in the true sense that all stakeholders come together and decide on a particular figure. The result is that sales have their own forecast and they continuously chase the supply chain for fulfilling their number, irrespective of sales actually happening or not. Furthermore, there is no metric to track the performance of sales against what they had forecasted and what they sold.
Lack of understanding of statistical forecasting
There is less understanding of how statistical forecasting can be used to generate a baseline forecast that can be moderated upon. The most commonly used method is moving average and multiplying it with a growth factor which is equivalent to what growth the company is experiencing. Such a method leaves out seasonality and events in the data.
Additionally, it has been seen that organizations do not want to believe the ROI that can come from a good forecast. The thought of investing in a technology that can help generate a good forecast is not encouraged. In fact, it is through multiple pilots that these corporations get convinced that a technology driven forecast can give more than 20% benefit on paper and at least 5% inventory reduction overall, which is a huge amount practically.
Forecasting models to suit Indian environment
While, there are so many statistical models available, in Indian scenario, with the kind of data and expertise available, it has been seen that a few give very good results. Hence, it may not be prudent to force fit the standard models in the Indian scenario. However, the lack of understanding in organizations of either the kind of data they have or these models, makes them use the more popular models only, which leads to a higher error.
All the above issues that have been studied in the candidate organizations can be classified into three broad categories as shown in the table below.
|Technology issues||Process Issues)||People Issues|
|Unavailability of reliable downstream data Unavailability of event and promotion related data Unavailability of cost effective demand planning software tools Less willingness to invest in tools||Collaborative S&OP process not part of most organizations||Concept of demand planner does not exist Understanding of demand planning and statistical forecasting limited|
LookAhead: The holistic people, process and technology framework for Indian Scenario
In all the organizations studied, a holistic approach was followed, to see if the forecasting accuracy improves on a long term basis.
The LookAhead Supply Chain solution is a framework that proposes
- To set up the processes and manage the change
- To provide technology to capture tertiary sales data, sales forecasting, supply planning, dispatch planning and analytics
- To provides a planner on demand service to run the entire solution
- And can finally make all this available as a subscription based service with cloud accessible software minimizing CAPEX and increasing ROI.
LookAhead Process Solution
First of all the acceptable process of collaborative planning has been setup in all the organizations. A snapshot of the process setup is shown in the figure below. This process brings together the stakeholders in terms of finance, sales, production, marketing and supply chain along with the executive management. It has been seen that simple events like promotions that would be run by sales have not been shared with supply chain in the past. And suddenly during the scheme period supply chain has been bullied into supplying more material. By collaborative planning, these proposed schemes have now been known beforehand, leading to better forecasting.
Though the acceptability of such meetings by the Sales organization has been very difficult to inculcate as a culture, it has been observed that if the head of the organization is available during the S&OP meeting, change management becomes easier. Also, ultimately, sales have to be made the owner of the forecast.
A good S&OP process also helps organizations improve their profitability. In times when demand outstrips supply, these meetings are helpful in identifying the more profitable product that will be sold hence resulting in better utilization of the resources.
A S&OP meeting also improves the overall productivity of the organization, as it was observed in these organizations that slowly, people started owning up to their numbers forecasted, and gave a better forecast. Therefore less time was spent in fire fighting.
Fig: Collaborative planning process as part of the framework
Technology Solution for the framework
The first and foremost is to be able to capture data at the lowest level. It is easy in developed countries where the levels of IT penetration are very deep in a small retail store. In India, issues like poor IT infrastructure, poor connectivity and the general mistrust in sharing data with the next higher level, leads to very poor or spotty data at the retailer level. We have to rely on surveys done by marketing companies to arrive at what is the market share at the retail level.
Hence, we have developed and deployed technologies that capture sales at the retailer level. These include a smartphone based sales capture system that can be deployed at the retailer or can be used by the sales person to capture the sales. However, there are still challenges of who shares the cost of the device as well as the data plan.
Also included is a modem that is attached to a distributors or a retailers system (if he has one) and it automatically logs n to the system and captures the daily sales and transfers it to a central server. Figure below shows the architecture of such a system.
Fig: Automatic capture of sales at secondary and tertiary level
It has been observed in the case of an FMCG company part of the study, that the sales forecast accuracy improved dramatically once it was known what were the tertiary sales.
Additional technology solution as part of the framework includes making available the technology part of the solution including the forecasting software on the cloud. This has drastically reduced the Total Cost of Ownership of organizations and now they are more willing to invest the reduced amount required and reap the benefits.
Statistical forecasting models for India.
As India is in a growth phase (low or high does not matter), it has been observed that instead of the traditional ARIMA or Holt Winters models, Logarithmic trend with 12 period seasonality gives excellent results.
It has been observed that in very large corporations in India, the investment in software for demand planning fails very quickly after implementation. This is due to no organization being available to run the process till it is ingrained into the system.
As a result, when these organizations were provided with a demand planner, and continuous support to run the whole demand planning process, it was seen that the organizations reaped the benefits of the advanced tools and technologies. Similarly it was seen that, when an external consultant facilitates the S&OP meeting, there is less acrimony and acceptance is much higher.
By deploying the people, process, technology solution framework in the studied companies, there has been a dramatic improvement in the sales forecast accuracy. The documented and acknowledged improvements range from 10% to 40% improvement in forecast accuracy.
Hence we as Gazelle Information Technologies have come to a conclusion that in emerging markets like India, a practical framework for improving demand planning has better long term results than just implementing a software.