It was a hot Atlanta day. The temperature in the cab was sweltering.  It was the kind of day that makes you feel like a grease spot on the pavement. As I bustled through the air-conditioned lobby, past reception and not catching my breath until I sat down in the deep leather chairs in the Chief Financial Officer’s conference room, I tried to compose myself.  However, the question that I was asked did not have an easy answer. I was in the HOT seat.
Bob, the CFO, opened the meeting with the statement, “Our SAP implementation was expensive. I know we need it, but I am trying to get a Return on the Investment (ROI).  I think that I can improve the ROI by using retail Point of Sale (POS) data to improve forecasting and replenishment.  How are others using POS data?”
I swallowed hard.  This question does not have an easy answer.  Most companies are struggling to answer the same question.  The company had built a repository for demand data and wanted to plug it into SAP APO, but I knew that it was not as easy as that.  I replied, “You cannot get there from here.  The data in the repository used by the account teams cannot just be plugged into APO for three reasons:  scalability, granularity, and usability. And, I started to tell him a story.”

The Story:  We Have Puzzle Pieces That Do Not Fit Together

The good news is that the sharing of retail data by retailers is increasing and the data is cleaner and more granular than it has ever been before.  For most consumer goods manufacturers, 40-60% of the channel is available as daily data received on a weekly basis.
The bad news is that POS technologies are over-hyped and confusing. In the fourth quarter of 2009, there were 72 instances of downstream data repositories sold in the North American market.  Often at more than one per company, and sometimes more than one technology sold into the same account team.  It is not unusual to have 4-6 different technologies to manage POS in the technology ecosystem of a branded manufacturer.
The market for downstream data repository vendors is very competitive and over-crowded.  To improve market positioning, several technology providers have added forecasting to their offering.  These companies have very little understanding of the differences between account level and corporate level forecasting.  At a minimum, it varies by focus (short term versus long term), the data model, and the depth of statistical forecasting.  (Account level forecasting is short-term (0-8 weeks) and corporate forecasting is longer term forecast.  On average the longer term corporate forecasting process uses demand data to project the period of 0-78 weeks).
Likewise, more than 80% of the market has an Advanced Planning System (APS) for corporate forecasting.  Most companies implement APS to model ship from data—either shipments or orders—on a monthly or a weekly basis.  Traditional APS technologies also use rules-based consumption to split the forecast into daily targets—usually termed “buckets” to drive replenishment. The problem is that rules-based replenishment is never right.  And, POS data requires a ship-to or a market-driven forecasting data model (outside in).
The data in the account team is used for weekly forecasting and replenishment.  It is isolated, seldom integrated and used for ad hoc analysis.  Very few companies have figured out how to plan globally at a corporate level to act with greater insight at a sales account team level.  Similarly, the data needs to flow bi-directionally to enable demand sensing by the account team to spotlight opportunities for improving sales through pull-based replenishment, minimizing costs through demand orchestration, and reducing channel and company inventories.  Today, 98% of companies do not have a synchronized demand signal and it is not as easy as just plugging POS data into APO. It takes more than that.  To use the POS data and drive maximum value, you have to cross the Great Divide.

The Answer:  It is Like Crossing the Great Divide

Demand data is the river that runs through the corporation that gives it life.  There are many—not just one stream—of demand data to harness.
The great divide is a name given to a mountainous region that forms a hydrological divide separating watersheds.  For example, the continental divide in the United States separates the watersheds that drain into the Pacific Ocean from those river systems that drain into the Atlantic Ocean.
In a similar vein, forecasting is the great divide between go-to-market and supply-side processes.  It is a rocky process— each company defining it a bit differently—and few companies being entirely happy with what they have.  In this time of demand volatility, where demand error is the largest risk to the corporation, it is often contentious.  To be successful with POS data integration, you need two hierarchies:  a demand and a supply hierarchy.  Most companies only have one:  a supply hierarchy. Therefore, they cannot cross the great divide, and they have no place to put POS data for modeling.
Good forecasting processes start with the end in mind.  Each of the functions have a different goal.
Let’s take a look at history.  As you read this synopsis, it is clear that the goal has changed, the data input choices have improved and the processes are being redefined to move from a supply-focused forecast to a demand-driven process.  It is a shift from an inside-out (supply side modeling based on shipments and orders) to an outside-in forecasting process (demand side modeling based on downstream data) to focus on market drivers.
Forecasting Supply
In the 1990s—the go, go years of supply chain management—new forecasting systems answered two questions for supply:
What should manufacturing make?
What should we stock in the warehouse to improve customer service?
These optimization engines used orders and shipments to generate a forecast for the future based on history.
Late in the 1990’s processes for internal collaboration were added—to align sales, marketing, finance, manufacturing and supply chain—around a common plan.  There was an assumption that the parties could give unbiased, accurate input, and that this could form the horizontal bridge across functions.  This assumption proved to be false.  To improve forecasting, bias and error accountability in consensus forecasting was adding in the period of 2005-2010.
Forecasting Demand
In the period of 2000-2010, short-term forecasting (weekly/daily forecasting for 1-8 weeks out in duration) in the sales account teams became a retailer expectation.  As a result, account teams started forecasting using point of sale data for Vendor Managed Inventory (VMI) programs to answer the question, “What will the retailer sell?”  As these account teams proliferated (the average consumer products team has 22 accounts teams just in North America), and point of sale data grew exponentially, companies started asking themselves the questions of:
How can I connect these account team forecasts to corporate demand planning?   How can I best use the growing availability of point of sale data to improve the corporate forecast?
This data is usually isolated in the account teams.

The Conundrum

Traditional APS architectures do not support the direct connection of corporate forecasting and account team forecasting.  Likewise, the direct usage point of sale data in corporate forecasting can be problematic.  There is no over-arching demand management architecture for companies to plan globally and execute at the account team level.  To make this happen, seven things need to happen:

1.  Change hierarchies. Corporate forecasting must be re-implemented to model ship-to relationships based on market drivers. The hierarchy needs to have account level granularity to ensure modeling flexibility.
2.  Assess Technology fit. The fit of the optimization engine to model downstream data must be accessed to determine optimization engine best fit and scalability.  This is often a re-implementation.
3. Build Process Governance.  A governance model needs to be established between account teams and the corporate planners.  Often the account teams own the bottoms-up forecast for the account for the short-term forecast and input these numbers into the corporate plan. Another alternative is to bring short term forecasting into the model as an indicator. Indicators are based on forecasting on multiple demand streams.  Alerts are then triggered by comparing indicators.
4. Eliminate Rules-based forecast consumption.  Move to daily statistical forecasting to determine a pull-based signal.  The greatest return on the usage of downstream data has happened making this move.
5. Infuse Discipline into Consensus Forecasting.  Insert bias and error accountability into consensus forecasting.
6. Persistence and Data ownership.  The demand forecasting process needs a persistence layer and a data owner. As POS moves from weekly data sharing weekly to daily data sharing daily, data volume grows exponentially, this requirement grows in importance.
7. Demand translation and Demand Orchestration:  To translate demand data across the watersheds of the Great Divide, focus on the definition of demand translation and demand orchestration processes.  Demand translation is the process of translating a ship- to or market-driven hierarchy to a supply hierarchy.  This requires a deep knowledge of mix, supply capabilities and shipment types/characteristics.       Demand orchestration is the process of pushing the demand forecast back from the supply-side to the demand-side hierarchy to analyze and maximize opportunity. These what-if analysis to evaluate the impact of demand shaping – sales incentives, marketing programs, new product launch, promotions, price management, and inventory obsolescence programs—on the corporate opportunity.

So, now you know the answer that I gave the CFO on his bulletin board on that hot Atlanta day.  My hierarchies were not drawn as pretty, and the board was full of scribbles.  It was a deep discussion, but at the end, he thanked me.
So, net/net, POS data sharing is a great opportunity, but it takes more than just pumping POS data through traditional forecasting systems.  As you work on this definition, it is important to keep in mind:

1. One number forecasting is a hoax.  Do not hire ANY consulting that tries to convince you to strive for this as an end goal.  Instead, focus on a common plan understanding that the data must be synchronized based on demand translation and demand orchestration.
2. Demand translation needs to be based on the operating strategy.  Get clear. Spend to time to develop the right strategy for your company to bring corporate financial forecasting, corporate market-driven and supply forecasting together.  If your strategy is to be demand driven, the market-driven forecast modeling drives the plan.  The output of this forecast is an input into the corporate financial forecast, but the market-driven forecast is never constrained by the financial forecast. The supply forecast is populated from the market-driven forecast based on demand translation of from the ship- to to the ship- from model.
3. Make sure there is discipline.  Many companies make the mistake in the formulation of the market-driven forecast of believing that they can just ask sales what they are going to sell.  WRONG.   This is a sales-driven, not a market-driven approach.  The sales organization, by definition, is COIN operated.  They are paid on bonus and commission, and their forecast is inherently biased.  In developing the market-driven forecast, focus on market drivers.  Spend time to get it right.

The good news is that the CFO let me finish the entire story.  He liked the analogy of the Continental Divide.  The demand hierarchies are like mountain ranges with different watersheds.  Demand translation is the path between them.
What do you think? Any stories of demand excellence that you want to share?  What would you have told the CFO?
<Please excuse the length of this post.  This is a complex concept asked frequently, and I thought that it deserved a detailed answer.>