December 21, 2007
CPGmatters: P&G Improves Forecasting to Lower Safety Stock
By John Karolefski
Through a special arrangement, what follows is an excerpt of a current article from CPGmatters, a monthly e-zine, presented here for discussion
Procter & Gamble is deploying demand sensing software as part of a global initiative to further improve short-term forecasting accuracy; to lower safety stock and increase the competitive advantage generated from its supply chain.
The software is being used in 20 percent of P&G’s businesses with the rollout scheduled to be complete in 2010. Using the Demand Sensing software from Terra Technology has yielded some impressive early results. Forecast error has decreased more than 30 percent and safety stock has decreased more than 10 percent, saving millions of dollars in inventory.
“The reduction in safety stock has been accomplished without increasing out-of-stocks,” said Nils Mueller, IDF Initiative Manager, SNS Demand Planning Services at the Cincinnati-based company. “In addition to these savings, we anticipate further benefits from increased productivity in the workforce. Demand planners, free from emergency changes due to inaccurate short-term forecasts, are able to focus on the longer term, analyzing long term trends and business intelligence about the marketplace and consumer behavior.”
Terra Technology introduced the first demand sensing solution for consumer products companies in 2002. Terra’s Demand Sensing (DS) analyzes downstream data daily and creates a forecast that responds to what is happening right now, not just historical trends. Forecast error typically drops by up to 50 percent.
For consumer products companies, DS is the first step toward building a Demand-Driven Supply Network, enabling reductions in safety stock, transportation costs and unplanned manufacturing changeovers. Promotional effectiveness improves and sales revenue and profitability increase. Terra’s Demand Sensing solution is a bolt-on addition to traditional demand planning systems and is completely compatible with demand planning systems from all major ERP and supply chain vendors.
“P&G is focused on winning at the two critical ‘moments of truth,’ said Mr. Mueller. “The First Moment is when the shopper is in the store and has the opportunity to choose which product to buy. The Second Moment is when the consumer uses that product. We must delight and win at both Moments of Truth. Supply chain excellence is critical to winning at the First Moment of Truth.”
Discussion Questions: What do you think of the potential of demand sensing technology to improve forecast accuracy and reduce inventory? What are the challenges of any solution looking to predict shifts in customer demand before they occur?
BrainTrust
Recent Discussions
Discussion Questions
Poll
The critical point to understand about demand forecasting is that it forecasts a sales level that is statistically most likely to occur. Statistically, it’s almost always more likely than not that that forecasted sales level won’t occur. It’s just the most likely.
There will always be variations between forecasted and actual, but forecasts that occur closer to the actual period being forecasted will be more accurate than forecasts that occur more removed. They simply take into account more timely data. So the business logic of demand sensing software is solid, but it cannot be expected to completely eliminate the variations.
That said, I believe the key to minimizing inventory levels is what you do once you have the forecast than what is forecasted. The key is in executing the plan at the merchandising level, managing the supply chain, and execution at store level.
Demand forecasting software can work wonders at developing statistical forecasts at the SKU level. Most of them are time series based–they use history to forecast into the future–which is great as long as current levels, trends and seasonality continues. Using them exclusively, however, is like “driving with the rear view mirror,” as one friend of mine likes to put it. Apparently the Terra system is more forward looking.
The real trick in using demand forecasting successfully is using causal market information effectively–those pieces of insight that history may not be accounting for, like a big promotion, a strike, or an upcoming recession. The software also tends to be pretty useless for forecasting demand on new products without a history.
Though I must admit I’m not all that familiar with the Terra system, it seems to be a significant improvement over other products on the market. Still, demand based supply chain forecasting systems have been around for years, in many ways reaching their apex in the 1990s. Terra would have to have produced something pretty extraordinary to revolutionize the way business as usual is conducted.
From my own personal experience I’m skeptical–forecasting software is a great tool, but it is only as good as the people using it and the amount of market information that they possess.
However good a forecasting system is, it usually cannot predict the random fluctuations or as we call it in statistics ‘random error’. Most CPG products have more or less stable demand drivers like seasonality and competition, in addition to factors retailers and manufacturers control like promotions and pricing. The shocks in demand are not driven by sudden changes in consumer preferences, they are driven by environmental shocks like a sudden increase or drop in temperature, a snow storm, a power outage, an increase in energy prices, etc. Some of these may be predictable but most are not.
Some Neural Network based forecasting systems have had some success in anticipating demand shocks, but these still remain the Achilles heel for consumer goods forecasting, so I am curious to know where this significant improvement in forecast error is coming from.
The technology might work great and help them reduce the amount of inventory in their supply-chain. The question is–like everything else–what happens when there’s kink in the system? We’ve all watched numerous other supply-chains which have been optimized to remove costs only to incur major supply disruptions as soon as something out of the ordinary occurs.
You can’t always predict what the weather will do, what unions might do or even the media…or a key supplier. Any of these can and will impact a supply-chain at one time or another.
In a business where an out-of-stock can result in loyal consumers switching brands the prize of saving a dollar in the short-term can negatively impact a brand’s share in the long-term.
Wonderful concept, but requires human interaction to actually “stock the shelf.” Wal-Mart has the most sophisticated replenishment system in the world but OOS are evident every time I enter a store. I suspect 90% of the OOS is due to “in inventory” items not being placed on the shelf by store personnel. Maybe someone could invent a system that organizes back room stock so a stocker can find product that is OOS at the shelf level. Hey, wait a minute, don’t all products in Wal-Mart have those chips in them?
Why can’t a scanner be used in the storage room to find and replenish OOS at the shelf? Sorry for the rant, but the majority of the problem isn’t in “systems,” it’s in the local store management.
I think the question answers itself–if the technology has proven itself to P&G, then it should be applicable across the board.
I applaud P&G for at least trying to “change the game” by using demand data instead of history.
The other comments about making sure the store executes are all valid points and need to be addressed as part of the bigger solution.
Intelligent forecasts and replenishment systems need to understand the forecasted demand using Demand Sensing technology while combining the result with shelf capacity, shelf minimums, local merchandising constraints, on-hand, on-order, and shrink numbers. An integrated system communicated to a store in a manner that can be executed will be the key to getting any benefit from a better demand signal.
Retail success is predicated upon controlling, communicating and planning every portion of the logistics cycle to mirror the demand needs of the organization. This is really what P&G is doing as they mirror the ever increasing fulfillment cycles of their customers. The logistic cycles are becoming more and more complex as the impact of the internet is becoming more of a factor in the retail environment.
P&G deserves its reputation for being one the best run companies in the world, over decades, not just compared to others in its industry, but compared to thousands of other companies in all industries. If P&G says that demand forecasting software can reduce safety stock, you can take that claim to the bank. P&G’s endorsement of a software package is a technology company’s dream come true.
Without a doubt, lowering safety stock and replenishing more frequently will lower inventory carrying costs, but at the risk of increasing out-of-stocks and upstream inventory voids. No way is P&G naive about this. Neither are they unaware that this objective (and its entire FMOC focus) depends intensely upon in-store implementation.
I’ve tiraded on this subject before, so I’ll try not to be too repetitive here. Tighter forecasting is desirable, but it can only succeed where store-level practices are elevated to match. This is not just about upping store management discipline or throwing more arms and legs at the job. It’s about about establishing embedded, daily practices for store-level implementation that enable compliance with category plans, promotion plans and shopper media plans.
Demand-based forecasting is one crucial input that must be incorporated into new implementation practices that our industry will adopt in the next couple of years. It must exist in a balanced ecosystem that also includes a continuous stream of shelf condition data, a networked data and communications platform, and a vastly improved compliance discipline incorporating what I’d call “intelligent loss of work.”
Keep the mantra, “plan-do-measure” in mind. Our present practices tend to shortchange the “do” too often. Understandable, because it’s hard. But no longer acceptable.
More Discussions
Has Same-Day Delivery Become an Expectation Across Retail?
Are expectations around retail delivery speed in flux?OpenAI Names the Infrastructure Standard for Agentic Commerce
Will OpenAI or Google (or another entity) set the gold standard for agentic commerce?Is Five Below Poised For Even Greater Growth?
Five Below is hitting serious growth numbers. Can this trend continue?
The critical point to understand about demand forecasting is that it forecasts a sales level that is statistically most likely to occur. Statistically, it’s almost always more likely than not that that forecasted sales level won’t occur. It’s just the most likely.
There will always be variations between forecasted and actual, but forecasts that occur closer to the actual period being forecasted will be more accurate than forecasts that occur more removed. They simply take into account more timely data. So the business logic of demand sensing software is solid, but it cannot be expected to completely eliminate the variations.
That said, I believe the key to minimizing inventory levels is what you do once you have the forecast than what is forecasted. The key is in executing the plan at the merchandising level, managing the supply chain, and execution at store level.
Demand forecasting software can work wonders at developing statistical forecasts at the SKU level. Most of them are time series based–they use history to forecast into the future–which is great as long as current levels, trends and seasonality continues. Using them exclusively, however, is like “driving with the rear view mirror,” as one friend of mine likes to put it. Apparently the Terra system is more forward looking.
The real trick in using demand forecasting successfully is using causal market information effectively–those pieces of insight that history may not be accounting for, like a big promotion, a strike, or an upcoming recession. The software also tends to be pretty useless for forecasting demand on new products without a history.
Though I must admit I’m not all that familiar with the Terra system, it seems to be a significant improvement over other products on the market. Still, demand based supply chain forecasting systems have been around for years, in many ways reaching their apex in the 1990s. Terra would have to have produced something pretty extraordinary to revolutionize the way business as usual is conducted.
From my own personal experience I’m skeptical–forecasting software is a great tool, but it is only as good as the people using it and the amount of market information that they possess.
However good a forecasting system is, it usually cannot predict the random fluctuations or as we call it in statistics ‘random error’. Most CPG products have more or less stable demand drivers like seasonality and competition, in addition to factors retailers and manufacturers control like promotions and pricing. The shocks in demand are not driven by sudden changes in consumer preferences, they are driven by environmental shocks like a sudden increase or drop in temperature, a snow storm, a power outage, an increase in energy prices, etc. Some of these may be predictable but most are not.
Some Neural Network based forecasting systems have had some success in anticipating demand shocks, but these still remain the Achilles heel for consumer goods forecasting, so I am curious to know where this significant improvement in forecast error is coming from.
The technology might work great and help them reduce the amount of inventory in their supply-chain. The question is–like everything else–what happens when there’s kink in the system? We’ve all watched numerous other supply-chains which have been optimized to remove costs only to incur major supply disruptions as soon as something out of the ordinary occurs.
You can’t always predict what the weather will do, what unions might do or even the media…or a key supplier. Any of these can and will impact a supply-chain at one time or another.
In a business where an out-of-stock can result in loyal consumers switching brands the prize of saving a dollar in the short-term can negatively impact a brand’s share in the long-term.
Wonderful concept, but requires human interaction to actually “stock the shelf.” Wal-Mart has the most sophisticated replenishment system in the world but OOS are evident every time I enter a store. I suspect 90% of the OOS is due to “in inventory” items not being placed on the shelf by store personnel. Maybe someone could invent a system that organizes back room stock so a stocker can find product that is OOS at the shelf level. Hey, wait a minute, don’t all products in Wal-Mart have those chips in them?
Why can’t a scanner be used in the storage room to find and replenish OOS at the shelf? Sorry for the rant, but the majority of the problem isn’t in “systems,” it’s in the local store management.
I think the question answers itself–if the technology has proven itself to P&G, then it should be applicable across the board.
I applaud P&G for at least trying to “change the game” by using demand data instead of history.
The other comments about making sure the store executes are all valid points and need to be addressed as part of the bigger solution.
Intelligent forecasts and replenishment systems need to understand the forecasted demand using Demand Sensing technology while combining the result with shelf capacity, shelf minimums, local merchandising constraints, on-hand, on-order, and shrink numbers. An integrated system communicated to a store in a manner that can be executed will be the key to getting any benefit from a better demand signal.
Retail success is predicated upon controlling, communicating and planning every portion of the logistics cycle to mirror the demand needs of the organization. This is really what P&G is doing as they mirror the ever increasing fulfillment cycles of their customers. The logistic cycles are becoming more and more complex as the impact of the internet is becoming more of a factor in the retail environment.
P&G deserves its reputation for being one the best run companies in the world, over decades, not just compared to others in its industry, but compared to thousands of other companies in all industries. If P&G says that demand forecasting software can reduce safety stock, you can take that claim to the bank. P&G’s endorsement of a software package is a technology company’s dream come true.
Without a doubt, lowering safety stock and replenishing more frequently will lower inventory carrying costs, but at the risk of increasing out-of-stocks and upstream inventory voids. No way is P&G naive about this. Neither are they unaware that this objective (and its entire FMOC focus) depends intensely upon in-store implementation.
I’ve tiraded on this subject before, so I’ll try not to be too repetitive here. Tighter forecasting is desirable, but it can only succeed where store-level practices are elevated to match. This is not just about upping store management discipline or throwing more arms and legs at the job. It’s about about establishing embedded, daily practices for store-level implementation that enable compliance with category plans, promotion plans and shopper media plans.
Demand-based forecasting is one crucial input that must be incorporated into new implementation practices that our industry will adopt in the next couple of years. It must exist in a balanced ecosystem that also includes a continuous stream of shelf condition data, a networked data and communications platform, and a vastly improved compliance discipline incorporating what I’d call “intelligent loss of work.”
Keep the mantra, “plan-do-measure” in mind. Our present practices tend to shortchange the “do” too often. Understandable, because it’s hard. But no longer acceptable.