Executive Briefings

By the Numbers: How New Analytical Methods Can Help Improve On-Shelf Availability

A conversation with Rahul Tyagi, associate director of retail and CPG with Tata Consultancy Services.

By the Numbers: How New Analytical Methods Can Help Improve On-Shelf Availability

The proper forecasting of retail demand has always required a certain level of intuition. But planners today can no longer rely on a “seat-of-the-pants” approach to calculating inventory and production levels. The consequences of being wrong are simply too high. The new era of retailing requires the application of sophisticated analytical and statistical methods to match supply with demand, as well as determine the impact of getting a particular forecast right or wrong. In this interview, conducted at Gartner’s 2014 Supply Chain Executive Conference in Phoenix, Ariz., TCS’s Rahul Tyagi describes the new, numbers-oriented approach to demand planning and forecasting – one that still requires the presence of humans to make it work.

Q: Why is on-shelf availability such a huge challenge for retailers and their supply chains, after all these years of trying to make it work?

A: Tyagi: There are basically two primary causes. One is external and market-driven, and the second is more internal. There’s been an explosion in buying channels over the last few years. Retailers that typically sold through stores and catalogs are now selling through mobile phones, internet and B2C. That has caused a fragmentation of demand. To capture all of these demands independently has become a big challenge.

Secondly, there’s been a demographic shift. Within five years, the overall buying power will shift from the Baby Boomers to the Millennials. Given the nature of Millennials – very fickle and not too loyal to a particular vertical or channel – capturing their behavior becomes very important. And this is a very well-informed generation. They use all possible forms of information.

In terms of internal challenges, there are a lot of things that need to come together. You have to have a perfect marriage of supply-chain planning and merchandising. The right mix of forecasting, replenishment and merchandising is not happening right now. These are the primary reasons why on-shelf availability is getting more important over the last few years.

Q: Why are companies having such a hard time getting forecast accuracy right?

A: Tyagi: As I mentioned, there’s channel proliferation happening, and data coming from all sources. A big challenge for any company, especially in retail and consumer packaged goods, is how to harness this treasure-trove of data coming from all these sources. How do you synthesize it to generate a good forecast?

With traditional forecasting solutions, there are two elements that are really difficult to get answers on. One is that most of these solutions are driven off of sophisticated items, which require a lot of different parameters that need to be managed actively. You need to have a certain level of statistical science know-how to manage that. Secondly, planners at CPG companies are typically managing forecasts for over a million item-location forecasts a week. As a result, one gets into approximations, rules of thumb, heuristics. The ultimate consequence is that the accuracy starts going down.

Planning, especially in retail, is an entry-level job. So you’ve got people fresh out of college who don’t have the experience or wherewithal to manage these forecast models accurately. This causes forecast accuracy to fall over time.

Q: When you say “science” is an answer, what do you mean?

A: Tyagi: Science means the use of analytical and statistical methods that can enable a better understanding of the demand pattern that underlies a particular item’s behavior at a location. Let’s say we’re talking about a 64-ounce bottle of Gatorade, selling in a store at a particular geography. In order to forecast that well, one has to have a very clear understanding of how the demand pattern for this item is related to the promotional response, and the seasonal response. Is it uniform, or does it vary by seasons? To understand this, we have to use analytics and statistical methods to completely deconstruct these different patterns into elements that can be reconstructed into a forecast model. What we have to look at is a better way of doing this deconstruction and reconstruction.

Q: Were these statistical models and analytical procedures not available to companies before? What has enabled them to take advantage of them now?

A: Tyagi: Some were not available. They have evolved over time, with the need for high-end, rocket-science techniques. Statistical and analytical techniques have been available for a long time – fifty-plus years. But there wasn’t enough technology, hardware or maturity to handle that kind of processing.

These are extremely computational-intensive techniques. Even 10 years ago, we didn’t have the processing power, and the cost of memory was too high. Now we’re able to do things that were unimaginable back then.

Q: Are the people to utilize this technology available today?

A: Tyagi: A lot of the things that a planner is expected to do has been abstracted into these statistical methods. I would say that over the next five years, we will see that become ingrained in the forecasting tools that are offered. What we’ll ultimately end up with is a forecasting tool that automatically does this deconstruction and reconstruction of the demand patterns, and also has the ability to self-correct. Techniques such as machine learning will ensure that the forecast error gets analyzed and gets linked back to the model, in a self-correcting loop.

These kinds of things are extremely computational-intensive. And given both memory and computational trends, it is a possibility now, and will be a reality going forward.

Q: We talk about a shortfall of human expertise. But are we to some extent taking humans out of the equation completely?

A: Tyagi: Absolutely not. Humans have a very important role to play. This is going to be a statistical demand-forecasting tool. It cannot have human intelligence – there’s no way for a tool to know how I change my promotions to ensure a better return on investment. Those kinds of decisions, which are a mix of experience and market assessment, will definitely be in the human domain for the foreseeable future.

Q: If companies are to invest time and resources into this effort, they’re going to have to know that it’s having some impact at the end. How can you measure the impact of forecast accuracy improvement?

A: Tyagi: That’s a million-dollar question. It’s the first one that any company going into an improvement initiative asks. What’s the dollar value? What are the savings for us? Right now there are some basic analytical methods to determine this. But the right approach is building a model that simulates changes in forecast accuracy and links them to losses, shelf-order stocks, inventory levels and safety stocks. We are seeing some retailers moving in that direction, trying to build a more robust platform for analyzing the real, measurable impact of forecast accuracy.

Q: Can you share some examples of companies that have deployed these methods in order to improve forecast accuracy? What methodologies have they used?

A: Tyagi: There are retailers thinking about the simulation piece. But quite a few have deployed an analytics piece. For example, we have worked with a supermarket chain that used fairly good statistical methods to improve their forecast accuracy. On the promotional forecasting side, they built a complete, homegrown tool, which integrated with their system of record. So they had a package that they were using for promotional forecasting. This complemented their off-the-shelf package perfectly, by providing them with those demand signals that had significant influence on the forecast for that week. To measure it, they set up a metrics framework which allowed them to isolate the impact of forecast-accuracy changes on losses and inventory levels.

In a completely different industry, life-sciences, a company was again using an off-the-shelf package to do its forecasting. We suggested that they try optimizing their parameters within that package they were using. The interesting thing was that after some time, they decided that it was better to devise the forecast outside, feed it into their system of record, then measure the accuracy of that forecast in terms of the same metrics – lost sales and inventory levels.

Resource Link:
TCS

The proper forecasting of retail demand has always required a certain level of intuition. But planners today can no longer rely on a “seat-of-the-pants” approach to calculating inventory and production levels. The consequences of being wrong are simply too high. The new era of retailing requires the application of sophisticated analytical and statistical methods to match supply with demand, as well as determine the impact of getting a particular forecast right or wrong. In this interview, conducted at Gartner’s 2014 Supply Chain Executive Conference in Phoenix, Ariz., TCS’s Rahul Tyagi describes the new, numbers-oriented approach to demand planning and forecasting – one that still requires the presence of humans to make it work.

Q: Why is on-shelf availability such a huge challenge for retailers and their supply chains, after all these years of trying to make it work?

A: Tyagi: There are basically two primary causes. One is external and market-driven, and the second is more internal. There’s been an explosion in buying channels over the last few years. Retailers that typically sold through stores and catalogs are now selling through mobile phones, internet and B2C. That has caused a fragmentation of demand. To capture all of these demands independently has become a big challenge.

Secondly, there’s been a demographic shift. Within five years, the overall buying power will shift from the Baby Boomers to the Millennials. Given the nature of Millennials – very fickle and not too loyal to a particular vertical or channel – capturing their behavior becomes very important. And this is a very well-informed generation. They use all possible forms of information.

In terms of internal challenges, there are a lot of things that need to come together. You have to have a perfect marriage of supply-chain planning and merchandising. The right mix of forecasting, replenishment and merchandising is not happening right now. These are the primary reasons why on-shelf availability is getting more important over the last few years.

Q: Why are companies having such a hard time getting forecast accuracy right?

A: Tyagi: As I mentioned, there’s channel proliferation happening, and data coming from all sources. A big challenge for any company, especially in retail and consumer packaged goods, is how to harness this treasure-trove of data coming from all these sources. How do you synthesize it to generate a good forecast?

With traditional forecasting solutions, there are two elements that are really difficult to get answers on. One is that most of these solutions are driven off of sophisticated items, which require a lot of different parameters that need to be managed actively. You need to have a certain level of statistical science know-how to manage that. Secondly, planners at CPG companies are typically managing forecasts for over a million item-location forecasts a week. As a result, one gets into approximations, rules of thumb, heuristics. The ultimate consequence is that the accuracy starts going down.

Planning, especially in retail, is an entry-level job. So you’ve got people fresh out of college who don’t have the experience or wherewithal to manage these forecast models accurately. This causes forecast accuracy to fall over time.

Q: When you say “science” is an answer, what do you mean?

A: Tyagi: Science means the use of analytical and statistical methods that can enable a better understanding of the demand pattern that underlies a particular item’s behavior at a location. Let’s say we’re talking about a 64-ounce bottle of Gatorade, selling in a store at a particular geography. In order to forecast that well, one has to have a very clear understanding of how the demand pattern for this item is related to the promotional response, and the seasonal response. Is it uniform, or does it vary by seasons? To understand this, we have to use analytics and statistical methods to completely deconstruct these different patterns into elements that can be reconstructed into a forecast model. What we have to look at is a better way of doing this deconstruction and reconstruction.

Q: Were these statistical models and analytical procedures not available to companies before? What has enabled them to take advantage of them now?

A: Tyagi: Some were not available. They have evolved over time, with the need for high-end, rocket-science techniques. Statistical and analytical techniques have been available for a long time – fifty-plus years. But there wasn’t enough technology, hardware or maturity to handle that kind of processing.

These are extremely computational-intensive techniques. Even 10 years ago, we didn’t have the processing power, and the cost of memory was too high. Now we’re able to do things that were unimaginable back then.

Q: Are the people to utilize this technology available today?

A: Tyagi: A lot of the things that a planner is expected to do has been abstracted into these statistical methods. I would say that over the next five years, we will see that become ingrained in the forecasting tools that are offered. What we’ll ultimately end up with is a forecasting tool that automatically does this deconstruction and reconstruction of the demand patterns, and also has the ability to self-correct. Techniques such as machine learning will ensure that the forecast error gets analyzed and gets linked back to the model, in a self-correcting loop.

These kinds of things are extremely computational-intensive. And given both memory and computational trends, it is a possibility now, and will be a reality going forward.

Q: We talk about a shortfall of human expertise. But are we to some extent taking humans out of the equation completely?

A: Tyagi: Absolutely not. Humans have a very important role to play. This is going to be a statistical demand-forecasting tool. It cannot have human intelligence – there’s no way for a tool to know how I change my promotions to ensure a better return on investment. Those kinds of decisions, which are a mix of experience and market assessment, will definitely be in the human domain for the foreseeable future.

Q: If companies are to invest time and resources into this effort, they’re going to have to know that it’s having some impact at the end. How can you measure the impact of forecast accuracy improvement?

A: Tyagi: That’s a million-dollar question. It’s the first one that any company going into an improvement initiative asks. What’s the dollar value? What are the savings for us? Right now there are some basic analytical methods to determine this. But the right approach is building a model that simulates changes in forecast accuracy and links them to losses, shelf-order stocks, inventory levels and safety stocks. We are seeing some retailers moving in that direction, trying to build a more robust platform for analyzing the real, measurable impact of forecast accuracy.

Q: Can you share some examples of companies that have deployed these methods in order to improve forecast accuracy? What methodologies have they used?

A: Tyagi: There are retailers thinking about the simulation piece. But quite a few have deployed an analytics piece. For example, we have worked with a supermarket chain that used fairly good statistical methods to improve their forecast accuracy. On the promotional forecasting side, they built a complete, homegrown tool, which integrated with their system of record. So they had a package that they were using for promotional forecasting. This complemented their off-the-shelf package perfectly, by providing them with those demand signals that had significant influence on the forecast for that week. To measure it, they set up a metrics framework which allowed them to isolate the impact of forecast-accuracy changes on losses and inventory levels.

In a completely different industry, life-sciences, a company was again using an off-the-shelf package to do its forecasting. We suggested that they try optimizing their parameters within that package they were using. The interesting thing was that after some time, they decided that it was better to devise the forecast outside, feed it into their system of record, then measure the accuracy of that forecast in terms of the same metrics – lost sales and inventory levels.

Resource Link:
TCS

By the Numbers: How New Analytical Methods Can Help Improve On-Shelf Availability