Last week we discussed the recent snafu made by Home Depot during a recent upgrade to its online website on February 1^st that Left Home Depot Customers Running in Circles and chasing their tails due to incomplete planning and testing. We noted that, despite the fact that it was breaking news for some analysts and bloggers, it is not something Home Depot needs to be concerned about. However, as discussed in following posts, Home Depot does have some serious technology-related problems in the doctor‘s view — problems that it may not even be aware of which, if left unchecked, may only amplify as time goes on. However, as discussed in our last two posts, the problem has an easy solution, and it starts by rolling back SARS and retrenching to ARS at the Local Area Depots.

Done right, an Automated Replenishment System (ARS) is the foundation for next generation inventory management. Just ask any inventory management software vendor (including SYSPRO, that was reviewed here on SI in this post and this other post). However, done right requires that the implementation follow a few simple rules

1. Forecasting at the item level is only done across a geography
   Trying to forecast item demand at the depot, and especially at the store, level is like trying to forecast the performance of an individual stock. It’s more or less impossible. For an item that moves erratically, a single purchase can shift the entire demand pattern. However, just like the performance of a broad mutual fund across an industry will be consistent over time, so will a forecast across aggregated demand across multiple Local Area Depots and the stores they serve.
2. Forecasting at the depot level is only done across a short time span
   Again, while the demand for many depots will be predictable with reasonable accuracy, the demand at a single depot will not be predictable with reasonable accuracy over a long period of time. So, the more fine-grained the forecast, the shorter the term the forecast must be for.
3. Forecasts can always be overridden and adjusted by the depot category manager
   
   No algorithm is intelligent. Not even close. An expert, who knows of a(n upcoming) promotion, change in local tastes or fashion, or trending feedback on product quality will always have the upper hand on certain items at the local level. Thus, the system should allow the local category expert to override each and every forecast and pull rule as circumstances dictate. As the rules get more fine-tuned, the need for this will decrease over time, but there will always be a special situation.
4. Forecasts and Rules are reviewed and updated on a regular cycle
   As per our last few posts, the forecasts and rules assume at the minimum a predictable, if not a perfect, world — and the world is never predictable. The system will have to be adjusted as time goes on.

Furthermore, if the retailer insists on maximizing use of the system across each and every one of its locations, including retail stores, then the ARS must be restricted to suggest mode. In other words, there’s nothing wrong with using the rules engine and forecast models to suggest what inventory should be ordered, when, and in what quantity, but the order for any item must be presented for review by the appropriate category or department manager, and corrected if required. Thus, there should be human involvement at the end of each and every order cycle. If the category or department manager knows that only certain items have been presenting (potential) issues, the manager must be able to quickly review the suggested orders for those items, make any mods, and then accept the entire order for submission. In other words, if the system has been doing a good job on the widget category, the category manager shouldn’t be forced to review the widget category on every order, but should always have the option just in case she has an inkling. However, if there has been a continual stock-out of sprockets, the category manager should be able to up the order, and, if necessary, change the rule at the local level (subject to review of the depot manager or category manager if required).

And it must be extremely easy for anyone to report an error in inventory count or historical data at any location so that model, and forecast, is corrected as soon as possible. Without accurate data, the system will never work.

That’s the secret to a successful implementation of an Automated Replenishment System — don’t set it and forget it and don’t force it where it doesn’t fit. Monitor and adjust it continually, and as time goes on, it will get more and more accurate with less and less adjustment for any items with steady trends, leaving forecasters free to focus on fashionable or seasonal items in an effort to truly minimize stock-outs and maximize sales.

In Part I, we began this series with a reference to a recent article in StoreFront BackTalk on how a recent snafu made by Home Depot during a recent upgrade to its online website on February 1^st Left Home Depot Customers Running in Circles and chasing their tails due to incomplete planning and testing. While this was breaking news for some analysts and bloggers, given that it likely won’t even make a blip on Home Depot’s bottom line when all is said and done, for reasons discussed in the post, it isn’t something Home Depot needs to be concerned about. However, as discussed in the second post of the series, Home Depot does have serious technology-related problems in the doctor‘s view — problems that it may not even be aware of which are only going to amplify as time goes on. And these problems are very serious because, as discussed in the third post of the series, they are likely resulting in dissatisfied customers every day in every one of the 2,200 stores across North America. And when you consider that it would only take 3 dissatisfied customers per day per store (which seems entirely feasible in the doctor‘s view) to create 2,200,000 dissatisfied customers over the course of the year, the unnoticeable drops in the bucket become a rip current that could cause some serious damage.

So what’s the problem? As discussed in the last post, it is SARS, short for Storefront Automated Replenishment Systems, which, to the doctor‘s understanding, they have rolled out to the store level across each and every North American store over the past year or two. Advertised by vendors as the ultimate solution to stock-outs and lost sales, as the system is supposed to automatically place purchase orders and replenish inventory at just the right time to insure an item is never stocked out and that the optimum quantity is always on hand, it is sold as a retailer’s dream when, in fact, it is actually a nightmare in disguise. As explained in the last post, these systems only work in a perfect world, but there ain’t no perfect world, and they inevitably break down due to imperfections in the system, incompleteness in the knowledge, and inadequacies of the human operators (including programmers, administrators, and users).

You see, like traditional Automated Replenishment Systems (ARS), also known as Automatic Ordering Systems (AOS), SARS assumes:

• Initial inventory counts are correct
  for each and every product in the store.
• POS-based inventory updates are regular and correct
  preferably, on a regular, daily, basis.
• Damaged merchandise is removed from inventory promptly
  and removed from the system just as promptly.
• The replenishment model is accurate
  and takes into account weekly, monthly, and seasonal variations in demands
• The world of tomorrow never comes
  because the model on which the inventory demand is modelled is supposed to repeat cyclicly with no change, ever.

But they are not Xanadu. And, in the doctor‘s view, the source of SARS is the same as that of the Kubla Khan because:

• A significant number of inventory counts are always wrong … and this number only increases with time.
  There’s a reason retailers typically have all-night inventory counting marathons on a regular, often quarterly, basis. Damage, theft, loss, and human error results in a large number of products having an inventory count that is off.
• Software is buggy and even the internet is not infallible.
  Errors in the POS system can result in the odd transaction not being included in the summary sent to the inventory system, the update file being cut off, or incomplete transmission. Plus, a poorly timed communication failure can result in the POS system thinking the transmission is complete when part of the file was lost.
• Even if it is removed from inventory, it’s often not removed from the system!
  A junior associate may remove the item from the shelf, but forget to update the system. This will cause the inventory counts to get wildly out of whack over time.
• The replenishment model is typically a randomly chosen best-fit model on available data.
  And depending on how much data is chosen, that model could change wildly.
• The arrow of time dictates that tomorrow always comes.
  Next Monday will not be the same as this Monday. Next February will not be the same as this February. And as soon as an unplanned promotion occurs on an unexpected item, something wildly different will occur.

In other words, at the store level, SARS does not work — at least not in an automated fashion. Thus, if Home Depot wants to turn things around, or at least insure that things get pointed in the right direction before it needs to turn things around, in the doctor‘s view, it needs to (partially) abandon SARS at the store level and go back to ARS at the (local) distribution centre level where, when done properly, ARS can be tuned to work like a charm. How? That will be discussed in the next post.