Automated Ordering: What are the possibilities?
By Al McClain
I recently met with SAF – AG of Switzerland, a company that promises that great things are in store for retailers who adopt automated ordering. Not being an IT guy, I was curious about what the real potential for automated ordering is, and what the stumbling blocks may be.
SAF’s system, under the product names SAF SuperStore and SAF SuperWarehouse, has recently been introduced to the United States. In a nutshell, the company promises to automate 95 percent of retailer orders, providing benefits such as reduced labor cost, improved inventory levels, lowered out-of-stocks (reduced to 2 percent instead of 7-10 percent, which they say is typical), and reduction of lost sales. They say “micro-forecasting,” meaning item-specific replenishment, provides the basis for this capability. The system takes into account issues such as seasonality, promotions, price changes and holidays. At the warehouse, there is a parallel system that helps in optimizing logistics, reducing inventory, reducing out-of-stocks, improving truck loads and adjusting the ordering process to account for supplier restrictions.
The way I understand it, many retailers in the U.S. currently take inventory and make orders based on a combination of computer assistance and manual work. Two critical factors that are hindering the adoption of automated ordering, according to BrainTrust panelist and RAM Communications Managing Director Ron Margulis, are shrink and random-weight products. Since shrink can be as much as 3-5 percent of inventory, this can be a real problem. With random weight products such as produce, the issues include mismatches between shipping and selling volume, in-store processing waste and, again, shrink. In addition, poorly trained cashiers will combine bags of similarly priced fruits or vegetables, resulting in inaccurate inventories.
SAF promises to address the shrink issue with a sophisticated exception tool that will enable a manager to intervene on about 5 percent of the order lines. Random-weight products are difficult from a perpetual inventory perspective. The unit of measure for most items is quantity in selling units. Random weight products also have the criteria of weight by which product is priced and sold. Random weight can be dealt with by continually updating the average weight of the items which requires significant discipline at store level. SAF says that random weight products account for a very small percentage of the items sold and should be considered last, if at all.
SAF’s European customers include the METRO Group and Woolworth Germany. In the U.S., Hannaford Brothers and Holiday Quality Foods are using the system, and two major retailers are in pilot.
The Founder and CEO of SAF AG, Dr. Andreas von Beringe, says what they do is “replace time consuming order review and order placement with the first truly automated store ordering system. We run thousands of items in a few minutes.” A customer, dm-drogerie markt, which has currently 1700 drug stores in eight European countries, says that they have been able to automate nearly 100 percent of their orders using the system, enabling customers to find more products on the shelf.
Discussion Question: What is the real potential for computer automated ordering? Should retailers get at least a pilot going now, or “wait and see”?
Ron Margulis believes that the random weight issue needs to be fully addressed for an automated re-ordering system to have storewide benefits. Also, merchandisers/buyers
at many retailers still have the culture of “I can order better than any machine”, so there is an institutional hurdle that must be defeated.
In my opinion, there is currently no ideal replenishment software system. Current systems rely on an initial inventory and subtract items that go through POS to determine product movement and reorder at a target level. What happens to the system when the product is in the backroom and not on the shelf? The system thinks that the product is on the shelf and is not selling because nothing is going though the POS. Nothing will happen until some person realizes that the product is in the backroom and restocks.
The systems will get better when we have a RFID tag on each salable SKU, and then we’ll know the product did not pass from the loading dock or backroom to the selling floor and then trigger an alert to stock the product.
Everything under the sun has some potential …BUT. As the introductory write-up states, there are two companies in the U.S. already using the SAF system (Hannaford Bros. and Holiday Quality Foods) and two major retailers are in pilot. So let see if something positive develops.
There are several other systems that suggest reorders and allow exception reporting for manual changes and approvals. Ordering and reordering systems, including forecasting methods, have existed for store locations, warehouses, wholesaler and manufacturers for over 30 years. Some of the software, such as IBM’s Inforem, predates point of sale terminals.
As has been noted, automated reordering is not a new technology. Microforcasting, while a new-ish term, is not a new feature. Services exist which will take a retailer’s own data and analyze sales patterns at the item level, returning back to the retailer the appropriate item selling “profile” or forecasting pattern.
In the non-perishables, non-food sectors, the key barriers to adoption have been cost and management time. Assortments vary over time, and with the shift toward local market assortments, they vary in large numbers. As new items are introduced, sales profiles must be determined from existing SKU’s. This is an inexact “science,” resulting in a lengthy period of adjustment and review…except when the “new” item is simply a new version of an existing item. The management time to identify, fine tune, and adjust these new profiles has led to the adoption of more generalized templates.
Many systems “self adjust.” The majority do so by using some version of an actual vs. forecast error factor applied to future forecast periods. Because trends do not stay constant, but instead fluctuate with competitive behavior and consumer shifts, this really means a constantly changing series of forecasts at the item level. The pricing model that most outsourced profiling services precludes doing these forecast adjustments as often as theory would require. Because of the cost, once again, time weighs in as someone must determine the SKU’s to receive weekly reforecasting, bi-weekly and so forth.
This isn’t new. Perhaps SAF has addressed the business process and reforecasting cost issues with its system. I don’t know.
Computer Assisted Ordering is a key component for large retailers, however, the only way it will truly be successful is if it is fully integrated in to additional optimization solutions. It’s no use forecasting and then deploying accurate stocking levels to store, only for a generic planogram to be sent out which doesn’t reflect adequate units on shelf; this just results in large amounts of stock in the back room or gaps on shelves.
Price, Advertising, Promotion, Assortment and Space planning need to interact in order to make the whole category review process seamless. CAO fits nicely into this vision and can certainly benefit from a market which seems to open to optimization solutions currently.
As long as a rapid ROI model can be proven, I think retailers will start to act quickly in relation to introducing this on top of existing in-house ordering systems.
In the arena of keeping retail shelves stocked, computers are providing raw data, but people are still needed to do the heavy lifting. There is definitely an opportunity to improve information systems in the world of inventory management.
When this happens, inventory professionals that are now monitoring SKUs at the store level could be deployed to make improvements computers can’t, such as working with buyers to provide more accurate initial forecasts on new products, working with suppliers to shorten lead times, thinking of ways to improve the logistics or analyzing the data to suggest changes that could improve space productivity.
What is disconcerting here is the idea that once again machines can do something better than humans. What about damaged packages? Spoilage for produce and meat? Can the automated system be predictive during promotional periods? What about system failure due to power outages and weather? You are still going to need someone present to watch and monitor any good system.
I think Ron is being too negative here. Shrink is trivial in a re-ordering context. If I have 24 cans of Tomato soup on the shelf and shrink is 5%, we’re talking a can. Gee – build the system to trigger when you are low plus one can. And yes, random weight is a problem for automated ordering systems. So what? If a system can fix 80% of a store (I don’t know the actual number), isn’t that an advance?
Automatic store ordering has been a dream of food retailers for years. Retailers started with modification of the old IBM Impact system in the early 90’s. Since every attempt failed, the process was re-named Computer Assisted Ordering (CAO).
There are many well documented reasons for these systems failures. First is not having an accurate retail inventory. The Retail Accounting approach does not require it, thus additional effort and cost are required to achieve it. To have an accurate retail inventory requires accurate scanning; something not being achieved. All areas of the store must have the discipline and process executing for inventory adjustment. Examples here are taking damaged product off the sales floor and telling the system or taking supplies from the sales floor. Additionally, continued cycle stock counting is required to include shrink. This cycle stock process requires more labor than order writing.
Second is the area of promotions. No one has every created a close to reasonable promotional forecast at either the store or warehouse. And then there is the issue of random weight.
The example used to justify this system is chain drug, not retail food. The differences are significant. First is the number of SKUs. Second is the velocity. Third is no processing or random weight, etc. Many retailers do have accurate retail inventories, but in every case the number of SKUs is smaller and the velocity much lower.
Automated Store Ordering has many advantages. I just finished a benefits article on CAO, based on recent projects. To read, click here.
A latest development not mentioned yet in the forum is Store Order Leveling by simply manipulating input parameters of the automatic replenishment system. The system uses free shelf space of set-to-minimum order algorithms, or redundant stock levels of set-to-maximum algorithms. Orders can be leveled to an ideal distribution over days of the week.
We have made many simulations for set-to-minimum and set-to-maximum order algorithms, and these simulations are tested in real-life situations. It is now possible to generate almost any order leveling distribution for any actual turnover distribution. It is also possible to shift automatic orders further to compensate for manual – non-leveled – orders.
As stated above, there is not much new technology here. Retailers have been doing this for over 25 years. Grocers have lagged behind, so SAF is putting a new spin on mature technology to try to entice the laggards. One of the reasons Wal-Mart, et. al. have been taking market share from traditional grocers is that they have deployed this and other technologies many years ago enabling them to operate more efficiently and therefore at a lower cost.
I am not familiar with the SAF-AG products, but automated ordering systems that make item level replenishment decisions have been used successfully for at least 30 years. Particularly at the warehouse level, automated replenishment is pretty much “a given” for fast moving consumer goods. Store level is still evolving and although Ron is right about the perishables and commodities issues, many of the prepackaged offerings in produce and meat have turned these traditionally commodity categories into branded ones. But the requirements of warehouse and store replenishment applications are fundamentally different.
The warehouse replenishment objective is to minimize acquisition cost while maintaining a desired service level. Warehouse replenishment systems must understand vendor ordering constraints and work to minimize acquisition costs (product, handling, carrying, and transportation) costs while all the while meeting service level objectives. This is usually done by achieving a balanced time supply of a vendor’s product line so the purchases consistently meet minimum quantities and acquisition costs are minimized. Multi-sourcing (the commodities issue) just introduces another level of complexity.
Store level replenishment is different because many of the product cost decisions are out of the hands of the retailer (they were negotiated by the buyer). Store replenishment is primarily concerned with handling costs. The store also introduces a new factor into the replenishment process. Stores must consider “presentation stock.” Usually this is some factor of the case pack to reduce the number of partial cases that must be handled at store level. Since 90% of the items in a store move less than a case a week and most stores have an order lead time that is less than one week, the whole replenishment exercise for all but 10% of the items is boiled down to establishing a reorder point that can sustain the store until the next delivery.
The 10% of items that move more than a case a week are often the promotion items. I agree with others that promotions can be a challenge but there are four factors that seem to determine the impact of a promotion on demand: category sensitivity to promotions (cigarettes vs. paper towels), depth of the price cut, external advertising (super blow, blow, or line item), and store display (end display, signage, etc). By knowing each of these factors you can make a pretty good prediction of a promotion’s impact on case movement.
Retailers, and especially supermarkets, have traditionally managed orders and shipments. The stores ordered and the buyers and warehouse made sure the product shipped. The focus must change to inventories and forecasts. Stores must constantly monitor inventory, with cycle inventories on high shrink categories to provide interim updates. They must also be tied into the forecasting systems for both regular turn and promotional activity so that the forecasts are accurate. With both inventory and forecasts correct, replenishment becomes easy.
I have worked with SAF for a number of years and am baffled by some of the responses above. This product was built from the store perspective and not from the warehouse perspective. Almost all of the replenishment solutions on the market today came from the Inforem base which is a warehouse centric product. The SAF solution is cost optimized and deals with attributes such as weather or the forecast of weather at the item level for each sku/store combination each time it is run. The SAF folks have a truly unique product which works extremely well in a hardgoods or basic softgoods environment.
Have we got to treat this as another potential war, with only an either or solution? Why can’t automatic ordering be used in part or in certain departments, supplemented by that good old human brain thingy that so many try to ignore. I’ve had to write to a bank today, for the fourth time, to try and get a real live person to make their computer stop churning out letters that I have repeatedly written and told them to stop sending. Anecdotes like these abound in every industry. Food retailing could be truly innovative by using a mix of (wo)man and machine.
The commentator is misguided in their perceptions. Inventory and reordering is a largely automated process at most larger retailers. The inventory function is a critical balance, since the product count is different than the electronic “scanner” count. This is a result of spoilage and damage which occurs at store level and because of shrinkage (theft). None of these can be accounted for in a totally automated system, so inventory is required to determine the difference between the scanned count and the actual count. Without this, the systems are wrong and the electronic ordering system cannot manage the product needed to eliminate out of stocks at store level.