All Things Inventory

Production Planning with Patterns

For decades, a major challenge in managing manufacturing plants in Chemical, Consumer Packaged Goods, Pharmaceuticals and other 'process like' industries (although they maybe producing discrete parts like ball bearings) has been the economical sizing of production runs, variously known as lot sizing, rhythm wheel, campaigns, cyclic schedules and so on. The difficulty is computational due to sequence dependence in changeover times and costs, capacity constraints and routing and recipe restrictions, and other secondary or tertiary constraints.

We achieved a major break-through recently, in a co-innovation project with ConAgra Foods, when we added a new constraint -- a very practical restriction that allows for easy implementation of solutions in the real world -- that insists that the production timing pattern of any item, on any resource, needs to belong to a small set of allowable patterns. This allowed us to create near optimal solutions to large size problems relatively quickly; ConAgra has implemented this already in several plants, and they are seeing 30% reduction in cycle inventory and reduced total changeover costs while meeting demand in a more responsive manner.

An academic paper on this is forthcoming in Operations Research. In the meantime, if you want to test our new algorithm out, please contact us and let us show you how your factories can run better. An enterprise strength module, that works with our EIO suite and connects to ERP/APS systems (including SAP's SNP and PP/DS), called PPPO (Production Planning with Patterns Optimizer) is available for purchase as well.

What is new in inventory models (part 3)?

A central problem in setting inventory targets is that historical data, even when available, may be limited. How to account for this? In a very charming paper published in Management Science, Bob Hayes (in 1969!) suggested the use of Expected Total Operating Cost (ETOC) framework to incorporate the "parameter uncertainty" in the distribution of demand within the news-vendor setting. However, he had to limit himself to a couple of distributions, such as Normal or Exponential. A larger conceptual problem, in my opinion, is how would do you -- the inventory manager -- know what the distribution is! Thus, this work, although cute, has been essentially useless for practice. 

It turns out that the ETOC framework is worth keeping. To create something useful, we (Professor Bahar Biller and graduate student Alp Akcay, and myself) decided to imbed a flexible family of distributions, called the Johnson Translation System (JTS), that can fit any possible first four moments to any available data set, into the ETOC framework. With this approach, the need for the inventory manager to guess at a demand distribution (or arbitrarily choose one) is eliminated. Our experiments, several with actual data from SmartOps customers, show that we can improve the inventory productivity by over 15%, while guaranteeing the desired service levels with even greater confidence. One more practical hurdle ---- limited historical data -- has been overcome. That is the value of academic research. This work is forthcoming in Manufacturing & Services Operations Management Journal, the flagship journal of our research community.

PS: To honor Bob, and because we liked how it sounds, we decided to call this new inventory policy: HIP (Hayes Inventory Policy), although it was not originally suggested by him.

SmartOps in New Zealand!

The year 2010 was a great year for us. It was our 10th consecutive year of growth, every year since our founding, and 7th year of profitability. In fact, 2010 was our second most profitable year, as we continued to serve our returning customers (such as Kohler, Danfoss, Lexmark, Estee Lauder, Dupont, ConAgra, Kellogg's, J&J, Celestica, Sysco, Clorox, Unilever, Eastman, Pfizer, Merck, Polaris, Honeywell.....I can go on and on!) as well as added many new ones--PPG, Dow, Lubrizol, Campbell's, CSL-Behring, Medtronic, Baker Hughes, and Vaillant, among others. Several of these were joint deals through our reseller channel partner SAP, a couple were executed jointly with our consulting partner J&M, and we are looking forward to continued profitable growth in 2011. We also created a joint agreement with Accenture in 2010 that should further help our growth and global reach. Our implementations in previous years have taken SmartOps to several continents; our new customers are taking us to some places we have been to, like Brazil, Spain, Italy, Thailand and England, and also to new places like China, France and, yes, New Zealand. And, Go Steelers!

The Lever of Riches

One of the delightful books I read some years ago is Joel Mokyr's The Lever of Riches (Technological Creativity and Economic Progress). A recent book by Brian Arthur -- billed as "the most important book on technology and the economy since Schumpeter" -- is an original (and extremely lucid) analysis of how we as a society have evolved from stone tools to  iPads. I am very pleased that Brian will be the keynote speaker at our annual Forum this year in Chicago (September 21).

This is SmartOps' seventh Forum, and the third one in Chicago. In 2005, when we first did the event in Chicago, the keynote speaker was Steve Levitt (before he was this famous and unaffordable) and he gave an engaging and funny talk on topics from his book Freakonomics. True to his self, and consistent with the attitide of being a tenured professor (at University of Chicago), he not only ignored my suggestions to skip potentially 'offensive' topics, but actually opened with them! Last year, again in Chicago, we had Geoff Colvin, the author of The Upside of the Downturn, as the keynote speaker and it was a wonderful talk. I expect that Brian's talk this year will bring refreshing clarity at a time of great uncertainty and confusion as to what actually drives the world economy. (FYI: Eric Schmidt, the CEO of Google, is a big fan of Brian and has applied Brian's ideas at Google.)

Of course, like at our previous Forums, there will be speakers who will present how thay have benefited from the new science and technology of enterprise inventory optimization, and how they are generating predictable margins and delivering stable and consistent service levels  in their supply chains even in a time of unusual uncertainty and volatility. This year Dupont and Danfoss will be highlighted, and I expect thay will be received as well as last year's presentations (from Celestica and Wyeth/Pfizer). There will also be an opportunity to test drive our technology through our educational sessions. As always, we will discuss new innovations and modules with the customers in our SmartOps User Group (SUG) sessions.

This year's Forum is co-sponsored by SAP and Accenture. SmartOps is a solution extension of SAP; Accenture and SmartOps have a strategic partnership to help clients in P-SIOP, Profitable Sales, Inventory and Operations Planning, which combines the twin benefits of enterprise inventory optimization and service level optimization to dramatically enhance the traditional S&OP process.

For book lovers, let me also suggest (Pulitzer prize winner) Thomas McGraw's Prophet of Innovation (Joseph Schumpeter and Creative Destruction) and Timothy Ferris' The Science of Liberty (Democracy, Reason and The Laws of Nature).

 

Service Level Optimization

Closely related to inventory optimization is the selection of the service level target itself (by item, or customer or location). A situation that is commonly encoutered involves meeting a group service level target (across items in a product category) with minimum investment in inventory. In this situation -- since the costs of various items are different, the volumes of their demands are different, the forecast errors are different, their lot-sizes are different, their lead times and the uncertainties in lead times can be different and so on  -- it is possible to determine item-specific service level targets that are higher than the median for some items and lower for others, that requires less aggregate inventory investment (when compared to a solution that has service level targets the same).

A variation of the above optimization problem is also common. Instead of specifying a group target, if data on profit margins and likelihood of lost sales (when an item is out of stock) are known, then one can find item level service level targets that maximize the total profit of this product category.

For folks interested in learning more about service level optimization, there is a webinar on August 26th. Please visit www.smartops.com/smartops-events to register.

What is new in inventory models (part 2)?

Something entirely different from the supply chain inventory models I discussed earlier has to do with inventory elements in video games.

Unlike early generations of video games that had advertisements 'hard coded' in the CDs played on individual consoles, the current generation of video games are played over the internet, and the games are shipped from the game developer with blanks where ads can be placed in real time while the game is being played. This is called scheduling of dynamic in-game advertisements, and is also the title of a paper (I am a co-author along with John Turner and Alan Scheller-Wolf) that was accepted (just last week) for publication in Operations Research.

Massive Incorporated (acquired by Microsoft, after Google acquired Adsense) and CMU have worked collaboratively to develop algorithms to optimally schedule ads. My friend Katherine was its COO, and there is a HBS Case on her and Massive. David  -- the CTO -- and his staff, in particular Frank, were the folks we worked with on a day-to-day basis. What is the optimization? The sales force of Massive negotiates contracts from companies such as Pepsi that pay a ceratin amount of money for a certain amount of impressions (of their ads), in specfic games (targeted at specific demographics) at specific times over a pre-defined time horizon. There are severe penalties if the impressions are under-delivered compared to contracted amount, and there are restrictions to avoid saturation as well as conflict (with competing brands) and so on. Now the demand for these ads is stochastic -- it depends on who is playing, in what part of the world, at what times and how many levels they manage to progress to and so on -- and so the algorithm aims to place the ads appropriately.

I expect that an on-line version of the paper will become available soon in the Articles in Advance series. I hope you enjoy it.

 

So what is new in inventory models?

Inventory modeling in academia -- specifically, the analysis of discrete time models for production and inventories, deterministic and stochastic, multi-stage and single stage, stationary and non-stationary, uncapacitated and capacitated -- is about 60 years old. (I have been at it for 20 years now!) In fact, SmartOps Enterprise Inventory Optimization (EIO) product has benefited from this rich heritage (as the underlying framework of SmartOps EIO is a discrete time, finite horizon, non-stationary, capacitated, multi-echelon model with batch size restrictions and so on).

What are some recent research publications that are useful? Let me highlight one now -- co-authored with my PhD students Nihat Altintas (now at Credit Suisse on Wall Street, and we are now using our inventory modeling skills to improve high frequency automated proprietary program trading algorithms) and Feryal Erhun (now a faculty member at Stanford working with Intel on risk management and other topics) -- the study of quantity discounts under demand uncertainty: How should a supplier optimally design a discount scheme to nudge a buyer towards buying closer to truck load quantities, anticipating that the buyer (facing uncertain demand) will follow an optimal policy himself? This research was based on interactions with Heinz as well as Shaw's Supermarkets, and the paper is published in Management Science. If the optimal policy of such a model is too hard to implement in reality, what are simplified policies that can be implemented that perform reasonably well? How are quantity discount schemes related to minimum order contracts? What are the upstream consequences of a discount scheme in terms of propogation of uncertainty?

One very interesting new phenomenon we discover here is the opposite of the very famous bullwhip effect. As the end-item demand at the customer increases, the volatility of the orders to the supplier decreases! The much maligned batch sizing in the bullwhip literature may not be all evil after all. I hope you have fun reading the paper and enjoy the various insights (and some are perhaps initially counter-intuitive).

 

Skin in the game

The first article in the first issue (of this year) of Operations Research describes a reward sharing collaboration between Deere and my company, SmartOps. Between 2005 and 2007, Deere and SmartOps worked collaboratively to simultaneously improve the logistics costs, inventory investment and lead times to dealers. This was done by tailoring the logistics strategy to the season -- in this case twice a year -- and so achieving over $10 million in savings additionally over all the other initiatives that were already in place. This was done through the creation of appropriate operations research (OR) models, gathering of sufficiently accurate data in a timely manner, deciding on what variables were going to be fixed and which were allowed to be optimized, testing the recommendations via simulation, negotiating contract terms with logistics providers, getting the IT infrastructure prepared for change and then tracking the performance of the associated implementation with great scrutiny.

Deere agreed to this reward sharing scheme to help me continue to build SmartOps by earning sizable revenue based on our expertise -- rather than by going to capital markets to raise money through equity offering or assuming debt -- and only get paid when Deere actually achieved the additional value. I thought this was not just a fair deal but a great deal for an entrepreneur like me as I could keep building SmartOps through the volatility of the economic conditions and the vagaries of the enterprise software market. By putting some skin in the game, we were able to work collaborately and effectively with aligned incentives to create immense value for both companies.

This was not the first project between Deere and SmartOps. In fact, it was the third. Our first project that began in 2001 helped save $1 Billion at Deere's C&CE Division (see Interfaces Jan-Feb 2005). A second project (in 2004) worked helped Deere manage the complexity of their broad product line and saved them tens of millions of dollars (see Operations Research, July-August 2007). This long term successful relationship was central to the reward sharing agreement, both in its genesis and in its execution. I personally cannot thank Deere enough for being such a good customer, partner and friend.

Deere is not the only long term partner of SmartOps. An earlier (and so longer) relationship is with Caterpillar that began even before I founded SmartOps (on March 8, 2000) when Caterpillar approached me at CMU. The first project with Caterpillar (in 1997-8) was the development of a rapid-response supply chain (see Operations Research, March-April 2000) that was also featured in FORTUNE (October 31, 2000). A second project focused on multi-stage inventory optimization using SmartOps EIO product (like in the case of the first Deere project) that helped stabilize product availability while reducing inventory investment by 15%, reducing average lead times by 20% (and its variance by 50%) and increasing revenues (due to reduced lost sales) by 2% (see Interfaces, July-August 2006). A third project (2006-8) between CMU, SmartOps and Caterpillar was in supporting a bundling and price sheet strategy; this is being rolled out in North America and Latin America right now. We presented the details of this project at MIT last summer at the MSOM Conference; a paper for publication is being written right now.

Long term relationships between companies, and between companies and universities have led to mutually beneficial results. I look forward to many more years of such successes, not just with Deere and Caterpillar, but also with other SmartOps customers such as ConAgra Foods and Kellogg's.

 

RNA interference increases shelf life of fruits and vegetables

I fell in love with RNA interference (RNAi) the very first moment I read about it, for its elegant simplicity and its potential effectiveness in a wide variety of areas. Today I read an article -- based on a recent publication in the Proceedings of the National Academy of Sciences -- that described how the shelf life of tomatoes (normally about 15 days evidently without refrigeration) is extended to 45 days by silencing genes that drive ripening of fruits and vegetable (and so slowing down their ripening). Research is underway to similarly silence the ripening genes in papaya and banana. This, if approved for widespread use, can help countries like India tremendously where refrigeration is prohibitively expensive and demand for food far outstrips supply.

In strong contrast to genetically modified (GM) seeds and foods (like Bt Cotton or Bt Brinjal), no foreign or new gene is inserted in the case of RNAi and so this technology is expected to have less social and political obstacles towards approval.

Separately (and completely unrelated to this research on fruits and vegetables), my charitable foundation (RAGS Charitable Foundation) has been funding research in RNAi in the hope of finding a therapy for a rare neurological disease (similar to HD and Parkinson) called spino-cerebellar ataxia (SCA). In this context, I have had the joy of communicating with Andy Fire (co-winner of the Nobel for the discovery of RNAi along with Craig Mello) and Phil Sharp (another Nobelist, although not for RNAi, and co-founder of Alnylam Technologies that is looking for RNAi therapy for HD and Parkinson). I am optimistic that over the next decade several great practical contributions of RNAi will become available.

Fix the Mix of Inventory

What is the minimal set of red wines to have at home? The objective is to have some everyday wines and some good ones (for special occasions). The point is zero inventory is not an option.

For the everyday, I decided on the Pinot Noir. But which one? From Oregon or Burgundy? I decided on one from New Zealand, from Marlborough region, not only because it was good and cheap, but also because it had a screw top. What about something good? There is a Chateauneuf-du-Pape (2005) that is decently priced and has been rated 95 by Wine Spectator, and many good choices of California proprietary reds. I decided to constrain the search by insisting that this also be a screw top. The 2004 Pillar Box Red (57% Cabernet, 32% Shiraz, rest Merlot) from Australia is very good. Try it out.

Technically, this is not minimal since two countries were represented. If I removed the screwtop constraint, then a one-country solution becomes feasible. For this 'problem', I chose a Burgundy for the every day wine and the Chateauneuf-du-Pape for the special occasion. Now one may complain that this is not minimal since two regions of France were involved. Then, a one-region solution probably needs to come from California, and in the limit, one may want a one-vineyard owner solution and so on.

Suppose now we look for a solution for a maximal one-case problem. For the everyday, I picked Pinot Noir (from New Zealand), Malbec (from Argentina), a proprietary red from Chile, Shiraz-Cabernet (from Australia), Grenache-Tempranillo (from Spain), and Barbera (from Italy). For the special occasions, I picked the Chateauneuf-du-Pape (France), Barberesco (Italy), Cabernet (California), Shiraz (Australia), Red Zinfandel (California) and Cotes-du-Rhone (France). Look at all the grapes, countries and continents covered in this proposed solution.

How many bottles should one have in the cellar? Well, let us look at the various parameters that may have a bearing on the answer. Since I am not a collector, the goal is to turn the inventory sufficiently quickly. Thus, it depends on how much inventory is depleted per week, and how long you want to keep a bottle (on average) in the cellar. Suppose the weekly base rate is 4 bottles (of everyday wines, but can fluctuate between 0 and 8), and couple times a month (assume 4 weeks long) there is a special occasion (just family, or a small group that depletes 1-4 bottles) and once a month there is a bigger party (for which some new wines may be purchased, some guests bring (good red) wines anyway, and typically depletes 10-16 bottles). The maximum 4-week depletion is 56 bottles, and so having 5 mixed cases (60) should be sufficient if you are able to shop every 4-weeks.

Now applying the maximal one-case solution above identically to 5 cases is the feasible solution I implemented this month. The turns are about 10-12 (per year), and the service level (measured by stockout probability or fill rate) is 100%, assuming perfect substitutability. Now, since folks who come to our parties have preferences, and not all wines are perfectly substitutable, the variety implicit in this solution forces me to have extra inventory of certain lower cost, but acceptable favorites. So I added a couple of mixed cases, one of my minimal screw-top solution and one of my France-only solution discussed above.

A lot of inventory target setting in Fortune 1000/Global 2000 companies have the similar dynamics that I describe above in Wine cellar inventory management. Demands fluctuate, there are lumpy orders, new products become available and replenishment frequency is periodic (and additionally there are stochastic lead times, supply variation, capacity constraints, bills-of-materials, multiple echelons, common components, engineering changes, holidays and events, batch sizes, pricing discounts etc etc). At the end of day, the inventory planner in a global supply chain is attempting to keep the best mix of inventory, at the right levels, at the right location for the right time period to maximize profitability through the right level of customer satisfaction (with low total cost of inventory investment as well as expediting and so on). And so, I founded SmartOps Corporation that provides the Enterprise Inventory Optimization solution to help fix the mix of inventories on an on-going basis in complex supply chains.