Ronan O'Donovan, Product Manager, Supply Chain Applications, ILOG, an IBM Company, starts this presentation at IBM Impact by detailing some common supply chain issues:
Ronan then runs through a number of common problems faced when trying to optimize the supply chain.
Network sourcing optimization
As you can see above, by adding an extra plant you may realize storage and warehouse cost savings.
You should also bear in mind that you can add value and increase working capital by optimizing where you hold inventory.
Service level optimization
Ronan uses the case study of a US-based beauty product retailer with a network of 3,000 stores served out of a central distribution center (DC) that introduces two regional DC's. The objective is to provide better service to the stores, reduce peak congestion at central DC and minimize incremental costs.
The following diagram illustrates the trade-off that needs to be considered:
Decentralizing the storage increases inventory costs at DCs but decreases store inventory costs and transportation costs. Looking at individual product demand variability can help isolate which products should be held centrally, and which regionally:
As you can see, it makes sense to hold inventory at the CDC when there is higher variation in demand.
Optimizing carbon usage
IBM ILOG has taken this into account with its LogicNet Plus tool. The user enters the factors used to calculate C02 emissions associated with various supply chain activities. LogicNet Plus optimizes the supply chain for lowest cost or highest profit and reports total carbon footprint to be used as an additional factor in decision making.
Under the Cap & Trade policy which is currently tabled, if organizations use less carbon than the level set by government, they have the opportunity to sell the difference to organizations that use more than their alloted level. This potential cost/revenue needs to be worked into the overall supply chain strategy. LogicNet Plus will optimize the network for maximum total profit, including the carbon credits bought or sold.
More information on ILOG Supply Chain Applications.
(Guest post by James Taylor of Decision Management Solutions)
One of the best things about being at DIALOG was the opportunity to meet a bunch of ILOG customers and learn how they are making better decisions in their organizations. It seems to me that every one of these customers is, in a very practical way, helping to build a smarter planet. The first group were a set of optimization customers - Alliance for Paired Donation, WinWorks KK and DecisionWare.
Alliance for Paired donation
This organization is making better decisions about kidney donations. Yes, that's right, kidney donations. The problem is that lots of people who want to donate a kidney to a loved one who needs it cannot do so - they are incompatible. Finding matched pairs - where one family's donor matches the other's recipient and vice verse is hard. Add to this the folks who just donate a kidney to help others and you have a situation fraught with complexity and opportunity. Because donations from living donors last longer than those from posthumous donors, the impact of getting this right is tremendous. The savings are pretty good too - dialysis is expensive. The Alliance for Paired Donation uses ILOG's optimization technology to find the pattern of donors/recipients that will result in the maximum number of successful operations. Each time a new kidney is offered, the optimization model runs through this constrained problem and finds the best solution - the one that will improve, perhaps save, the most lives. This works so well that they can sometimes use a single donation to trigger 10 successful transplants. Not a classic use of optimization, but a fascinating one.
Winworks is another optimization customer with a more classic focus - workforce scheduling. Winworks supports retailers, hotels and others with large workforces and handles work scheduling and forecasting. Using the optimization engine they are able to help clients use fewer work hours to get the job done and ensure that work is allocated more fairly to temporary and contract staff.
A Colombian company, DecisionWare is bringing supply chain optimization models to South America, where they are significantly less common. Most companies in South America do not have an operations research or decision sciences group and so are disadvantaged when it comes to competing with companies in countries where these groups are common. DecisionWare offers optimization-based decision support for companies to use in negotiations, with other companies or with unions, and in planning every step of their supply chain.
These optimization customers were all interesting - nice to see a range of solutions using optimization. The other customers were rules customers.
One of ING's operations is managing define benefit pensions. Companies with these pensions typically outsource the management of them completely and ING handles everything from tax and legal issues to answering questions for benefits holders. The complexity of the situation comes from conflicting regulations (department of labor, IRS etc) as well as from the acquisition and divestment of businesses covered by the pensions. Their legacy system for calculating benefits used to take 18 months to change to handle a new customer - not exactly a compelling time frame. Back in 2002 they re-imagined the system and used ILOG's rules to replace the procedural benefit calculation engine on their mainframe with a rules-based one that could be managed by business users. Requiring only the same kind of skill as using Excel, they completely redid the calculation engine. Adding business user-friendly regression testing and test harnesses, they empowered the business to define the rules completely. Even a complex client now takes only 6-9 months to get on-board. A great example of empowering the business to own rules that are complex, but complex in business terms not technical ones.
Travelers uses the ILOG rules products to manage small commercial underwriting. Before they started on this project in 2006 they relied on manual decision making by their independent agents. This meant that two businesses might get the same rate (because they are apparently similar small businesses) even though their risk profiles were quite different. This leads to being adversely selected against because the better risk can find a better price elsewhere leaving Travelers with only the worse risk. Travelers implemented a decision management solution that uses business rules and predictive analytics in concert to automate the underwriting decision more comprehensively. Not only did they boost automated underwriting from 17% to 70% they also increased the number of attributes being considered in the price to around 40. This improves accuracy and risk management tremendously. This system is credited with helping them increase their quotes per agent by 26%, increase the number of agents who want to quote Travelers' policies by 19% and drive an overall increase of 50% in new business quotes. I liked this one because so many of the insurance underwriting cases are for home and auto insurance and this one was for the more complex small business policies, showing the increasing scope for business rules.
Equifax is best known as a data company and provides financial data to companies in the financial industry, telecoms and others. wanted to turn this data into decisions for their clients to add value while maximizing the use of their data and implemented ILOG rules. They have 100M txns a day so performance was critical. These transactions are often short running - the company calling equifax might only have 10-15 seconds and so Equifax is often only given 1-2 seconds.They have more than 100 companies who maintain their own risk management rules and scorecards in these services. this allows their customers, for instance, to change their risk rules the night before "Black Friday"! Something they would never have considered before. This is the kind of agility that can make a real difference - just in time responses to critical events.
I blogged about RCI earlier They use ILOG BRMS for member visibility and inventory segmentation (35 years of contractual obligations), pricing (more than 50M calculations daily), reservations, exchange fees, discounting, communication rules and more for both members and business partners. Combining these rules with a search engine builds customer trust because they never see a property in the search that they won’t be able to exchange or rent - the search finds the properties but the rules narrow it based on the member and their constraints. The rule engine constantly interacts with the search so that all the business rules that are relevant are executed as the search is being conducted - not after the fact but as “part of” the search. Revenue analysts, marketing and IT operations people enter rules and all these rules get pushed into the rule engine. This then sits behind all the channels providing consistent, accurate results.
British Airways first saw the need for rules in their website. Starting as a brochureware site this has evolved into the single largest source of bookings for BA. In the process it has become rather rigid and hard to change. This lack of agility is exacerbated by a lack of confidence in any change being correct which drives a very long and complex regression test before anything can be deployed. To bring business agility to the website while also improving the other systems that interact with customers and partners, BA invested in ILOG's rules platform and built a high-availability, high performance platform for use by the website and other systems. Their model is to develop lots of small, focused decision services that can be easily managed and changed by different business units. by deploying these on a robust platform they are able t deliver agility while meeting their performance and reliability goals. One example is an upsell offer made to those with bookings who return to the site. This has gone from a very fixed offer to a much more dynamic and carefully tuned one, particularly important in the current highly constrained market for business class tickets. The formal release process is now just two weeks from business initiation to implementation, increasing the agility of even core systems, and the flexibility of the rules-based decision services helps manage risk when changes are made. For instance, when they discovered a problem with part of a major release they were able to simply add rules to exclude that particular aspect of the change and go ahead with the rest of the deployment. Before the whole release would have been delayed - now they have fine-grained control delivered by rules at lots of different flex points in the architecture.
Optimal workforce schedules and supply chains, perfectly matched kidney donors, business ownership of complex systems with thousands of rules, agility with the confidence to make changes whenever they are needed and consistent, accurate decisions delivered across channels. Truly better decisions and steps towards a smarter planet.
Erica Klampfl of Ford Research & Advanced Engineering presented an overview of optimization efforts in the automotive industry during a time of increasing uncertainty, in fact downright “distress”, with many of their suppliers.
Using OPL, underpinned by CPLEX as the solving engine, she and her small team were able to meet an eight week deadline for novel analysis involving five separate phases of development. What amounts to a nonlinear Mixed Integer problem was decomposed into tractable portions, still of sizable dimension: a capacity submodel consisted of 1.69 million constraints and 852 thousand variables, while a utilization submodel had 1.78 million constraints and 8.48 million variables. This corresponds to 42 products, 229 requirements, 57 facilities, 37 processes, and 11 assembly plants. Three to four major iterations were required for convergence, requiring about 25 minutes per iteration (10 for capacity, 15 for utilization). Such modeling clearly is not for the faint of heart!
Why did they use these ILOG products? CPLEX’s reputation for solving large MIPs was a starting point. But performance tuning (barrier optimizer, etc) within OPL was important, as was overall ease of modeling setup time, and a facility for performing what-if studies. Perhaps the single biggest key was the need for an interface to Excel to bring the data into the modeling environment.
The results look compelling. The project confirmed the existence of excess capacity, it demonstrated a hub-and-spoke supply chain was optimal, and it identified the need for a new hub in southeastern Michigan. Overall five-year savings are estimated at $50 million, with $40 million of that up-front.
Tom Dong presented this educational track explaining the basics of optimization and its application in different business cases.
When it comes to business cases, ILOG Optimization is most heavily used in these four main verticals: manufacturing, transportation and logistics, energy, and financial services.
What is optimization? A good example is when you sit on a plane and realize that the person to your left and the person to your right has each spent a different amount on their ticket. Why so? The prices have been determined using optimization software analyzing various factors tied to supply and demand, all of which vary over time.
Another example: what price for Xbox consoles and games will maximize profit from Xbox sales? You charge more and the number of sales goes down. You charge less and you cut your profit margin. Optimization software helps you find the sweet spot.
The problems handled within optimization can be categorized by frequency: long-term, short-term and detailed scheduling (very short term):
As this is a broad math model (frequently dealing with optimizing cash and time resources), it can be applied across a broad range of industries:
The one thing that most of these applications have in common is that they focus on problems where there are too many options to be considered manually. Optimization tools allow you to isolate the area of feasibility. This involves looking at scenarios and alternatives as there is often no optimal value - this only exists theoretically. The structure of these optimization models generally looks like this:
When should you use optimization approaches over other technologies? It depends on the sophistication of the intelligence, according to a model developed by Accenture/SAS. At its most basic lies standard reports (often compiled manually) which tell you what happened. More advanced are alert systems which signal when there is an issue that needs to be addressed. And then, at the most sophisticated end of the spectrum lies visualization, business rules and optimization (which all fall under the decision intelligence umbrella in this model):
Note that these systems are not necessarily mutually exclusive. For instance this IBM/ILOG YouTube demo shows visualization, business rules and optimization working in concert.
A common question is around the difference between business rules and optimization. A rough distinction would be as follows:
Rules are more transactional (part of middleware, eg. JRules fitting into Websphere). In contrast, optimization is more about looking at a large set of data and making a decision. This solution is more mathematical, rather than computer-science related.
How long has it been around? Optimization isn't new - in its modern form it has been around since the Second World War. CPLEX itself came to life in 1988.
Some CPLEX figures:
Examples of optimization in action:
Space Management at Hallmark
Transportation Planning at Michelin
Mine Planning at BHP Billiton
Optimization solutions can also bring a number of intangible benefits:
The latest development in optimization and operations research is application development: the use of wizards to build an interface with some components (eg. 'what if' analysis) pre-defined. Once you build this OPL model, you can easily create a user interface (ODM). This can then be distributed to users not specialized in OR (eg. economists for economic modeling). This opens up the technology to a whole new market.
If you want to learn more about operations research, an excellent resource is The Science of Better.
You probably know that FedEx Express is the world's largest express transportation company; they move 3+ million packages daily, using an air fleet of nearly 700 aircraft (flying through ten hubs worldwide) and a ground delivery fleet of over 44,000 vehicles. Would it be any surprise that they face huge planning problems? Kevin Campbell gave us an inside look at how they apply Optimization to their operations.
Their planning goes from ultra-long range (3 to 20 years) for fleets and facilities, down to tactical decisions. Applications using CPLEX include fleet planning, vehicle allocation, hub ground support equipment, aircraft arrival/departure timing, and manpower planning. A typical model attempts to minimize pound-miles, subject to various constraints at hubs and ramps Kevin gave examples of Mixed Integer Programming models with over 100,000 constraints and 1,000,000 variables, of moderate density (8 to 10 million nonzeros). And that is after Presolve removes any obvious redundancies and trivial bounds!
FedEx Express has made meaningful use of the technical support offered by ILOG. Performance tuning on models like these, by using the advanced features CPLEX makes available to power users, helped get solution times down from the weekend range to more like overnight. Definitely a leading-edge model to put into production usage, and it makes me really happy to hear of their success.
(Guest post by Steve Núñez of Illation)
Sandy Carter, in her opening keynote, made a good case for the Smarter Planet and going green, and I don't say that because my (adopted) home state of Tasmania is also the home of the Greens and the Green Party. One of the interesting case studies she described was that of Stat Oil, who ensure the cleanliness of the oceans around their drilling rigs by attaching RFID tags to blue mussels. These mussels close when they are exposed to crude oil, making them the 'canaries of the sea'.
Beyond keeping the planet clean, there are a number of sound economic reasons for going green.
Who would have thought?
Steve Nunez, Principal Consultant for BRMS, Illation Pty Ltd
OK, so it seems I've got a bit of a fixation going with this whole medical theme, but I felt it wouldn't be right to move on without mentioning this unique Optimization customer case study session to be presented at DIALOG 09 by Michael Rees MD, PhD. Not only is Michael a real MD, but his presentation will show just how ILOG CPLEX is helping The Alliance for Paired Donation to save real lives.
The Alliance for Paired Donation finds matches for patients who have kidney failure and have willing but incompatible donors. Kidney paired donation matches one incompatible donor/recipient pair to another pair with a complementary incompatibility, so that the donor of the first pair gives to the recipient of the second, and vice versa. With over 73,000 people on the kidney transplant waiting list in America alone, 70 participating transplant centers, and an estimated one third of all willing kidney donors who have an incompatible blood type with their intended recipient, selecting the best matches among the myriad of possibilities is a monumental yet vital task.
Dr. Rees and his team are working with ILOG optimization experts to tailor their software application for their needs and thus improve the performance and accuracy of paired donation calculations or "swaps." ILOG CPLEX is recognized around the world as the leading mathematical programming optimizer and helps thousands of leading companies to improve efficiency, quickly implement new strategies and increase profitability. This application goes way beyond improving business efficiency, however, and the ILOG Optimization team is justly proud to be helping one of America's most vital non-profit organizations to save real lives.
Dr. Rees' presentation -- Optimization in Kidney Paired Donation -- is scheduled on Wednesday February 4 at 2:00-2:30 pm.