Managing Sewage Demand and Holiday Demand

When it rains it pours – that is, sewage overflows into streams, rivers and lakes and then into our beaches and drinking water (yuck!).   (NY Times 11/22/09)  For example, cases of serious diarrhea at one Milwaukee hospital increase when local sewers overflowed.

The problem with our old sewage systems is that they are capable of handling normal “loads”, but when it rains, the demand on the system exceeds capacity. With no place to inventory the excess water, it is spilled into local waterways.

So what is the solution when peak demand exceeds capacity? The knee jerk reaction is to increase capacity. But this can be foolish – it can be very costly to increase capacity when it will be used only at peak demand times. A better strategy is to smooth out the demand peak.  Of course, we can’t control rain, so that won’t work. But we can slow down the flow of rain into the sewage system by how we design our cities. In particular, planting trees helps to absorb the flow of water (maybe we can interpret this as an increase in capacity). Planting organic roofs can also slow down the flow of water – instead of rushing off the roof, it drips from the roof.  Or, if we could design sewage systems so that rain water was processed in a different system than toilet water, then the peak demand problem is solved - peak rain water can be safely spilled into the river and it would be unlikely to have a coordinated demand peak of toilet water, i.e., separate smooth demand from the variable demand. (I never did believe the stories about the problems induced when too many people flush during a Super Bowl commercial.)   

So what is the connection between sewage and holiday demand? Retailers face the peak demand problem in spades. The following graph plots sales of general merchandizers from 1992 to 2008 and clearly illustrates the annual holiday spike in demand

An interesting feature of this paper, contrary to what you hear in the press, is that the end-of-year spike is actually getting smoother. To illustrate, the following graph shows the % of annual demand that occurs in November and December:

From a peak of 25%, the fraction of annual sales occurring in November and December has been steadily declining, and now is about 21%.  Why is this? It could be that consumers want to smooth their consumption (they can’t wait for that holiday present) or it could be that retailers are encouraging this demand smoothing (via pricing). 

Another change in the data is the “back to school effect” - August’s sales relative to September’s sales has been showing a steady increase. Back-to-school is now the mini-Christmas of the year.

H1N1 Production and Forecasts

The following table provides a summary of the H1N1 vaccine forecasts and actual availability:

Date	Forecast (million doses)	Actual
7/30/09	120 by October.
9/12/09	50 by Oct. 15,
9/26/09	40  by mid October
10/17/09	28-30 by the end of October.	11.4
10/26/09	30 by the end of October
10/28/09		23.2

Why the error in their forecasts? My favorite quote (NY Times Oct 25, 2009) is

Dr. Thomas R. Frieden, the director of the Centers for Disease Control and Prevention, “We really thought that having five different manufacturers would buy us some insurance, that they wouldn’t all have problems.”

The implicit assumption in this quote is that yields would be independent across flu manufacturers.  This may be true if yield primarily depends on manufacturing choices that could differ across the 5 manufacturers. But the manufacturers all used the same seed stock (the virus injected into eggs that is suppose to replicate within the eggs to make the vaccine).  Consequently, if the seed stock isn’t very good, it doesn’t matter if you have one manufacturer or 100! If you are going to diversity your risk by choosing multiple suppliers, then you should make sure that their yields are uncorrelated. (On a related point, there was no shortage of nasal spray vaccine, because they had a much higher yield … and used a different seed stock.)

There are some other reasons for the errors.  In July the forecast was based on an assumption of a high yield. However, the actual yield would not be known until the first batches were completed in August. Therefore, it seemed premature to put any faith in the July forecast without better information.

So why were the September forecasts wrong? The government basically asked the manufacturers’ for their forecasts, and not surprisingly, they got optimistic numbers. I am not saying the manufacturers lied. Instead, they probably thought it was possible that they could improve yields quickly enough to make the numbers. For example Sanofi Pasteur’s initial yield was 1.5 doses per egg and they did manage to increase it to 3 doses per egg.

Some of the shortfall was due to shortages in parts. For example, MedImmune, maker of the nasal vaccine, had more vaccine than it could put into nasal sprayers (their yield was apparently 5 times higher than expected), despite having their supplier of nasal sprayers work 3 shifts, 7 days a week.

Finally, one of the manufacturers, from Australia, satisfied Australian demand first.

The lesson from all of this, dare I say, is “don’t count your chickens before they have hatched”! Production yields are uncertain, and the government could have benefited from learning more about what determines those yields, when information would be learned about those yields and relying less on manufacturers’ forecasts.

It isn’t your father’s PC industry anymore

My first PC was an IBM PC with an Intel 8088 microprocessor, two floppy disk drives and a whopping 64K of RAM (not 64MB or 64G, the 64,000 variety) – and it cost about $4000 in 1985 (but my father worked for IBM so we got it for the employee discount price of something like $2700). HP is currently selling a laptop through WalMart for $298 (or $148.47 in 1985 dollars, http://data.bls.gov/cgi-bin/cpicalc.pl). 

The PC industry has gone through many stages in which one firm was on top. Apple started it, then IBM took over. IBM tripped in the early 90s and Dell took over. Dell started to stumble about 5 years ago and now HP is on top as we see in the following graph reported in WSJ (HP wields its clout to undercut rivals, 9/24/09):

So how is HP able to do this. First, they are working with small margins, razor thin 4.6% margins. Next, the article gives some other clues to their strategy.

 (1) “Simplifying the specifications of the product”

i.e., reduce product variety so that contract manufacturers can have higher volumes and thereby offer lower prices. This is a standard recommendation in an OPs class.

(2) ”By getting orders in earlier, H-P could save on component and manufacturing costs, which are cheaper if they’re ordered far in advance.”

This line is intriguing. If component prices are falling, then ordering early is a disadvantage, not an advantage. This suggests several possibilities. First, component prices may not be falling rapidly and HP is better off giving suppliers a long lead time to get an advance purchase discount from them. Second, component prices are still falling but it is cheaper for HP to take on that risk than to let the suppliers take on that risk – i.e., if they take on that risk then they have to charge more, which is passed on to HP.

As I said, it isn’t your father’s PC industry anymore. What makes me think it could be entirely different in another 5 years?

A great operations management blog … check it out!

Our friends at Kellogg, Marty and Gad, are writing their own blog on operations management …

And here is the link:  http://operationsroom.wordpress.com/

Their style is very similar to ours (and maybe better!) – comment on current events as they pertain to operations management, especially the teaching of operations management.  This will be a great resource for people looking for examples to illustrate points (or for people looking for new research ideas).

Starbucks goes lean to pull out of its slump

In June Daniel Corsten pointed out to us a very nice article about Starbucks and how they are continuing to rethink their operations (WSJ, “For Starbuck’s  Baristas, It’s Back to the Grind, 6/17/09).  In some cases Starbucks had let efficiency reduce the customer’s experience, such as grinding coffee only once a day.  Now, they will grind throughout the day so that the shop is filled with the aroma of freshly ground coffee – less efficient but better for the customer.  That said, they are not backpedaling on all efficiency.  For example, they are making their coffee brewers more flexible (instead of brewing only one variety, now they will switch) and consequently customers will experience stockouts of their favorite variety less often. And they have a team of 10 people focused on bringing to their shops “lean thinking” alla the Toyota Production System. Given that labor represents 24% of revenue, even slight improvements in motion can make a significant difference.  That means looking for how the company can reduce the walking, reaching, and bending that goes into the making of a cup of java.  Classic process analysis. (WSJ, Latest Starbucks Buzzword: ‘Lean’ Japanese Technique 8/4/09).  The changes seem to be having a positive effect – they reported better than expected fiscal third quarter profits this year (WSJ, 7/22/09).  They’ll need to keep it up – McDonald’s and Dunkin Donuts also understand the importance of lean in how they make their coffee.

Apologies for the layoff

It has been a long while since our last post, back in Feb 2009! The radio silence is not for lack of interest, but rather for lack of time – it has been an especially busy period for us. Of course, it has been also a very busy period for the world. So we have a long backlog of posts to get up there and we look forward to being back on a more regular schedule.

Toyota in the down-turn

Bad news about the auto industry has been all over the news and Detroit executives are becoming regular visitors in Washington these days (no matter if they drive there or if they take the corporate jets…). While GM has had some time by now to get used to big losses, 2008 was a bad year even for industry darling Toyota. After years of dramatic growth in volume and in profits, Toyota now reports a multi-billion loss. Will Toyota be able to master this crisis “The Toyota Way?”. So far, all we know is that the company continues to honor its life time employment promise – the cost cutting so far has only affected temporal workers. Yet, vehicle inventory continues to grow and it is unclear how long Toyota can afford to produce more that it is able to sell.

 

See http://www.nytimes.com/2009/02/15/business/15toyota.html?pagewanted=1&_r=2&sq=toyota&st=cse&scp=3 for some updates on Toyota, including their new management team

 

http://www.nytimes.com/2009/02/15/business/15toyota.html?pagewanted=1&_r=2&sq=toyota&st=cse&scp=3

Lean manufacturing gone too far?

“5S” in the world of lean manufacturing stands for “sort”, “straighten”, “shine”, “standardize”, and “sustain”. While it is hard to argue with the success of lean manufacturing on the assembly line, how far should the concept be taken? For example, can an employee add a hook on a door to hang a sweater? According to some 5S proponents, the answer is “no” – it doesn’t contribute to “aesthetic uniformity”.  Put a box of papers on top of a file cabinet – another “no no”.  This raises the question of how far a employer should go to dictate how employees do their job. One might argue that the employer should care only about output and not about process. Another may argue that employees will not maximize output because they will not choose the right process.  It would be nice to see some data on whether application of 5S to office environments indeed yields improvement.

Wall Street Journal, October 27, 2008 – Neatness Counts at Kyocera and at others in the 5S Club

Prediction Markets and their Applications

Forecasting demand, especially for new products, has long been the weak spot for those of us interested in matching supply with demand. No matter the amount of market research a company does, forecasts are off target more frequently than we would like. While there exists a number of tools that support companies in planning for and absorbing this demand uncertainty (e.g. the Newsvendor model, models of risk pooling, reactive capacity), nothing solves this problem better than a good forecast.

Over the last decade, prediction markets have often been praised as the new way of creating demand forecasts. Prediction markets resemble betting on sports games: if you see that the betting return on $1 for one horse is $5 and on another horse is $50, you don’t need to be a jockey to figure out that the second horse is most likely a slower one. A fundamental advantage of a prediction market is that it implicitly puts a higher weight on those people’s opinion that have a stronger opinion. Ask 100 people how the weather will be tomorrow in Alaska. You could just take the average over the 100 forecasts that you get. But, chances are that many of the people you ask have little or no information about the weather in Alaska, while others are local experts. In a traditional forecasting approach, everyone is weighted equally. If, however, you ask people to place a bet about tomorrow’s weather in Alaska, you force them to “put their money to where their mouth is”. Those living in Florida, most likley, will not participate in this prediction market or place very small bets. In contrast, the meteoroligist living in Anchorage might be willing to bet a lot more. Since you, as the company who needs to forecast typically do not know who will have the best information for your forecasting problem, you should just let the (prediction) market decide.

Prediction markets have been used a lot for politics, including the US presidential elections (see e.g. http://iemweb.biz.uiowa.edu/graphs/graph_Pres08_WTA.cfm). However, as is discussed in a September 16 Wall Street Journal article, the number of corporate prediction markets is increasing. The article describes how Best Buy uses prediction markets to forecast the sales of new products and services. Every one of Best Buy’s 115,000 employees is allowed to participate. Yet, economic theory suggests that someone selling washing machines will be hesitant in placing bets on new lap-tops…

See WSJ, September 16: Best Buy Taps Prediction Market

More flights overbooked, but some things are changing

Airlines often sell more tickets on a particular flight than they have seats available. Not because they are mean, not because they can’t keep track of what they sell, but simply to protect them from passenger no-shows. If passengers, especially those with flexible tickets, change their travel plans at the last minute, the airline has to fly with empty seats. And, given the high fixed cost of operating a flight (fuel, labor, and landing fees), that is the airlines biggest fear.

The number of passengers that get bumped is publicly reported in the US by the Department of Transportation. This number is up, once again. About 1.16 passengers per 10,000 are bumped. Not a lot – until, of course, you are the one who gets bumped. 

Two things have changed recently. First, airlines are now better predicting how many passengers show up. Part of this is they just get smarter (using more sophisticated modeling tools), part of this is that the choice set of flights for most customers is shrinking (many airlines have cut the number of flights they offer). Second, planes are getting fuller. The load factor (the percentage of seats that are occupied in a plane) has increased some 10 percentage points in the last year.

How much to overbook remains an interesting problem for the airline. There are elegant analytical models available, that trade-off the cost of the empty seat with the cost of bumping a passenger.

http://www.nytimes.com/2008/08/23/business/23bump.html?em