On the power of randomization: Selling diapers to pregnant women

I posted before on how Priceline uses randomization to conceal the minimum price they would be willing to accept in the Name your own price channel. Here is another example on putting randomness to work. This one is from retail.

Whether we like it or not, retailers do accumulate a wealth of data about our purchasing patterns. Every time you use a credit card, coupon, or a loyalty card, transaction data is logged and stored. The data is obviously used to send you more coupons, ads and peddle new products. Now, how much can retailers learn from this data? It turns our quite a lot. As this article from the forthcoming NY Times magazine describes, Target can actually predict whether a woman is pregnant just by analyzing change in her shopping patterns.

Pregnancy is a sensitive matter, however, so when a week after shopping a bunch of coupons for maternity clothing and baby products arrives in mail, women can get upset. It also looks awful lot like spying – does not it? Lesson #1 – one has to be very careful with this kind of data. Lesson #2 – even if data suggests something, do not necessarily pursue the opportunity at full speed. Here is what Target does: they randomize. Put a coupon for wine glasses next to diapers and office furniture next to pacifiers. Combinations seem to be quite ironic. But it does seem to work – the same article reports substantial growth in maternity and baby product sales for Target after they started doing this.

This is possibly very big!

This thing is making a big splash on the Internet right now with potentially big implications for wireless technology and (Why not?) could make Nicola Tesla‘s dream of wireless energy transmission reality. Watch and enjoy!

More at http://www.wesolveforx.com

Labels and supply chains

Interesting information you can get from reading product labels. And I am not talking about food ingredients. What I am talking about is how long does it take from the moment a product is manufactured until it is sold? Or even, how long can a manufacturer afford this lead time to be? Lead times are tricky and rarely reported by firms. Longer lead times mean more working capital and pose challenges for forecasting, because firms have to decide how much to produce well in advance. In fact, in this paper I argue that retail sales forecasts (and inventory budgeting) for the next year are done 6 to 12 months before it starts. Anyways, because lead times are so tricky, I always welcome first hand data that documents them.

In this case the data comes from two product labels – one from Ikea and the other is CB2. Both related to furniture bought by me in the beginning of February. It turns out that Ikea manufactured that product (it was a chair) on May 26, 2011 in Mexico. Moreover the cover for that chair was made on the 16th week in 2011 (that is around Apr. 20). The chairs from CB2 were made in Taiwan by vendor Elegant Products and shipped from there on Aug. 4, 2011. Which gives almost 9 months lead time for Ikea and 6 months for CB2. Again these are the lead times after the product is manufactured. Actual decision about manufacturing them had to be done before that.

Give or take, it seems that Ikea’s lead times are about 50% more than CB2’s. And it makes sense, given that Ikea’s assortment rarely changes (aside from seasonal items) and CB2 tries to follow the contemporary trend.

 

Does Boeing know operations?

Operations management is indeed about continuous improvement and this WSJ article illustrates the point brilliantly with Boeing. What they do is applying principles of Kaizen to their processes. Shaving off 15 minutes from an activity may not seem much in the context of aircraft production but things do add up. What is interesting about the article is the numbers.Here are some excerpts:

Workers here recently boosted 737 output to 35 jets a month from 31.5, and Chicago-based Boeing aims to produce 42 planes a month in 2014. Executives said they are studying ways to eventually reach 60 a month as they plan a retooled version of the plane called the 737 Max, a jet that Boeing expects to begin delivering in 2017. The company is trying to pare an order backlog of some 3,700 jetliners, including about 2,300 of its best-selling 737s.

“How do you produce more aircraft without expanding the building?” is the question Boeing managers in Renton continually focus on, said Eric Lindblad, vice president for 737 manufacturing operations. “Space is the forcing function that means you’ve gotta be creative.”

Boeing today takes about 11 days for the final assembly of jets at the Renton plant. That’s down from 22 days about a decade ago, but the company has for years set goals to go even lower. Mr. Lindblad said the company’s near-term goal is to whittle that number to nine days.

First of all, 11 days to assemble an aircraft is impressive. I’ve been following the development of a new Russian passenger jet SSJ 100 and I suspect they are dreaming about this kind of numbers – since April’11 just 6 have been delivered so far – which is about 2 aircrafts every 3 months. Boeing produces 35 a month. Moreover they want to produce even more. And here comes my question – do they know operations? The answer is –

Process improvement: Innovating pick and pack.

Imagine an order fulfillment center for a big online retailer. How does it work? One can probably think about hordes of people running with order lists picking items and putting them in a bin: pick, bring, pack, ship. This looks like a labor intensive process and it is. In fact this labor intensity was one of the factors that led to demise of Webvan – online grocery store. Fortunately, people learn from mistakes and sometimes change such a mundane process as pick&pack in a bold way. This video explains how (hat tip to Benn Konsynski for posting the link).

It is quite astonishing. The company that provides this sort of solutions is Kiva Systems and apparently Staples uses it in their fulfillment center. What made me think though is the simulation of shelves moving on the warehouse floor. It looks like the system is actually self-organizing: the most popular items will naturally be situated closer to pick workers, just because they are frequently demanded and will not have much time to drift away. More over if demand changes over time, the system will reorganize itself. Perhaps it will still make sense to put Teddy bears out front for Valentine’s day, but if you forget about chocolates they will get there anyways.

One thing that is not shown in the video is how do you replenish inventory. A solution it seems would be to ‘decomission’ an empty shelf and replace it with a full one. Given that all pick operations are logged, it should be fairly easy to figure out when the shelf is about to empty and prepare a replenishment. What happens if there is more than one kind of items on the shelf? It’s inevitable that one SKU will be sold out before the other, but I imagine grouping complementary products can help.

How do you make money forecasting?

In the Operations management class that I teach this semester this week is dedicated to forecasting. While we mostly focus on methods, for example, time series and causal models, it is also important to think about business aspect of forecasting. That is – how do you make money doing it?

Suppose you are interested in predicting an election outcome.Traditionally, companies like Gallup or Rasmussen run surveys and polls, and this can be quite expensive. If you can substantially reduce costs, then you can lower price and compete successfully with bigger companies. How?