sadfsaf

I will go first

Worked on a consulting contract at Shaw's Supermarkets (for you non-New Englanders, Shaw's is the 2nd largest supermarket chain in NE) Headquarters for 18 months.  Half of the time it was inbound freight cost optimizing.  But the other half was redoing Shaw's 15 year old forecasting process.

In an attempt not to bore you too much, this will only cover the items that are going to go on sale for a week in the future.

This exercise starts 8 weeks before the actual sales week. 

There are many factors that drive the selection of the items going on sale that week.  They are:

· The item

· Amount of discount (supplier plays a big role here, how much are they willing to drop their price to Shaw's.)

· Time of the year

· Other items to be in the same flyer (don't offer more than 2/3 main meat items)

· Which stores (with over 200 stores, some items just do no sell, even on sale, in certain locations + what is the competition?)

· When was it last on sale

The old forecasting system took a rolling 8 week forecast to look at the sales average for this item when it was not on sale.It had an error rate of about 5%.

We 1st reviewed the last 5 years worth of data for every sale item during that period of time.  (great to have a very powerful mainframe computer)Sliced and diced the data every way possible.  Reviewed what had a very low error rate vs what had a very high error rate  (error rate was the actual sales vs the forecasted sales numbers)

The initial data showed that the 8 week forecast for all was not the best way to go.  Many items had very low error rate.  So we recommended that they stay using the 8 week forecast.  The biggest driver for sales was the price (no surprise there).  Then the next phase was to look at the stores that had low error rates to see what they were doing and what was their market segment.

With so much data the decision was to make the sample size smaller using the largest 50 stores in sales

The end result was that 6 different forecasting tools; the old one plus 5 new ones.  The new ones all used history but for various lengths of time.  Plus the weight given to each of the variables listed above.

So the process was to meet with the sales representative and determine how low should the price be and how much would the supplier contribute.  About 50% of the time, the sales rep would start the process.  After that, the models would be run in total then a break down by store.  That would tell what stores would have the greatest sale for that item and which ones would not.  (Pastene, a manufacturer of high quality can tomato products was only sold in about 50 stores.  This was due to the % of Italian shoppers were in that stores market area).  So when Pastene went on sale, only 50 stores sold it.

Many items put on sale would increase the number of shoppers during the week of the sale.  So the profit estimate was made for each and every sale item.  Many items are Loss Leaders.  That is they drive the sale of secondary items.  Example:  put hamburgers on sale and you will sell at full price, rolls, condiments, paper plates, chips, etc.  So putting hamburger on sale would mean a low profit for hamburgers, maybe even a loss but an increase in profit for all the related items.  So the forecast was just not for the on-sale item but for the other items.  This is called the "Halo Effect".

Your turn:

.