Are We Playing Sudoku or Forecasting a Business?

Most, if not all, of you are familiar with Sudoku puzzles.  Now I must admit that I’m actually not very good at Sudoku.  On the rare occasions where I attempt a puzzle – usually in an inflight magazine – I quickly find myself turning to the answer key for a hint.   Nevertheless, the game has grown in popularity rapidly outside of Japan since The Times in Britain introduced the puzzle in late 2004.  New Zealander Wayne Gould is credited with stumbling across a semi-completed puzzle in a Japanese bookstore and taking the idea to The Times.

According to Wikipedia, there are a stunning 6,670,903,752,021,072,936,960 solution grids for the classic 9×9 Sudoku puzzle – no wonder I have to cheat!

Any given Sudoku puzzle has one right, one perfect answer.  The challenge, of course, is figuring out how to organize the numbers in each box, row and column to arrive at the one correct solution.

I missed my tube stop!

The first letter to The Times was from a reader stating he had missed his tube stop the day before because he was so wrapped up in the puzzle.  So what does any of this have to do with forecasting?  One of the key questions any demand planning organization must answer is ‘at which level(s) should I generate forecasts?’ Of course, the correct answer is largely dependent upon the purpose of the forecast – the downstream business processes that will ultimately leverage the information.  A forecast driving store deliveries requires a different level of detail than a quarterly financial forecast, which would also be quite different from the more finite needs of a labor forecast.   However, even once the consuming business processes are identified, there are multiple – in fact, many – possible forecasting levels.   There are at least three dimensions to be considered – product, location and customer.  Suppose each of those dimensions has three or four levels and you need to figure out what unique combination of those levels maximizes your forecast accuracy.  Now consider that there may be many forecasting methods and parameters available.  The challenge facing supply chain forecasting practitioners is to figure out which unique combination of product hierarchy, location hierarchy, customer hierarchy, forecasting method and parameters maximizes their accuracy.  Beginning to see the connection to Sudoku?   While there may not be   6,670,903,752,021,072,936,960 possible solutions, there are certainly too many to simply try them all out.   Organizations often then fall into one of two traps.  They spend an enormous amount of time and money attempting to find that one right answer, often resulting in a solution that is still highly debated.  Or, they simply pick something simple which appeals to common business sense.   The challenge with both of those routes is that they often leave millions of dollars’ worth of assests in play from inventory and labor to space and transportation as a result of suboptimal forecast accuracy.   It’s really easy to get so wrapped up in the effort of figuring out the perfect answer that you miss the tube stop – your project deadline, budget or day job.

Introducing Automated Parameter and Level Optimization

The good news is that, thanks to new capabilities in JDA’s demand management portfolio, it will find that answer for you so you can keep your eye on the rest of the business and be less likely to miss your tube stop.

JDA has long had a market-leading ‘Demand Classification’ capability that leverages strong algorithms tuned with over 30 years of empirical evidence to identify the right forecasting method and starting parameters.

In our recently announced 9.0.0.0 release, JDA Demand Classification has been greatly improved to also recommend the right level at which to forecast so as to maximize accuracy for the consuming business processes.   Forecasting boxes of deck nails in retail stores for the purpose of positioning inventory?  Individual boxes of nails in a single store are likely too low a level of detail to capture robust seasonality trends.   You could aggregate on the location dimension and forecast a box of deck nails across the United States.  Doing so would likely cause some seasonality distortion though – one would expect deck building season in New England to be quite different from that in Texas.  Similarly, you could aggregate up the product dimension and forecast all nails in a store.  Again, the trends for deck nails are probably different than the trends for finishing nails.  Somewhere between a box of nails in a store and all nails in all stores lies the right answer.  Now, with JDA’s Automated Parameter and Level Optimization, you’ll find that right level – and you’ll also be provided with the insight as to the right method and starting parameters for that level.

Don’t spin your wheels trying to find that one unique solution manually.  Considering a JDA Demand upgrade?  An upgrade is the perfect time to revisit the forecasting hierarchy.  Discovering that right answer just might uncover significant inventory savings – more than justifying the upgrade effort and give you the free time to try one of the 6,670,903,752,021,072,936,960 in that puzzle you’ve been working on from last month’s paper.

No Comments

Be the first to start a conversation

Leave a Comment