Inventory modeling in action

Inventory modeling and inventory optimization attempt to drive out unnecessary inventory from your systems, to improve service levels to your customers.  This does work and can drive very significant reductions in inventory, but, if you lack discipline around execution you will not get as much value as you should.

The list of 10 watch-outs that follows is based on my experience.  Some of  these are mistakes I've made and learned from, others are mistakes I have observed.
  1. You do need a good model.   I've seen a lot of inventory models, some are more "unique" than useful.  This is an area with a solid analytic/statistical framework available that has been real-world tested.   Here's a link to my online calculator  [How much inventory do you really need] or check out this Wikipedia entry for a basic introduction   http://en.wikipedia.org/wiki/Safety_stock  You are not going to build something with common-sense or street-smarts in Excel that can come close.  Do the necessary learning, hire someone that already has it or buy into one of the commercially available packages.
  2. Better models yield better results.  If inventory really matters to you, you may want to invest in a more rigorous level of modeling.  A basic, statistical model will generate results if used well.  A model that better fits your reality will let you cut deeper/faster.  For a simple example check out: [Inventory modeling is not "Normal"].  Some other areas that may warrant extra work/investment: 
    • fitting the most appropriate distributions of uncertainty
    • capturing demand uncertainty effectively
    • handling multi-level distribution networks
  3. You need people who know how the model works.  This does not have to be everybody that ever touches it, but someone either in your organization or that you have easy access to must understand this.  I've seen system-implementers (consultants in this case) hamstring a perfectly good commercial package because they did not understand how the models worked and set it to provide bad recommendations. ("Garbage in - Garbage out")
  4. Integrate inventory recommendations into your planning system.  With good models you will generate unique targets by product and location.  To make this part of the planning process this data must be integrated in to the planning systems: it's completely unreasonable to ask planners to keep this amount of information "in their heads" and to act on it appropriately.
  5. Get supply planners involved. Don't ignore one of your best assets - the people that work with product supply every day.  Their intuitive sense of what works is probably not far wrong, whereas computers have no intuitive sense.  The model should help them challenge their ideas  and refine them it does not replace sense.   Get them involved!  Teach them to work the model inputs, to understand why the model does what it does and create ownership of the answers.  If you implemented months ago and you're still hearing "Your targets aren't right..." you need to work on this.
  6. Measure compliance.  You will, I'm sure, be measuring what happens to your aggregate inventory and service levels.  These should be moving steadily in the right direction.  You need to look down in the weeds though to know if you are getting full value.  Excess inventory mounts up a with few cases here, a weeks extra supply over there and it all adds up. Exception based reporting to find item:locations that are not under control drives deeper/faster results.
  7. Use the right target for each decision.  Your system probably has just one place to embed a safety stock or replenishment point value for each [product:location] combination.  If you have done your modeling work effectively this value will represent your primary replenishment process.  But, what happens if you have alternatives?  Perhaps sourcing from another (expedited) supplier or re-deploying inventory from another facility?  The assumptions you made to generate your target are wrong for this alternative option.  Ignore this and you can spend a lot of money very quickly. Check out [Balancing safety stocks across DCs
  8. Continually update/revise.  Let's assume that your system will automatically record and update statistics around demand uncertainty so you do not have to touch every model every week.  Even so, things change:  manufacturing constraints change over time, your understanding of your own supply chain improves, your willingness to take risks shifts, ... things happen.  If you lock in your inventory targets once a year and forget them your supply chain will under-perform.  Continually tweak and revise the models.  these targets drive your supply systems, the better they are the better you will look.
  9. Resist the temptation to react to one-off events.  It is the nature of uncertainty in your supply chain that most of the time, most orders for most products are filled completely.  When there is a failure, orders get cut, service level for that product drops and it becomes the center of  attention for a period of time.  Understandably, this can become uncomfortable.  However, if your aggregate service-level metrics are in-line and this  appears to be a one-off event, reacting to it by pushing up inventory achieves only that - it pushes up inventory.  
  10. Be tough on inventory AND tough on the causes of inventory.  Your inventory models are not just a way of setting accurate inventory targets.  They are key to understanding what changes to your supply chain would drive lower inventory.  What is the value of:
    • 5 points of improvement in forecast accuracy?
    • a 25% reduction in lead-time from production?
    • replenishing once a week rather than once a month?
    • a delayed deployment strategy for hard-to-forecast  products
    • reducing service levels by 0.5 points
    • stratifying service targets for different groups of products (typically based on volume and/or demand uncertainty)
    • allowing service levels for individual product:locations to float in a wider range and using optimization to minimize the overall inventory while maintaining the aggregate service level target

A successful inventory modeling/optimization project does need a good model but it also needs  great execution.