The Infosys global supply chain management blog enables leaner supply chains through process and IT related interventions. Discuss the latest trends and solutions across the supply chain management landscape.

« Crunching Procurement Cycle Time Through Business Workflow Optimization | Main | Why my EAM implementation is not giving me as I expected? »

How far "Pull" concept should drive your Supply Chain Activities?

In this blog I will try to answer one of the most pertinent question every supply chain operating manager face in his or her day to day working. To set the context consider a simplistic model of a manufacturing supply chain.


In this supply chain on extreme right we have finished goods warehouses from where dispatches happen to end customers (Also called 3rd Party Customers). These finished goods warehouses draw the goods from regional warehouses. Regional warehouses in turn order the goods from finished goods factories depending on which product is produced in which factory. Few products can be ordered from subcontractor as well. Finished goods factories require some raw materials (or components) and some semi finished products. For example - If finished goods factory is manufacturing car then gear box assembly (A semi finished product) could be coming from some other factory. Semi finished factory in turn will require raw material and components for its operation which it will order from supplier. This is logical flow of demand. Now let us consider forecasting flow.


At finished goods warehouse demand forecasting is done for a reasonably long horizon, generally for 24 to 36 months into the future. Assume lead time of 1 month between finished goods warehouse and regional warehouse. Regional warehouse will see the demand forecast 1 month before finished good warehouse. Lead time could be different for different finished goods warehouse. So ultimate demand picture of regional warehouse is combination of all these lead times. Same will happen at factory level. Factory will see demand off set further based on lead times between regional warehouse and finished goods factory. Factory will have its own manufacturing lead time so semi finished factories will see demand even further before (to the extent of manufacturing lead time plus transportation lead time between finished goods factory and semi finished goods factory). And finally raw material and component demand will have additional lead time offset added because of semi finished product manufacturing lead time. It may so happen that supply chain cumulative lead time offset could become extremely long. Raw material we are ordering today at semi finished factory could be for a finished good warehouse demand for 3rd party customer which is 18 to 24 months into future. My premise is as follows....


"How logical is it to carry out all your today`s supply chain operations at upstream warehouses and factories based on a demand forecast for a period which is more than 12 months into the future?"

But this is what we generally do....right? We think there is no alternative. Famous "Pull" concept of supply chain management tells us to do so. We are told that if we do not do this we will end up building unwanted inventory in the supply chain, inventory is a "muda", it affects our responsiveness, it's bad for quality and so on. But think for a moment, is this logical? Put yourselves in shoes of a demand manager at finished goods warehouse. What is your prime motivation when you do demand forecasting? It is to do forecasting correctly so that you do not face stock out situation. You will try to be 100% accurate for next one month (If lead time from regional warehouse is 1 month). You will try to be reasonably accurate for next 3 to 6 months because you have good market intelligence for that period. You will attempt to be accurate for next 12 months because all the financial results are computed for a year and that has bearing on your performance appraisal. But beyond 12 months do you really care what is your forecast accuracy? You give this forecast more as obligation for upstream players to plan their work with almost zero interest in its accuracy. You always know that you will have multiple opportunities to revise this forecast before you are held accountable for its accuracy. If forecast beyond 12 months is developed with this philosophy, how fair is it to base your upstream operations based on this forecast? Answer in my opinion is clear. "Pull" concept should not drive supply chain if demand forecast beyond 12 months is used for driving it.

There are many arguments put against about above statement prime being "We know that forecast accuracy beyond 12 month at individual finished goods warehouse is wrong but when aggregated at regional warehouse level or at factory level it will balance out". Which means one finished goods warehouse will have positive forecast error and other will have negative error hence on an average accuracy beyond 12 month will be as good or as bad as it is in first 12 months. There are statistical theories to support this argument. This is not a convincing argument in my opinion. Forecasting is as much a science as an art. Final forecast number is arrived using algorithmic logic, market intelligence and gut feeling about future. Forecast beyond 12 month is based more on gut feeling for future. It has more macro considerations than micro. These factors are more or less same whether you are demand manager at one finished good warehouse or other so evening out of forecasting errors rarely happen beyond 12 months. So what is the answer?

In my opinion we should apply following methodology at each node in supply chain.

1)    Determine forecast accuracy at each node. For example, if you are say buyer for raw material at semi finished goods factory, compare demand thrown on you through "Pull" with actual consumption. Demand here means propagated "Pull" demand that comes on this raw material from forecast done at finished goods warehouses. By consumption I mean goods issued to factory for doing production.

2)    If the forecast accuracy for this raw material is drastically different than short term forecast accuracy  at finished goods warehouse, no point in using "Pull" concept at this node. Why should this node deal with more uncertainty than finished goods warehouse? Additional drop in forecast accuracy at this node has nothing to do with market uncertainty. It is induced by "Pull"

3)    Instead of using Pull demand create fresh demand forecast at this node based on consumption history

Planning done on the basis of this "independent" forecast will be much more accurate than planning done based on "Pull" demand. Do this analysis for each and every node in your supply chain where you do demand and supply matching in your supply chain network. You will be amazed to see that 70% of the nodes, mostly upstream, will fall in this category. Managing them as an "independent" forecast entity will do world of good for their planning than managing them as "Pull" entities. In few cases putting a strategic safety stock where forecast accuracy is a problem is much better approach. In summary just delink each node from being demand carrier for upstream. Make demand manager at each node a real "Demand Manager" rather than converting him / her into a postman who merely transfers the demand upstream by doing inventory netting. If you organize demand management this way, everyone is responsible for forecast accuracy rather than just handful of people.

A somewhat unconventional thought.




Nikhil, interesting thought. But I think there are ways and means to reduce negative effect of lag in forecast accuracy at downstream and upstream nodes in supply chain. Adapting 'stand alone' forecasting model for each node essentially means putting walls between the nodes and not allowing 'market intelligence' to flow from near-market nodes to upstream nodes and allowing bullwhip effect to manifest itself. Forecasting at each node completely based on its own history will have this handicap of not using forward looking information. Hence instead of this model, supply chains try to
1. Reduce lead times - to ensure that downstream nodes are planning and acting on near term forecasts (which are more accurate)
2. Increase forecast accuracy at granular level for periods which are meaningful for upstream nodes
3. Use strategies which reduce commitment to final products at very early stage - such as 'postponement of final assembly' or modular product designs etc.
All these are easier said than done.
You have mentioned that your thought is unconventional, but have you come across any live supply chain example where the 'standalone forecasting at each node' model is working. Would like to hear about it.

Post a comment

(If you haven't left a comment here before, you may need to be approved by the site owner before your comment will appear. Until then, it won't appear on the entry. Thanks for waiting.)

Please key in the two words you see in the box to validate your identity as an authentic user and reduce spam.

Subscribe to this blog's feed

Follow us on

Blogger Profiles

Infosys on Twitter