Virtual Pull Systems
article written by Don Guild, Synchronous Management

This is the first in a two-part series. This article is an introduction to the concepts of Virtual Pull - and how to size kanban supermarkets. The second installment will show you how to execute a visual pull system - without the manual work typically associated with visual systems.

So, have you implemented kanban yet? Have you been unable to roll it out - or just abandoned it? Most companies who begin kanban implementation struggle to finish the job. In too many cases, the following questions not only go without answers, they go without asking:

  How do we size and resize supermarkets to reflect process and demand changes?
  How do we kanban hundreds or thousands of parts?
  How do we schedule a level load and mix with batch production?
  How do we account for make-to-order requirements?
  How do we use our pull system to focus process improvements?

As a result, two major problems persist. First, kanban quantities may not account for major supply and demand constraints. Often, chaos still reigns and improvements in service and lead times remain elusive. Second is the assumption that kanban systems must be manual to be visual. So, most pull/kanban implementations occur via "kaizen" events. A team is structured and trained, and kanbans are made and implemented on a few items in a few days. Almost always, this approach loses steam because the struggle to apply it over and over again to additional products is difficult and costly.

On hundreds of implementations, our clients have overcome these stumbling blocks with Virtual Pull systems. Virtual Pull views "pull" as a system, not as a series of kaizen events. The nature of pull systems is that they connect value streams and loops with each other; they cannot be effectively implemented locally.  So, Virtual Pull:

  Quantifies inventories based on constraints in supply and demand.
  Calculates, by item, the maximum inventory required to maintain material flow.
  Eliminates non-value-adding kanban cards, containers, and boards.
  Provides visual schedules accessible anywhere on your network.
  Incorporates mixed-model level-loading (heijunka) into the scheduling process.
  Incorporates non-kanban demand into level-loaded visual schedules.
  Reduces non-value-added scheduling and expediting effort by up to 90%.

Virtual Pull is being used to size supermarkets and to level schedules in a wide variety of environments. These include offshore procurement, consumer products distribution, and daily production scheduling. The results are reductions in total enterprise inventories of 20-40%, improvements in service levels to all customers to 99%+, and 70-90% reductions in scheduling workload across the organization.

Of course, other lean tools (e.g. cells, changeover reduction, zero defects) are important to the long-term success of an enterprise. But, Virtual Pull does not require that any of them be in place before implementation. And it does not use proprietary software. Your Virtual Pull system is constructed in MSExcel or MSAccess using data you already have available.

Implementation
Virtual Pull is implemented in two phases. The first is to calculate supermarkets - or maximum inventories required - item by item. With such a model, we can also recalculate the supermarkets on demand as conditions change (e.g. engineering changes, new products, process improvements).
The second phase is the actual execution of pull and the level-loading of suppliers - both internal and external. If supplier level-loading is not required (e.g. nuts, bolts, fasteners), then this phase of Virtual Pull can be skipped. The application of inventory replenishment tools such as order point or min/max may suffice.

Sizing Supermarkets
A supermarket is inventory of an item which is strategically sized and placed to perform several functions. First, it ensures an uninterrupted supply of material. Replenishment is triggered based on actual consumption - not a forecast. Second, it links and protects the flow within and among value streams. Third, because it is quantitative, it can be deconstructed into its root causes. This provides a valuable means for focusing additional value stream improvements.

A Virtual Pull model is typically constructed using commonly available data from your planning and scheduling system, including:
  Historical usage or shipment patterns
  Forecasted demand rates
  Processing data, such as bills of materials, changeover times and cycle times
  Resource data, such as manning, efficiencies and utilization
  Supplier data, such as lead times, replenishment cycles and vendor inventory agreements
  Part data, such as sourcing, unit costs, container quantities and minimum order quantities
  Current open customer order requirements
  Current on hand, in process and/or in transit inventories

In most cases, all of these data can be automatically downloaded from your requirements planning system each day. Then, changes in supermarkets can be simulated and schedules can be updated at any time.

Each supermarket consists of at least three elements, each of which must be determined separately for each item (see Supermarket Elements):

  Average Forecasted Demand During Replenishment Lead Time: This is the average amount of
    consumption from the supermarket which is expected to take place during the supplier's lead
    time to resupply.
  Buffer Stock to Cover Deviation from Average Forecasted Demand: This is the extra inventory
    required to cover variation from the average forecast. Buffer stock is based on historical average
    consumption, and deviation from the average, factored for the forecasted demand and required
    service levels.
  Average Forecasted Demanded During the Replenishment Interval: The replenishment interval is
    the highest frequency with which the supplier can replenish the supermarket. This interval, when
    multiplied by the usage rate, determines the average quantity to be released or produced. Note
    that the replenishment interval is related to, but not the same as, the supplier's lead time. For
    example, if a supplier ships daily, and it takes five days to transport the goods, the replenishment
    interval is one day, and replenishment lead time is five days. The replenishment interval is the
    same "interval" referred to in the lean term, "every part every interval (EPEI)".

Virtual Pull Example
To illustrate the structure and execution of Virtual Pull, let's use a simple case study. Shipment History shows thirteen weeks of shipment history for seven items (your horizon may be different). Several of the items are made to stock (MTS = supermarkets) and several are made to order (MTO = no inventory). Shipment History shows the calculated average historical weekly demands, and the standard deviations for the stock items.

Forecast shows the forecasted average weekly demand for each item. The forecasted average demand is used by the model to calculate all three components of each MTS supermarket. Each MTS item's historical standard deviation is factored by its forecast to produce a forecasted standard deviation which in turn can be used to calculate its buffer stock.

Next is the calculation of the replenishment interval. Let's assume that one machine is required to produce these seven items, that they are the only items produced on that machine, and that one week of further processing is required after the items are machined. Production Processing Data shows the processing information (changeover and cycle times), the calculation of the number of hours required to produce one week's forecast for all items, and the number of hours required to cycle through the changeovers on all items.

Replenishment Interval shows the calculation of the interval. Assuming the machine is manned 40 hours per week, and its uptime is 95%, then 38 hours a week are available to run product and to conduct changeovers. If 35.3 hours a week are required to run product, then 2.7 hours a week are available to perform changeovers. If 2.7 hours are required to set up all items one time, then all items can be set up and run once per week. Thus, the replenishment interval in this case is one week.

Supermarket Calculations shows the calculation of the supermarkets of the MTS items, based on current supply and demand constraints. Note that the two week lead time includes the replenishment interval plus one week of further processing, and the service level used to calculate buffer stock is 95%. The average order quantity will be one week's worth, although the actual quantity released may vary from week to week depending on consumption. And, as input conditions change (e.g. demand patterns, changeovers, forecasts), the supermarkets can be recalculated on demand.

Next installment - learn how to pull these seven items through production with a level schedule using their visual "fuel gauges" instead of manual kanbans.