Most obvious difference with push and pull production is that in MRP system production is scheduled based on demand or predicted demand and raw materials are pushed into the production based on that production schedule. In pull production every workstation is authorized for the job based on the storage in the next step.
In other words push production is based on external information whereas pull production is based on internal information (or stock levels). Of course the last storage is always affected by external demand.
Biggest advance in pull production is based on its intention to control WIP levels. Work is not authorized until stock is at certain point in the next phase. Pure push production inputs material into production regardless of WIP stocks.
This leads into that that theoretically in push our WIP stocks are infinite and because of the outages in the system they continually increase. In practice stock levels are monitored and they cause changes in the production plan. So in practice MRP system is hybrid of push and pull production.
The magic of pull production
The beauty of kanban system is based on the WIP regulation not the cards itself. Production signal can be given in many ways and the simplest way is card. When there is no cards in the work station it means that next stock is full. During breakdowns floor is not filled with WIP like it would happen in the pure MRP system.
This also adds flexibility because when jobs are not released they can be modified. There is very little WIP in the system and future production can be modified quickly. This flexibility is highlighted during the outages.
Low and actually pretty constant WIP keeps also cycle times smoothened. Cycle times in pull production are less variable than in push production. This improves customer service as we know cycle times with higher probability.
According to Little’s law cycle time increases when WIP increases so cycle times in pull production are lower than in push production. But this is not corollary of pull production but result of WIP regulation.
In JIT production setup times and other outages are tried to reduce because this is the only way to keep same throughput with low WIP levels and short cycle time. Notice that this is not because of pull production but because of low WIP all outages quickly affect whole production.
Low WIP brings all disruptions at sight when they are easier to notice and fix. Also slight buffer creates pressure to intervene failure sensitive workstations.
Quality is also often combined into the JIT. Because of no buffer every quality problem leads into shutdown at next stations. Big buffers can cover up problems and when they are detected we have plenty of faulty production. Because of low WIP quality problems are quicker and easier to notice. Also production system that is sensitive to interference creates pressure to intervene work methods and make quality at the first time.
Belief that when worker claims parts from previous station he can check the quality is also often combined into kanban system. But this can not be done if parts are moved with forklift by another worker. Then parts are kind of “pushed” into next station. This kind of quality control is possible only if parts are small and work stations close each other.
Even though kanban is the most known application of pull production where stock levels are controlled at each work station it is not necessarily the most simplest model. More simpler would be to regulate WIP in the whole system. This is called CONWIP as in constant work in process because WIP in the system is nearly constant.
Idea is that we have certain amount of production cards that are attached into the products. When product is finished its production card is removed and returned to the first station.
Every station is allowed to do jobs without limitations. Unlike in kanban system where station needs production card from the next station in addition to the raw materials. In CONWIP production card kind of follows the parts so stations are authorized for production when they have raw materials.
CONWIP vs. MRP
Push production is much more difficult to control because it tries to control the output when pull production controls the WIP in the system. From the previous posts about variability you must have noticed that throughput depends on multiple variables. That’s why its prediction is very difficult and the production planning based on capacity predictions is difficult in push production.
If we input too much parts into production our WIP levels explode. If we input too little we lose throughput and sales (revenue)
Same output (TH) can be achieved in pull production with smaller WIP. According to Little’s law this means that cycle time is smaller in pull production if the throughput is same.
Pull production is also more robust to different throughput levels. Profit in push production decreases sharply when we are out of the optimal production. Profit in pull production is fairly same at troughput levels of 70 – 140 % from optimal and does not decrease a lot even at 40 – 160 %. Profit in MRP production is negative already at 120 % production and decrease is fairly linear between 0 – 100 %.
This only highlights the opinion that capacity forecast is difficult. In MRP production we should be able to plan almost optimum production so that we could have maximum profit. CONWIP production is more easier to control by controlling only the amount of WIP. In addition the profit does not decrease much if we are not near the optimum production. This means that it’s much more easier to end up near optimum production and profit in pull production rather than in push production.
CONWIP vs Kanban
Yes, they both are pull productions so they have same benefits compared to push production. But they have differences too.
Biggest difference is the difficulty to decide the amount of production cards. In kanban system we have to decide card amount for every station. In CONWIP we limit only one thing – whole line. That’s why it is the most simplest application of pull production.
Secondly in kanban system production cards are product specific which means that each card orders specific product. In CONWIP system one card means just one job and jobs are released based on separate production list that holds the order of products meant to manufacture.
Big amount (variety) of cards can become a problem in kanban system that produces a lot of different products. In theory it could be possible that workstation orders wrong parts that are not meant to produce for a while.
Lead time in kanban system is 0 when in CONWIP system it’s “small”. This means that kanban system is make-to-stock system where parts are made to WIP stocks and next station can take parts from there immediately. In CONWIP system cycle time is kept small by keeping WIP levels small. Price from that flexibility is this small addition to lead time.
Because in kanban system we can define stock levels at each station we can compensate the bottleneck machine with bigger stock in front and after the bottleneck. But if we produce multiple products which production times at work stations vary we have an problem.
Think that when producing A product our bottleneck is machine 2 and with B product bottleneck is machine 3. If we produce products in batches we could adjust the card amount when changing to different product family. But if we produce products with constantly varying mixes it is not easy to adjust card amount. CONWIP system controls only the number of jobs in the whole system so it adjust automatically where the WIP queue in the line will form.
At the beginning I wrote that pull system is controlled by internal data. Even though production can be controlled by internal or external data it must face the external demand at some point. But we can influence where inventory and order faces. In this interface make-to-stock production changes into make-to-order production.
Simple example comes from restaurant world. Think about fast food restaurant that makes hamburgers from order. Fridge is filled based on stock level so it is filled with MTS principle. Inventory/Order interface is located in the fridge and all work phases from this point forward is made for customer’s order.
But buffet table is filled based on the food there is left on the table. So in buffet restaurant food is prepared and cooked with MTS principle. Interface is in the buffet table where food is kept warm waiting for the customer. Usually customer takes his own food from the warming table but in some cases it can be also served and packed to take away.
Hamburger restaurant is more flexible because products in the interface are raw materials and can be cooked in many ways based on what customer wants. Price that we pay from this is time. In buffet restaurant customer gets his food immediately.
Solving the place for interface is difficult and there is no unambiguous answer for that. This depends completely on customers expectations and habits. You need to draw line between speed and flexibility. The line can also be variable. Example many hamburger restaurants cook hamburgers ready during the rush hours. This way the most common hamburgers can be served immediately.
Product variations affect too. Example it’s barely reasonable for the plywood factory to produce for orders if plywood can be bought only in couple thicknesses. The most reasonable option would be to manufacture products ready (MTS) when the interface is located in the warehouse.
One good example more is paint in the hardware store. Only white paint is stocked and it’s then tint into the tone that customer wants. It would be insane to stock every tones and no one would. This would mean that only basic colors are stocked and special tones would be ordered from factory. Special tones would be expensive and their lead time big. Interface has been brought near the customer while adding variables, keeping lead time small and keeping stocks low. Win-win situation for all!