Dear Gemba Coach,
We’re a tier two supplier and we’re trying to implement a pull system. We’re leveling our production schedules, but our customer’s variation in the demand for products immediately makes us miss shipments. Our OTD is not getting better. Although production runs more smoothly, people are getting frustrated. Where would you start?
That definitely is a tricky question, and I’d start at the gemba. Are your truck preparation areas all set up? The only practical way of guaranteeing shipments is to have the truck prepared in advance and in full. The first place to go is shipping, checking that truck preparations areas are under control. This means that for every customer destination, shipping staff need to understand:
- Truck destination
- Truck departure time
- Preparation zone (where are the parts for this truck)
- End of preparation time (at what time the truckload should be complete)
- Status (are we ok or are we late)
- Comments (if we have missing parts, why?)
As a kaizen focus in shipping we should try to have all trucks prepared in full by the end of preparation time. The gap between end of preparation and truck departure is meant to be able to respond an emergency if a box is missing. Some would argue that we’re creating muda of handling the parts as opposed to picking them directly in a warehouse and placing them in the truck, and that is the case. When we grow up we’ll be able to do so. But in the meantime we use the truck preparation areas to (1) visualize our mis-shipments, (2) be able to react immediately and fight for the completion of every truck, and (3) ask why? when a truck preparation area is missing crates at closing time.
The next step is to wonder about how the parts get to the truck preparation area. If you’re running a pull system, then parts should not come from a finished goods warehouse, but from a shop stock on the line itself. The production line, cell, or machine owns its own stock, and has a supermarket of parts with dedicated lanes for high runners: A, B, C, and a general purpose lane for low runners, X.
Parts are picked up from the line’s supermarket by parts withdrawal kanban. These kanban instruct the train operator (or the forklift) that a crate of A has to be withdrawn from the shop stock and brought to the corresponding truck preparation area at the specified time.
Time is specified by a leveling board. This is a big box with slots for kanban cards. On the vertical side, the box is organized by destination first, and within a destination, by parts numbers – each part number has its own line on the board. If the same part goes to two different destinations, it can appear twice on the leveling box – each at that customer’s takt time. Horizontally, the board presents time-based pigeonholes in which the kanban cards are placed – several slots for every hour. A truck preparation area should be consolidated from a start time to an end time.
Learning by Leveling
This leveling board is like a bicycle crankset. It converts the pull from customer trucks into withdrawal on the line. Without a pull system, the MRP would order to pick up all the necessary parts at once on the line (in the finished goods warehouse, more likely). What the pull system does is consolidate the load in the truck preparation area at the same rhythm as parts are used on your customer’s assembly line.
Why a physical leveling board and not simply instructions to the material handler via some electronic device? Kaizen and visualization, of course. The leveling box is a learning tool to teach us to do two rather difficult things:
- Consolidate truck loads to guarantee complete shipments
- Transmit takt time to the lines to level production
We need to learn to do this because it’s far from obvious. Truckloads are dictated by customer’s requests. In an ideal world, they’d be perfectly smooth, but we all know that customers change their minds constantly and at the last minute, so truck loads tend to vary a little in volume (most customers do try to fill their trucks) and a lot in mix (aaargh!). On the other hand, takt time is a leveling device: it’s the total daily operating time divided by the total daily production requirement. The production requirement is issued out of a leveled plan that averages out customer demand per reference and spreads it evenly by day. Basically, these numbers correspond more or less and we need to learn to manage this more or less ever more finely in order to 1) guarantee shipments as well as 2) maintain an even pull on the line.
The assumption here is that customer variation is not that wild. Particularly when the customer is an assembly manufacturer, he’s got a plan as well, and although at tier two you might be experiencing a lot of spot demand variation, experience shows this tends to be created variation – variation created by the supply chain mechanisms themselves.
The basic idea behind leveling the production plan is that what the customer has not picked up today, he’ll pick up extra tomorrow. Conversely, what he’s asking for extra today, he won’t ask for tomorrow. On average, customer demand for any part tends to be fairly stable, because (barring seasonality or promotions) market behavior tends to be stable. A given size of market will have a given rate of purchase replacement, and so will buy products at a given rhythm. There are all sorts of reasons volume varies from one day to the next, but the law of large numbers works in our favor. On large volumes, variation is actually quite small in percentage, and, more than anything else, it’s steady.
The leveling board, therefore, represents first the leveled production plan – the weekly customer demand averaged by day – in order to pull smoothly and in small quantities on the production lines – but also needs to take into account the real customer order for this truck. The leveling board should visualize start and end of truck preparation.
So if the customer orders more parts than the plan, where do they come from? Or if they order fewer parts than the plan, where should they go? Our assumption is that if they order fewer parts than the average planned, we will still build them. Certainly, this smacks of overproduction, but the belief is that if the customer doesn’t want the parts today, he’ll want them tomorrow on top of the averaged demand – and we wouldn’t be able to deliver because of fully loading the production line. In essence these are not parts in stock, but parts in waiting for a delayed order.
In order to learn how to deal with the discrepancies between the leveled production plan and the real truck order, we will start by visualizing the situation: we will set up a buffer stock, a “pool stock” in lean parlance which is not an inventory, but a waiting area for crates corresponding to the variation of created demand.
So, if the customer is asking for nine crates instead of ten today for truck number one, the tenth card will be placed in the leveling board, with a sign saying that the destination of the parts is not the truck preparation area (which is now complete) but the pool stock. Conversely, if the customer asks for eleven crates rather than the average of ten, the eleventh card will appear in the leveling board with an instruction to pick it up in the pool stock.
Inventory sitting in this pool stock is tracked part by part with the kanban cards, which get their own board. Variations in the pool are the key to understanding real demand and customers’ behavior. Part by part, the amount of crates in the pool stock should be fiddled with endlessly to stick as closely as can be to real customer behavior. This stock is not, well, a stock so to speak, but a tool to understand the real nature of customer demand. For instance, it should be distinguished physically from any safety stock, which is kept to protect the customer from fire, strike, machine breakdown or a comet hitting the plant.
Handling the pool stock is not easy, but great kaizen material. You’ll discover there are many combinations of things going wrong, such as trucks not arriving on time (or with no set schedules), parts going to different customers (and at different takt times: don’t put them together, but treat them separately), or not receiving firm orders until the very last minute (fill the leveling board according to the leveled plan, and complete it as new information comes in). But that’s the fun of it.
At the gemba, check that your leveling board actually reflects trucks. In practice, most people who work with leveling boards are more focused on their impact on production (withdrawing from the line’s supermarket), and not on consolidating the truck (and delivering exactly to customers). I suspect the key to unlock your question lies in using the pool stock to absorb the difference between the leveled plan and the actual truck load. Make sure the leveling board visualizes both dimensions of the problem and that the people in shipping understand the kaizen they have to do on that matter, and I am willing to bet that your OTD will improve visibly.