Three Strategies To Deal With Incomplete Data In Production Scheduling

Day in, day out we have been reading blogs, articles and papers on data growth (e.g. Moore's law stating that the amount of data doubles every 2 years), new technologies enabling new insights into customer behavior, and lately also about the Internet of Things making us understand - in real time - the current state of entire supply and value chains. As compelling as these trends are, and as fascinating as the technologies driving these trends (or riding its waves?) are, in my view they often miss an important point. Typically, these "macro thoughts" are presented to us with references to large enterprise, and how they can benefit from catching the next wave. However, small enterprises typically seem to live in yet another universe than large enterprises and hence many of these new ideas miss the important point of the SMB reality. As much as all these new thoughts make us believe that we all have "all" data always at our fingertips, the reality speaks a different language at many SMB manufacturers. In this blog, I share some thoughts how SMB manufacturers still can propel on their production scheduling even if they have to face incomplete data.

Think I sound a bit cynical when saying that SMBs live in their own universe with (massively) incomplete data? Well, this is not my intention. I am just stating the obvious that we do not live in a perfect world and this is also true when it comes to job shops, machine shops and small make-to-order enterprises and the "wealth" of their today's data. The core consequence of this inevitable situation is that the planner has to deal with highly volatile throughput times. This results in hardly predictable

• Set up time
• Production times
• Tear down times
• Transfer times

#1 Schedule, not optimize

When it comes to assigning jobs and tasks to resources, when it comes to determining the best sequence of processing jobs and tasks, when it comes to delivery time commitments vs. resource utilization, when it comes to squeezing in rush orders, when it comes to taking into account production campaigns and avoiding set-up times, many folks initially hope to find a software where they can press the magic "solve the puzzle" button and get the perfect production schedule.

Good news: this software exists and it tends to get called APS (advanced planning and scheduling) software. Bad news is that it tends to be built on rather sophisticated and powerful optimization algorithms. Well, why is this bad news at all? Here is the thing: without explicitly stating, this kind of software implicitly expects a perfect and always current data input. A great optimization algorithm is pointless if it is fed with poor data, because garbage in results in garbage out. 

In a nutshell: in the described situation with highly volatile throughput times an APS just makes the user believe that he deals with an optimal planning strategy. But this optimum will never be realizable as data will be different in reality. Moreover – as APS are very complex and high maintenance software – this comes along with high running costs, low transparency and restricted time to react.

So my recommendation to SMB manufacturers is not looking at an optimization solution. Instead, walk before you run and start with applying some rather basic scheduling principles, which depend significantly less on perfect real-time data. These scheduling principles could (and should!) take into account

• Basics strategies like ASAP and JIT,
• task dependencies as well as
• job priorities and
• machine utilizations. 

The ideal software should be easy to use, means being always transparent, letting the planner understand what happens if and allowing him to react immediately and fast when he gets some information that makes the world a little more perfect. The resulting plan that you will get with applying these principles will be most efficient and constructive.

#2 Simplify, not complicate

So, obviously there is a software bit associated with production scheduling - even for SMBs with incomplete data. And many times, when software starts playing a role, an interesting phenomenon seems to happen. On the one hand, software developers typically are smart guys, and often they prefer building something super powerful, sophisticated and heavy-duty mind-blowing. Obviously, this is anything but simple. On the other hand, customers who so far have "survived" scheduling with pen and paper, a whiteboard or some rudimentary Excel spreadsheets start to think in long list of requirements when they start specifying a production scheduling application.

This is classical hen and egg scenario: Are customers as demanding as they know what can get achieved with software? Or is software that complex as customers are so demanding?

Anyway. Remember the garbage-in-garbage-out aspect that I raised above. Remember that we live in a world of incomplete data. Let's make things simple, not complicated. Of course, I refer this to both the software that is built as well as to the software that is required. Here are some ideas that might help you identifying a simple enough software.

• It is very visual, so that you can simply see your schedule and take actions "graphically" via drag & drop rather than flipping around through a hell of menus.
• It connects to what you are most likely used to: Microsoft Excel.
• It does not have massive "powerful" functions hidden in some really mighty right-click context menus.
• It does not request you to make settings and configurations that take you forever.
• It works as cloud based software, so that you can get it running whenever you want - without any installations.
• It is "self-sufficient" - so no training courses needed to understand it.

#3 Focus action, not analysis

Here is another thing that tends to happen in times of incomplete data: folks try to cope with this by analyzing all available data "to death". This often ends in a paralysis by analysis type of situation. There is no imminent point that better analytics will help you better schedule. Of course, understanding patterns and understanding past behavior is great, but a look in the mirror will never tell you what is ahead of you.

Rather than spending too much time looking back, use your data to look ahead (this is what scheduling is all about!) and make sure that your data are presented to you in an actionable form:

• They must be super easy to access and to intuitive to change
• Ideally, they give a kind of direction where you need to take action first. Just as example: which jobs will be late based on the current workload? 
• Being actionable also should mean: being "simulateable" so that you can simply build scenarios and then decided for the best. Seeing options for things to come is much more actionable than trying to predict things based on past patterns. 

Summing this up, I fundamentally believe that a "simple" software (in the above sense) can help SMB manufacturers easing their production scheduling even in times of incomplete data. With just plan it, we have the ambition to exactly provide this. Give it a try!