Custom manufacturing is an attractive business to be in, whether you’re founding a manufacturing startup or expanding a mass production factory into the personalization market. Mass customization of products—like items engraved with customer-specified text or a logo—is an entirely different arena. Rather than being just another lookalike item, your customers get to make themselves part of the product story and form a stronger connection to it. These products are more likely to be seen as artisanal or unique, even if you’re a large company, and customers are more willing to pay a premium for something they get to “make their own.”
While these benefits are alluring, they come with corresponding challenges on the implementation end. Designing lean manufacturing systems to manage mass customization will have all of the usual software challenges, but there are three winning strategies that companies can apply to ensure their systems operate efficiently: 1) Iterating over the manufacturing process and software system in tandem, 2) evaluating workflow exceptions and choosing how to deal with them, and 3) planning for automated handling of variable production bottlenecks.
Lean manufacturing takes iterating in tandem
When pursuing mass customization, the new production process should be developed in an iterative manner along with the custom software system. This is not a new idea for either startups or existing operations— manufacturing pioneered the “lean” concept and software arrived at the Agile revolution. However, it’s important to iterate over both process and systems to reduce the shakeout period when bringing a plant online, especially if other parties are developing the equipment or software. While it’s tempting to assume something will work in practice the way we design it in theory, it often doesn’t. It’s worth the time to design a small part of the production process, test with measured observation, and refine to smooth out issues, rather than assemble the complete line and software system and try to resolve the interdependent issues that arise, all while maintaining the projected capacity of the facility. “Big bang” integration can be painful enough in a pure software system, but the financial impact only worsens when factory workers and expensive machinery also have to be put on standby.
Handling exceptions: Pay now or pay later?
While creating the custom software system for the line, one design facet that adds development cost is dealing with exception cases. Detecting an error and notifying the user is relatively easy, but what can take considerably more time is designing an alternative workflow for processing items in an exceptional state. Deciding how to best handle these cases is key to lean manufacturing.
As an example, consider a workflow for engraving a widget that might include:
- Assigning a serial number
- Activating the engraver
- Instructing the factory operator to put the engraved widget back on the conveyor
The computer system won’t know when the engraver doesn’t physically mark the widget properly. Either the computer system has to provide an alternate flow for the floor operator to record a failure and remove the faulty widget from the current order, or the floor operator will have to ask a system administrator to update the system on his or her behalf.
While it’s possible to save substantial amounts of up-front development time by only supporting the “happy path” workflow, it comes at the cost of extra operational support needed to deal with the missing features. The cost is not just from worker time spent resolving issues; there’s often a larger organizational impact: Either the company has to hire and train support staff to fix issues, or burden the existing staff with the responsibility. In many small organizations or startups, this existing staff may be your key developers who are now distracted by frequent interruptions from the floor. At best, they will complain, but more likely they will quit, increasing staffing and turnover costs and detracting from general morale.
Before you decide to try saving time or money by A) not designing an alternative workflow for a potential exception or B) overdesigning an alternative by assuming the factory will halt if you don’t, analyze the effort your staff will expend in dealing with that particular exception. It can be difficult to accurately gauge the likelihood of a scenario, but it’s usually pretty easy to assess the impact. Do the math before deciding.
If you do decide to pay later, be sure to validate your assumptions by recording and tracking exceptions that your staff has to deal with on the manufacturing floor. Over time, you’ll either learn that you’ve made the right choice, or that it’s worth going back and investing in alternative workflows to lessen support overhead.
The variable bottleneck in mass customization
There’s a well-known adage in manufacturing that you can only produce as fast as the slowest point on the line. Even business professionals outside of manufacturing are often familiar with bottleneck constraints through Eliyahu Goldratt’s lauded book, The Goal. When mass-producing identical goods, bottlenecks on the production line are generally static and easy to identify. One of the challenges typically introduced by large-scale manufacturing facilities going custom is dynamic bottlenecks.
Identifying the variable
Imagine a mass manufacturing scenario in which it takes one minute to engrave a widget, and five minutes to pack a case of 24 widgets. When producing an order of 1,000 widgets engraved with a company name or sports team logo, it would be difficult for 20 workers engraving 100 every five minutes to outproduce five workers packing 120 every five minutes. The bottleneck is at engraving. Suppose that you finish making your 1000 widgets, and start manufacturing single item orders with individuals’ names engraved. Each of these is going to get put into a small, individual box that takes one minute to pack. Now you’re engraving 100 individual widgets every five minutes, and only packing 25. The bottleneck has shifted, and the production line will get swamped simply by changing the type of orders you’re fulfilling throughout the day!
Breaking down the bottleneck
While there are degrees of sophistication in how the software system deals with this challenge, observability is a prerequisite. My colleague Yogi wrote an excellent article on building observability into your systems that’s incredibly useful for startups building up their processes or established organizations breaking them down into targeted chunks. The short version is, IT organizations across all industries are being called upon to support reporting closer to real-time rather than the traditional overnight report, and there are many solutions ranging from free to enterprise-grade tools to achieve this.
Once you have observability in place, a relatively easy prevention measure that more than pays for itself is overflow detection and alerting. While many bulk manufacturing facilities have this as a safety measure, it is especially valuable to be automatically notified when and where a large bottleneck is detected. This frees up factory supervisors to focus on activities that are more valuable than carefully watching for pile-ups on the floor.
Moving up the feature chain is automated prediction. Rather than simply detecting when a pile-up has occurred, the system can combine past metrics with the current operating conditions to determine the likelihood and immediacy of a bottleneck. Making these predictions allows the factory to adjust its work before any serious balance issues occur.
A carefully designed system can also aid in preventing exceptions. In addition to building in observability and storing past data for prediction, it can be engineered to run workload simulations. The simulations would look at the current orders to be fulfilled, and project the different results with a variety of factory floor configurations and adjustments. This allows the system to recommend the best strategy for staffing the various stations on the floor, maximizing production while avoiding unnecessary bottlenecks.
Designing for simulation requires careful forethought for a new system, and can involve extensive rewriting of an existing one. The software must allow current orders to be read without being affected, and then completely simulate the entire factory process without actually doing anything in order to comparatively score all possible approaches. Simulation has the potential to drastically boost the productivity of a floor that experiences dynamic shifts in the type and location of possible constraints.
If you don’t invest in automated handling of variable bottlenecks, you doom your floor supervisors to deal with it themselves—carefully monitoring the state of the floor, maybe creating standard operating procedures to process orders in a particular sequence, and manually adjust the floor configuration throughout the shift. While this approach can work, it will cap your productivity (and your profits) by requiring humans to learn how to make fine-grained scheduling decisions, guesstimate how to adjust the staffing between the different operator stations, and stay tied to the floor in order to detect potential backup catastrophes.
Having automated metrics and predictive capabilities, on the other hand, will help more accurately forecast capacity under a variety of order configurations. This is critical to maximizing your production, especially during peak consumer buying seasons like the December holidays.
“Implement metrics to understand potentialities; design systems to react to those in reality.”
Thinking through these aspects may seem daunting and make you wonder if bootstrapping a new mass customization line can be successful, or question if your legacy system is really as productive as it could be. Mastery is possible. Iterating over both the manufacturing steps and software system in tandem is a workflow that, like any new habit, can be introduced slowly with a simple reminder system or basic updates to standard operating procedures. For workflow exceptions, a manufacturing startup may have difficulty guessing in advance the frequency of occurrence, but the impact should be fairly understandable and will dictate the “handle or don’t handle” decision that can be revisited later. Existing facilities can either review their incident reports, if they have them, or ask their line supervisors if they don’t. Workers remember very well frequent problems with machines they rely on to do their jobs! Finally, automatic handling of variable bottlenecks can seem unachievable, but it doesn’t have to be an “all or nothing” implementation. Start with observability and simply collect baseline metrics at first. Then gradually add in detection, prediction, and prevention capabilities, with each building on the other to achieve a truly lean manufacturing system. Be patient and take the long-term viewpoint—your customers are discerning and paying a premium for a customized product, which only multiplies the value that your careful custom software investment will provide.