Manufacturing Metrics that Matter: How to Optimize Your Operations

Improvement and optimization efforts in manufacturing processes require a clear definition of measurable aspects to optimize. This finite set of manufacturing metrics: yield, cycle time, cost, on-time delivery, and throughput, supports business needs and customer expectations. In many cases the production goals could be in competition with each other. The article addresses the different metrics and provides a guideline to keep customer on-time delivery expectations in mind.

Improvement and optimization efforts require clear definition of measurable aspects to optimize or cooptimize, subject to constraints. Multiple sets of manufacturing metrics, or key performance indicators (KPIs), ranging from 15 to more than 30 metrics have been commonly proposed^1,2,3 to track and assess project health. For modeling and optimization, it is desirable to focus on a vital few. The dual criteria of necessary and sufficient can be applied to ensure that the smaller set of few vital metrics assesses and satisfies expectations of the key stakeholders for manufacturing— primarily the internal business expectations and external customer expectations. Focusing on these quantitative metrics enables us to generate a concise and digestible dashboard to quickly assess and respond to continuously evolving project health indicators. Qualitative metrics to augment these vital indicators will be addressed in a secondary article.

Internal business expectations are generally financially driven—that the manufacturing operations must support the financial expectations of the business in terms of revenue and profitability. The external customer expectations primarily involve quality and delivery: delivery of the promised quantity of the expected quality of the end product at the promised time.

Starting with the internal business expectations, revenue is closely linked to manufacturing throughput. Eliyahu Goldratt proposed a business approach, throughput accounting that highlights this linkage.⁴ Similarly, profit for the business is closely related to manufacturing throughput multiplied by the difference between the price and cost per unit. Manufacturing yield also has a strong influence on unit cost, since scrapped units impact the business financially while not providing revenue. These financial metric linkages indicate that three key manufacturing metrics are throughput, unit cost, and yield.

External customer expectations of quality also relate to manufacturing yield. Assuming that customer expectations have been translated into product requirements, and the products are tested or inspected or otherwise assessed against those product requirements that reflect the customer expectations, units that do not meet those requirements are scrapped—consequently, manufacturing yield is closely related to quality.

Customer expectations of delivery relate to manufacturing yield and manufacturing cycle time. The delivery of the promised quantity of product depends on the manufacturing line consistently achieving yield expectations. The delivery of the product at the promised time depends on the manufacturing line achieving cycle time expectations. These customer expectations add two more metrics resulting in five key manufacturing metrics: throughput, unit cost, yield, cycle time, and on-time delivery.

Based on this discussion, these five proposed manufacturing metrics might be sufficient to satisfy both business and customer metrics; it is reasonable to question whether each is necessary, or if any of these are somewhat redundant. For example, could on-time delivery be removed as a manufacturing metric, since it is a function of yield and cycle time? Alternatively, could cycle time be removed since it seems to be connected with on-time delivery?

A quick thought experiment can explore these questions: is it possible to have good yield and good cycle time in manufacturing, and yet have poor on-time delivery to customers? Yes, if the manufacturing area overestimates yield and underestimates cycle time when it makes promises for delivery dates. Is it possible to have good on-time delivery to customers and still dissatisfy customers or the business with poor cycle time? Again, the answer is yes—if the manufacturing has very long cycle times, impedes them from making promised delivery dates, and meets those elongated delivery dates a high percentage of the time, the business will disappoint customers and delay revenues with those very long delays for deliveries, especially if competitors have shorter cycle times and therefore faster delivery.

We can therefore assume this set of manufacturing metrics is necessary and sufficient for focus and for cooptimization:

• Yield
• Cycle time
• Cost
• On-time delivery
• Throughput

Industrial engineering and manufacturing engineering can model each of these metrics or KPIs separately; however, this can lead to suboptimization, wherein optimization of one metric has an unfavorable impact on another metric. The ideal, then, is cooptimization, perhaps with a primary focus on the metrics that the business considers most important while setting constraints that limit the acceptable impacts on the other metrics.

Projects can be chartered towards achieving this cooptimization; for example, Lean Six Sigma and/or Design for Six Sigma efforts can develop charters for projects involving a manufacturing process. These metrics are intimately tied into the define and measure phase of define, measure, analyze, improve, and control (DMAIC) and help to define success for the project, for the team, and for the manufacturing process. While metrics also drive behavior and accountability, the metrics can also help the team assess the financial benefit of the project.

Three of these manufacturing metrics—yield, cycle time, and cost—relate to the famous aphorism, “Good, fast, and cheap—pick any two”. This humorous quote implies that there is a trade-off among these manufacturing metrics that cannot be overcome. The last term, cost or cheap, is often seen as the most problematic, but the goal of a business is generally more aligned with maximizing profit rather than minimizing cost. So, a more appropriate goal would be to cooptimize yield, cycle time, and profitability—pick all three.

Cycle time is directly relevant in terms of the responsiveness to customer requests for product and indirectly relevant in affecting the probability of on-time delivery: the manufacturer starts the material into the manufacturing line on a certain date and the product is delivered after a delay corresponding to the manufacturing cycle time. The distribution of cycle time can be approximated with a gamma distribution bounded by a lower threshold corresponding to the theoretical minimum cycle time. The cycle time is also affected by the percent utilization of the bottleneck step, as described by Kingman’s equation⁶ (see Figure 3), and by the throughput and work in process (WIP) inventory of the manufacturing line, as described by Little’s Law.^7,8 Cycle time can also be effectively modeled and optimized using discrete event simulation. See figures 2 and 4.

A subset of these manufacturing metrics are tied directly to financial benefit—gross margin, related to profit:

This equation requires the manufacturing costs to be allocated to fixed costs that are independent of throughput, and variable costs that rise with throughput.

The ratio of variable unit cost over yield suggests a means for cooptimization through optimization of yield. Increasing yield not only improves quality “good”, but also reduces the cost per good part and thereby increases the gross margin. See Figure 5.

Throughput can constrain the revenue obtained, and throughput is often, in turn, constrained by the bottleneck of the process.⁹ Little’s Law relates cycle time, WIP inventory, and throughput. Since throughput can limit revenue, and therefore profitability, and can also impact the cycle time, and therefore on-time delivery, ensuring sufficient throughput to meet the demand (and allowing for upside) is vitally important. This requires that the organization identify potential bottlenecks and either take actions or set up contingency plans for actions to mitigate the bottlenecks, including:

• Identify potential bottlenecks, both among suppliers and bottlenecks in internal manufacturing
• Ensure sufficient capacities to ensure that other suppliers will not become the bottleneck
• Ensure sufficient yield that a supplier will not become the bottleneck
• Ensure sufficient availability (uptime) for suppliers and internal manufacturing
• Anticipate and plan for foreseeable disasters (hurricanes, earthquakes) that can shut down a manufacturing site (internal manufacturing or supplier)

Improving yield also helps with the probability of on-time delivery, the probability that you will fulfill the trust of your customers by delivering the promised quantity of product by the promised date. When the manufacturing organization starts production, they can start more product than needed to allow for yield loss. If manufacturing yield falls below the assumed yield, fewer products will be delivered than committed, and the customer will be dissatisfied. See Figure 6.

There is a tendency to assume that yield is a constant; for example, if the manufacturing team knows that yield has historically run at 50%, they can start twice as much product and deliver the promised quantity. However, yield follows a distribution—typically a beta distribution, bounded by 0% and 100%; an average yield of 50% suggests that the yield can sometimes be below and sometimes be above the average of 50%. The relationship and analogy of a continuous beta distribution to a discrete binomial distribution indicates that the variation of yield will follow a parabolic function of the average yield: the variation of the yield distribution for a mean of 50% will be higher than the tighter yield distribution for a mean near or at 100% yield. Using the approximation of the yield distribution with a beta distribution and the relationship between the mean and the standard deviation, a manufacturer can start enough product to allow for the uncertainty of yield. See Figure 7.

Figure 8 illustrates the flow down of the manufacturing metrics. The screenshot of an Excel spreadsheet shows an implementation for an actual manufacturing line. The ability of the manufacturing team to visualize these manufacturing metrics, with constantly updated values and predictions, allows the team to plan, optimize, and react rapidly to issues that could impact financial results and customer relationships.

“The global microchip shortage has gotten a lot of attention, and it has halted or delayed the production of vehicles across the automotive industry.”¹⁰ Analog Devices, a major microchip supplier to the automotive and other industries, was part of this supply chain issue. Top executives asked the manufacturing organization to propose ways to improve; the Vice President of Manufacturing had a Six Sigma Master Black Belt certification and decided to give the manufacturing metrics approach a try.

The primary manufacturing metric he chose to focus on was on-time delivery. He and his team entered current information from a manufacturing site into an Excel workbook in which these manufacturing metrics were linked, and then used the Crystal Ball Excel add-in to bring in the variabilities and uncertainties for the information.

The results from the Monte Carlo simulation reflected a high variability of on-time deliveries that were consistent with ADI’s frustrations in delivering to customer orders and expectations (see Figure 9).

Sensitivity analysis from the Monte Carlo simulation indicated that the main source of variability was the availability (% uptime) for the test system used to test the microchips prior to shipping to customers.

The team was then asked to model the availability of the test system, making slight adjustments to a Bayesian model for availability.¹¹ The Bayesian model suggested approaches that could improve the availability of the test system and predicted the availability of the test system with those improvements. The team subsequently entered the improved availability into the manufacturing metrics model, and ran a Monte Carlo simulation—which predicted considerable improvement in on-time delivery of the microchips (see Figure 9).

Improvement and optimization efforts require a clear definition of measurable aspects to optimize or cooptimize. This finite set of manufacturing metrics: yield, cycle time, cost, on-time delivery, and throughput, support both business needs and customer expectations, and can be cooptimized to satisfy internal stakeholders, the business, and the customers.

These manufacturing metrics were flowed down and linked to equations that enable both optimization of the requirements and exposure to trade-offs and decisions needed for optimization subject to constraints.

The application to a manufacturing metrics improvement for a semiconductor company provides a case study demonstrating the value of this ability to drill down to specific decisions and approaches needed to improve customer and stakeholder satisfaction with the results.