Yield was 96% yesterday
That number, on its own, tells an engineer almost nothing actionable. It doesn’t say whether 96% was a good day or a lucky one, and it doesn’t say what would need to change to make it 98%. The engineer still needs to know: Why wasn’t it 98%? What was different about the 2% loss: a specific product, shift, piece of equipment, operating condition? The number itself isn’t enough; contextual investigation is how those questions get answered, and it’s the difference between reporting a result and understanding one.
Where yield and material loss hide
Losses are frequently distributed across many small events rather than appearing as one obvious failure. That’s precisely why they survive inside aggregate reporting for so long; no single event is large enough to trigger a review. Common sources include:
- Washdown and cleaning cycles
- Startup/shutdown transitions
- Product transitions and changeovers
- Overfilling and giveaway
- Rework and off-spec production
- Process instability and equipment behavior
Each of these can look routine in isolation. The investigative question worth asking is not just, “Did this happen?”, but “How consistently does it happen? And is the pattern the same every time, or is one recurring version of it quietly worse than the rest?”
Start with a production event
Rather than reviewing a month’s worth of aggregate data, narrow the investigation to comparable operating events (i.e., a batch, production run, shift, cleaning cycle, or startup/transition). Event-based analysis is what makes the comparison meaningful: two batches making the same product, under the same nominal conditions, are a much more useful comparison than a monthly average against a prior monthly average, because the event boundary removes most of the noise that would otherwise obscure the difference.
Compare good runs and poor runs
With a defined event in hand, the useful questions become: What separates high-performing runs from low-performing runs? Which signals behave differently, and when does the divergence occur – at startup, mid-run, or during a specific transition? Is the difference repeatable, and does the same pattern show up across other products, shifts, or equipment?
Answering these well requires holding several runs side by side on a common timeline, not scanning them one at a time within separate reports.
Add the process context
One signal rarely explains a loss on its own. A real investigation might require material flow, density, valve state, pump behavior, production phase, and cleaning event data together, aligned to the same window of time. The objective isn’t to correlate everything with everything; it’s to establish enough context to identify a meaningful relationship, and to rule out the conditions that turn out not to matter as quickly as the ones that do.
Example: Tracing material loss through process conditions
In a recent webinar, Marzetti shared how they were able to reduce waste in an egg yolk process by 66% by following the contextual analytics methodology. Material usage in the process varied from run to run, but the pattern wasn’t visible in standard production reporting. It only became clear once the team began comparing individual runs against each other rather than against a monthly average. In practice, this involved:
- Observation: Material usage showed unexpected variability run to run
- Investigation: Process data was examined alongside density and flow-rate conditions
- Context: Product being flushed during overnight system washdowns
- Improvement opportunity: The relationship created a specific, actionable operational target
- Reported result: 66% Waste Reduction
The washdown behavior had been present in every run; it simply hadn’t been isolated as distinct, addressable condition until it was examined in the context of density and flow rate. That’s the pattern worth generalizing: the loss was never hidden in the data, it was hidden in the aggregation.
Look beyond material loss
The same contextual approach extends naturally to cycle time variation, over-cycle impact, utility intensity, quality variation, equipment behavior, and rework. In each case, the mechanics of the investigation are the same, even though the loss shows up on a different line of the P&L.
Ecolab applied a Golden Batch framework to define and pursue consistent, better-performing operating patterns, producing a 7% increase in total batches and a 10% decrease in over-cycle time impact – worth more than $1M in potential value across the fleet – without adding new equipment. Batch reactor cycle times at Ecolab had expanded from roughly 5-20 hours to 13-36 hours as the product portfolio grew more complex, with up to 12 hours of variation between runs of identical products. That range is easy to miss in a monthly cycle time average because the long runs and the short runs cancel each other out mathematically, even though they represent very different operating realities. Applying a Golden Batch framework improved capacity from simply operating the same equipment more consistently.
In another example, Solvay cut raw material waste by roughly 30%, increased OEE by 10%, and represented approximately $400K in savings per year by moving to a condition-based washing practice for critical filters. Fixed-schedule filter washing was producing roughly 30% precoat material waste on filters that didn’t yet need cleaning, while filters that were already inefficient kept running and quietly dragging down OEE. The calendar-based schedule was wrong in both directions at once, and neither error was visible from a single OEE or material consumption metric alone. Moving to condition-based washing, triggered by actual filter condition rather than a fixed calendar, led to significant gains in material cost and operating efficiency. The broader lesson: any maintenance or changeover practice still run on a fixed schedule rather than a measured condition is a candidate for this same investigation.
From one investigation to a repeatable workflow
A successful investigation shouldn’t disappear into someone’s notebook. Every proof point above started when a single engineer noticed something in one dataset. The value only became tangible once the finding was turned into something reusable: a standardized definition of the event, a shared method for the comparison, and a way to keep watching for the same condition going forward. Organizations should build toward reusable analyses, standardized investigation methods, shared definitions, ongoing monitoring, user enablement, documentation, and governance.
That progression from a single investigation to a repeatable, monitored practice is the bridge between understanding a loss and acting on it consistently. Investigation answers “what happened and why.” The next capability an organization builds is the ability to recognize the same condition early enough, and reliably enough, to intervene before the value is lost again; turning today’s finding into tomorrow’s alert, and eventually into an operating routine the team doesn’t have to rediscover from scratch each time.
It’s worth tracing this back to a single thread: process variability becomes operational behavior, which shows up as yield, material, cycle time, or quality loss, which drives production and resource impact, and ultimately surfaces as a financial number. In the examples shared, an engineer walked that chain in reverse: starting from a cost and tracing it back to the process behavior that caused it. Seeing that full chain, rather than just its two endpoints, is what a contextual investigation makes possible.
Where does your organization stand today?
When investigating yield or material loss, ask: What event are we analyzing? What does good performance look like? What changed when performance deteriorated? What process context explains the difference? Can we validate the relationship? Can we turn the finding into a repeatable improvement method?
Take the next step with the Yield & Resource Loss Self-Assessment, a short set of questions covering material loss visibility, batch comparison, cycle time variability, connecting yield loss to process conditions, contextualizing utility and resource losses, distinguishing normal from avoidable variability, and repeatable investigation methods. Get a “gut check” to help you identify where to focus investigations first.