Every plant has a number nobody tracks: the time between a question being asked and the answer arriving. Why did OEE on Line 3 drop last Tuesday? Which shift is generating the most rework? What changed in the furnace cycle before the energy bill spiked? These questions get answered eventually. Someone pulls data from the historian, someone reconciles two Excel sheets, someone schedules a review meeting. The answer arrives in three days, a week, sometimes at month-end. The answer is usually correct. The problem is everything that happened while you were waiting. We call this gap decision latency — the elapsed time between “we have a question” and “we can act on the answer.” It doesn’t appear on any P&L line. But it quietly inflates three numbers your finance team already watches: lost output, scrap cost, and energy cost. And it degrades a fourth thing that’s harder to measure but more expensive than all of them — the quality of decisions made on stale information. This piece puts arithmetic on each one. The figures below are illustrative; the method is the point. Substitute your own numbers and the conclusion tends to get more uncomfortable, not less. 1. Lost output: the problem keeps running while you investigate OEE losses are not events. They are rates. A line running at 62% OEE instead of its demonstrated 74% isn’t losing output once — it’s losing output every hour until somebody identifies the cause and fixes it. This is what makes decision latency expensive in a way most cost reviews miss. The cost of a problem is not the size of the problem. It’s the size of the problem multiplied by how long it persists. And how long it persists is largely determined by how long the answer takes. - Demonstrated OEE: 74%. Current OEE: 62%. Gap: 12 points. - At a theoretical output of 100 units/hour, that gap is 12 units/hour. - At a contribution margin of 500 per unit, the line is bleeding 6,000 per hour. - Two shifts a day, that’s roughly 96,000 per day — every day the cause stays unidentified. If the answer involves exporting data, merging it with quality records, and waiting for the weekly production review — the honest answer is three to seven days. At the rates above, a five-day investigation costs nearly half a million in margin before anyone has even started fixing anything. The engineers are not slow. The investigation is slow, because the data lives in four systems and someone has to manually stitch it together before analysis can even begin. Your team’s diagnostic ability was never the bottleneck. Their access to the evidence was. 2. Scrap: every hour of latency produces more of it Scrap has a property that makes it uniquely sensitive to slow answers: the process that creates it doesn’t pause while you investigate. If a parameter drift, a tooling issue, or a material batch problem pushes your rejection rate from 2% to 5%, the line keeps producing — and keeps rejecting — at 5% until the cause is found. Unlike a breakdown, which announces itself and stops production, elevated scrap is silent. - Line output: 2,000 units/day. Normal rejection: 2% (40 units). Drifted rejection: 5% (100 units). - Excess scrap: 60 units/day. At a fully loaded cost of 800 per unit, that’s 48,000/day. - Detection lag: 2 days → 96,000. - Investigation lag: 4 days → 192,000. - Total cost of one drift event: ~288,000 — of which the fix itself might have taken twenty minutes. The repair was cheap. The latency was expensive. Six days of cost, twenty minutes of correction. Multiply by how many drift events a plant sees per year and scrap latency alone often exceeds the annual budget of the quality department tasked with preventing it. 3. Energy: the cost that drifts quietly and gets reviewed monthly Energy is the slowest feedback loop in most plants. Consumption is metered continuously, but it’s reviewed monthly — when the bill arrives. A compressor leak, an idling furnace, or a chiller running outside its efficient band can persist for three to five weeks before anyone is even prompted to ask why the number went up. - Monthly energy spend: 4,000,000. A 6% drift: 240,000/month. - Built-in detection lag (bill cycle): up to 4 weeks. - Attribution lag (which machine, which shift, which state): 1–2 weeks. - A single drift episode found six weeks after it began: ~360,000 — for a fix that was likely a setting, a schedule change, or a maintenance job. Energy waste is the purest example of latency cost because the waste itself is usually trivial to stop. Nobody debates whether to fix an idling furnace. The entire cost is the time it took to see it. 4. Stale decisions: the cost that doesn’t show up anywhere The three costs above are at least computable. The fourth is worse because it compounds invisibly. When answers take days, decisions get made on old data — leadership steers with a delayed picture, and the delay is exactly equal to the decision latency. And the quieter damage: questions stop being asked. When every question costs someone two days of data assembly, people ration their questions. The plant doesn’t just answer slowly; it gradually asks less. The improvement opportunities that would have surfaced from those unasked questions never enter the pipeline at all. Sizing your own latency tax You don’t need a study to estimate this. Take your OEE gap during a deviation, your theoretical output, your contribution margin per unit, your operating hours, your average time-to-answer, and your number of deviation events per quarter — and multiply. The interactive calculator in this article does the arithmetic as you move the sliders. In our experience, leadership teams that run this exercise rarely arrive at a small number. Collapsing the latency Almost none of this cost comes from missing data. The historian has the OEE story. The quality system has the scrap record. The meters have the energy trace. The data was always there. The cost came from the assembly time — the days spent pulling, merging, and formatting it before anyone could reason about it. That assembly time is what Prompt to Insights (PTI) removes. PTI connects to the industrial data sources you already have and lets your team ask operational questions in plain language — “Compare Line 3 OEE this week against last month, by shift” — and get the insight back in seconds, as a chart or a table, drawn from live data. The same questions that took a three-day data assembly exercise become a thirty-second exchange. Nothing about your team changes. The engineers who were always capable of finding the cause simply get the evidence at the speed of the question. The latency multiplier in every calculation above collapses toward zero. The slow answer was never free. It was just unbilled.