A note on method first: the orders of magnitude below come from the diagnostics we run and exist to calibrate expectation, not as a promise. The only way to know your own number is to measure your own operation. The purpose of this note is to hand you the five right measurements, so you arrive at any vendor conversation already knowing where the money is.
1. The gap between the event and the record
In nearly every mid-market plant there is a structural delay between what happens on the line and what enters the system. The stoppage was at 14:00, the entry was made at 18:00, the reason was picked from memory out of a list of twenty, and next week’s root-cause analysis works on that record.
How to measure: take a sample of 30 downtime entries from last month and compare the event timestamp with the entry timestamp. Also measure how many use the reason "other" or the first item on the list.
Order of magnitude: when more than 20% of entries use a generic reason, your availability analysis is blind to the largest single cause of loss. The gain does not come from buying a new system, it comes from reducing the friction of recording at the point where it happens.
2. Rework that is never classified as rework
Scrap has a code, appears in a report and someone answers for it. Rework, in most operations, is absorbed: the part comes back, someone adjusts it, the order closes, and the cost dissolves into labour hours and machine occupancy that show up as normal production.
How to measure: compare hours booked per production order against the standard time for the same order, and look at the distribution rather than the average. The tail of orders that consumed 1.5 times the standard is your unclassified rework.
Order of magnitude: this is the point that most often surprises the executive team, because the information already exists in the system and has never been cross-referenced this way. Revealing it requires no technology investment, it requires one cross-reference.
3. The per-product margin nobody actually calculates correctly
The company knows its company-level margin and believes it knows its per-product margin. In practice, indirect cost allocation is a historical average set years ago, setup cost is not attributed to whoever causes it, and the cost of idle inventory never reaches the item that generated it.
The predictable result: the sales force is incentivised to sell exactly the items that destroy margin, because those items look profitable on the price list.
How to measure: take the five highest-volume products and the five highest-priced ones, and recalculate margin attributing setup, rework and inventory turns to the item. Compare against the official figures.
Order of magnitude: when the recalculation inverts the profitability ranking of any meaningful item, correcting commercial policy is usually worth more than any industrial efficiency project in the queue.
4. Procurement reacting instead of planning
Without reliable demand forecasting, procurement works reactively: buy too little and pay expedited freight, or buy too much and tie up cash. Both errors are invisible on the P&L, because one becomes logistics expense and the other becomes inventory on the balance sheet.
How to measure: total expedited freight and emergency purchases over the last 12 months, and cross-reference against the inventory coverage curve for class A items. If both ends are high at the same time, the problem is not negotiation, it is forecasting.
Order of magnitude: this is where applied AI produces the fastest and most measurable return in mid-market manufacturing, because the required data usually already exists and the effect shows up in cash within one purchasing cycle.
5. Critical knowledge living in three heads
The adjustment only one operator knows how to make. The sequencing rule that exists in the supervisor’s experience. The alternative supplier’s contact sitting in one person’s phone. None of it is a problem today. All of it is a problem on the day of the absence.
How to measure: list the ten processes that stop the operation if they fail and, for each one, how many people can execute it without asking anyone. Count how many have the answer "one".
Order of magnitude: this one is not measured in margin percentage, it is measured in risk. It is the leak that costs nothing until it costs everything, and it is the only one of the five whose fix depends more on documentation discipline than on technology.
The pattern running through all five
In none of the five is the answer to buy a system. In all five, the data needed to see the problem is already inside the company, and has never been cross-referenced the right way by anyone with the time and the mandate to do it.
That is why we insist diagnosis comes before investment. An operation that measures these five before hiring technology negotiates better, chooses better and frequently discovers that two of the five resolve with no project at all.
How to use this note
- Pick two of the five, the ones that most resemble your reality.
- Run the measurement described, using a closed month of real data rather than an estimate.
- Take the result to your next executive meeting with no conclusion attached, just the number.
- Only then assess vendors. The conversation changes completely when you arrive holding the measurement.
If you would rather run the five measurements with support
That is precisely the content of week 2 of AI Inception, the X-Ray. But you do not need to hire anything to start: the 90-minute session with a partner costs nothing and serves to pick which measurements make sense in your operation.
Book the 90-minute session See AI Inception →