How many batches do you test each year? What does each of those controls cost, sampling and waiting included? And what would a result obtained during manufacture, rather than after, be worth to you? These three answers are what a return on investment that fits your process is built from.
The price of the instrument is what weighs least. What decides is the cost of one control multiplied by the number of controls in the year. The same instrument pays back quickly over dozens of batches a year; over a handful, another reason carries the project, and it is worth naming.
A control breaks down into four parts
| Part | What it holds |
|---|---|
| Sampling | Operator time, opening the equipment, getting the sample to the lab |
| Analysis | Technician time, consumables, report: an assay in pharma, an oven or an extraction in food, a titration in chemicals |
| Waiting | The batch, the vessel or the line held until the result |
| Material | The product taken, which never returns to the batch |
The last two are the ones people forget.
PAT-INDUSTRY holds CIR and CII approval (2026-2028), and our invoices separate R&D operations from innovation operations. Funding the project.
This calculation has been run and tested on a real granulation case. With your figures, it can be redone in forty-five minutes. Book a call.
Four parts on the other side too, of a different nature
- The investment. The instrument, the mechanical adaptation of the measurement point, the wiring and the feed to supervision. Once.
- Development and validation. Governed entirely by the calibration burden: a sample campaign and a reference method, or nothing at all. Once, but the amount moves considerably from one calibration regime to the other.
- Maintenance. Annual contract, periodic verification, source consumable. Every year.
- Model monitoring. Reviewing the indicators, keeping the sample set, enriching it. Every year, and the part most often missing from budgets. What becomes of a model over time.
Orders of magnitude, hardware only
| Type of installation | Order of magnitude |
|---|---|
| Handheld spectrometer for identification at goods-in | €25k to €35k |
| Contact-free in-line NIR measurement point | €40k to €50k |
| In-line probe on qualified equipment | €30k to €90k |
Amounts excluding VAT, for hardware only. They include neither integration, nor model development, nor maintenance, which are the other items in this list. They vary with the configuration; the amount for your project is established at scoping.
Full project, services included: in the order of 1.5 to 1.8 times the hardware. That is what two real pharmaceutical projects show, from feasibility to validation and deployment. Excluding the client’s internal costs: reference analyses, line time, quality.
Two parts are paid once, two are paid every year
That is the whole difference between the two columns. Traditional control has recurring parts only: it costs on every batch, indefinitely. In-line measurement concentrates most of its spend at the start, then costs little and almost constantly, whether you make ten batches or a thousand.
Which is why the comparison only means something over a period written in advance, and why it swings sharply with throughput.
The variable that decides is the number of batches a year
The break-even is one line. On one side the investment and the development, paid once. On the other, what the measurement saves each year: the unit cost of a control avoided, multiplied by the number of batches, less the annual cost of the measurement itself.
That quotient carries a consequence nobody enjoys hearing. On a line running a few batches a year, no in-line measurement pays for itself on control savings alone. Where it is still justified, the reason is another one: a batch saved, a release brought forward, process knowledge that was missing. Then say so, and defend that reason rather than a table.
At the other end, on a line running a hundred batches, the question hardly arises. The annual saving exceeds the initial investment, and the decision moves to technical feasibility and to the level of regulatory ambition.
Between the two lies the zone where the calculation deserves to be done properly, because the answer there depends on assumptions open to argument. That is exactly where they have to be written down.
Run the calculation on your own figures
Two columns: what control costs today, and what it would cost with an in-line measurement. The tool assumes no gain on your behalf: you say what changes. Reworks stay the same until you change them.
Today
With an in-line measurement — your assumptions
The investment
Annual cost today
—Annual cost with the measurement
—Annual difference
—Investment, full project
—Payback of the investment
—In pharmaceuticals, replacing a control with an in-line measurement goes through method validation and, depending on the case, a variation to the dossier: this calculation assumes that step has been passed. Indicative order of magnitude; nothing is stored or sent, the calculation runs in your browser. The calculation on your real data is done at scoping.
Three assumptions to write before costing anything
A cost study is never argued on its result. It is argued on its assumptions. Better to set them out at the start, where they can be discussed.
- The hourly cost, and which one. It is what turns time into money, and it governs the sampling part as much as the analysis part. Ask finance, not the laboratory. And say which cost is meant, because two notions travel under the same word. The loaded cost is gross salary plus employer contributions — for a laboratory technician in France the common order of magnitude is thirty-five to fifty thousand euros a year. The fully absorbed cost adds premises, equipment, supervision, IT and quality. Taking one for the other changes the result twofold. Our own cost studies retain seventy to eighty thousand euros a year for a technician on a thirty-nine hour week, and that is the value used here. Depending on how your finance department aggregates the lines, it will reach you as a high loaded cost or as a fully absorbed one: that is the first thing to pin down, before the calculation starts.
- The real cost of an assay. Consumables for a chromatographic analysis come to a few euros per sample — two, in our assumptions. The column costs a thousand to fifteen hundred euros and is written off over its number of injections. Human time remains, by a distance, the dominant part.
- The value of what is taken. On an expensive active, a few doses per batch become the first line of the traditional column. On an excipient, that line disappears. It is the assumption that moves the result most from one industry to another.
Orders of magnitude used in our own cost studies, French pharmaceutical projects, 2023. They are working assumptions, not results: they exist to start a calculation, and the first task is to replace them with yours. Values drawn from your own site are also the ones that hold up best in review.
Three ways to get the calculation wrong
These are the three objections a cost study meets in review, and they are well founded each time.
Taking the maximum cost of an analysis for its usual cost
A difficult batch triggers extra analyses, an investigation, sometimes a second sampling. That case is real, it is not the usual case. Costing the traditional column at its maximum inflates the gap by a factor that does not survive the first serious reading. Take the average, and give the maximum separately.
Comparing two different scopes under one heading
An in-line measurement often reduces the number of development batches needed, because it gives continuous information where the laboratory gave a point. The gain is real, and it is large. But it belongs on its own line: folded into a cost line labelled as though the number of batches were the same on both sides, it becomes a bias instead of an argument.
Forgetting the period when both are paid for
Traditional control does not stop on the day of commissioning. It runs in parallel through the equivalence demonstration, then through whatever period of confidence quality asks for. That double spend is predictable and bounded, and it is missing from almost every calculation. What an equivalence has to demonstrate.
What the calculation does not say
A break-even built on control savings alone is a floor, not the value of the project. What escapes the table is often what decides.
- The batch not lost. A drift seen during the operation is corrected; seen at release, it is recorded. One batch recovered sometimes pays for the whole installation — but that event cannot be budgeted, only told afterwards. Unless reworks come back regularly: reworked or downgraded batches are then counted from the history — workshop time, lost materials, the cost of a customer complaint — and enter the calculation. That is the line that links a complaint to an amount.
- Cycle time. A blend stopped when it is homogeneous rather than at a fixed duration releases capacity. That gain counts in machine hours, and it belongs to production, not to the laboratory. Stopping an operation on a criterion rather than on a clock.
- What is learned about the process. A continuous measurement shows blending and de-blending phases that a final sample could not see. That knowledge serves beyond the control, and it is often what justifies an installation in development before production.
Conversely, a calculation that only holds thanks to those three lines is a fragile calculation. They strengthen a decision, they do not carry it alone.
The calculation covers one operation and one year. Yet the same measurement capability often serves several times: in development to explore conditions, at scale-up to compare two pieces of equipment, in validation, then in production. Each of these uses has its value. A case that adds them up describes the project better than one that keeps a single use, on one condition: the measurement stays the same from one stage to the next, with a model that transfers.
Frequently asked questions
From how many batches a year does the question arise?
There is no universal threshold, because the unit cost of a control varies by a wide factor between an excipient and an expensive active. The useful first step is therefore not to look for a threshold, it is to cost a single control honestly, part by part. The threshold follows from one division.
Should our own teams' time be counted?
Yes, and on both sides. Sampling and analysis time on the traditional side, monitoring and model upkeep on the in-line side. A study that counts internal time on one side only always produces the result it was looking for.
Does the calibration regime change the calculation?
It changes which part decides. A full calibration loads development with a sample campaign and reference analyses. A pure-component calibration moves that load onto qualifying a spectral library. The recurring cost moves little. What each regime asks for, and when each is legitimate.
Can an in-line measurement be justified without control savings?
Yes, and it is common in development, where the value lies in process knowledge rather than in an analysis avoided. The only requirement is to name it: a project defended on one reason and budgeted on another is judged on the wrong indicator a year later. What is measured a year after commissioning.
A rare but costly incident: how do you count it?
As a risk: its frequency multiplied by its cost. A batch lost once every five years weighs, each year, one fifth of its cost. In the calculator above, enter it as 0.2 batches reworked or downgraded per year.
The frequency comes from your deviation and investigation history, not from a guess. The cost covers material, line time, the investigation, and what the line did not produce in the meantime.
One condition decides everything: the measurement is only worth that amount if it would have seen the deviation in time to act. That is what a proof of concept establishes, on your process. Like the other lines in this section, this amount strengthens a decision; it does not carry it alone.
We have none of these figures. Where do we start?
With one batch and one control. The time actually spent sampling, the number of analyses run, the delay between sampling and result, and the mass that does not return to the batch. Those four readings are enough to place your case, and they are taken on an ordinary batch, with nothing instrumented.
The full cost of a measurement point, as industry has costed it
Two published sources give useful orders of magnitude, and they do not carry the same weight. We cite them with their origin, so that you can weigh them yourself.
A NAMUR working group (AK 3.6, 2017), bringing together chemical and engineering companies, holds to two orders of magnitude: the return on investment of an in-line measurement point usually falls between one and three years, sometimes less; and 50 to 70 % of the whole-life cost of a measurement point comes from maintenance, not from the purchase. Three levers follow: follow one global piece of information rather than each component, multiplex several points on one analyser, and think in life-cycle cost from the specification onwards.
A sampling supplier, Swagelok, offers a multiplication rule: the full cost of an analyser is about three times the hardware, four times once the system is commissioned, five times counting project management. This marker, drawn from a sampling specialist's experience, usefully complements NAMUR's.
Both markers point the same way as the breakdown above: the instrument is the most visible item and the least decisive. A costing that takes in maintenance and integration from the start holds up over time, and that is the one we build at the scoping stage.
Tell us how many batches you control in a year. We will tell you which side of the break-even you are on.
Forty-five minutes are enough: break one of your controls down part by part, set the assumptions that are missing, and see whether control savings carry the project — or whether it has to be defended on another reason.