Four phases, four deliverables: structure, prove, deploy, sustain. Each deliverable keeps its value even if the next phase does not happen. Between two phases there is a decision, taken against criteria agreed before the trials: each phase can therefore close either way, and that is what lets a project be scoped one phase at a time.
Why a project is often bought as a single block
An in-line measurement project is often bought as a single block: the instrument, the integration, the model and the training in one commitment. Splitting it into four gives you three points at which to look at real data before spending the next tranche. Three things a preliminary analysis brings out early, and each of them is cheap to settle at that stage: where the process connection goes, how much scatter your reference method carries, and which line of the operating account the gain lands on.
The four phases, named by what they produce
Structure: scoping and diagnosis
Which measurement question is really being asked, how fast you need to decide, and where it sits. The gain is costed here, on your data.
Deliverable: a scoping note, naming the conditions that have to hold.
Prove: proof of concept (PoC)
Trials on your batches, at pilot scale or on the process wherever access allows, compared against your reference method, including on a difficult batch.
Deliverable: a report and a decision, argued against the criteria set at the outset. It is a complete report either way, with its data and reasons. Eight situations described, with their outcome.
Deploy: integration
Installation, connection to the information system, qualification, procedures. Where the proof of concept ran on the process, the model carries straight through.
Deliverable: a qualified system. Instrument, model, procedure and documentation, giving in production the results it gave in trials.
Sustain: training
The teams who run the measurement, maintain it, and justify a result to an auditor. Residual monitoring, detection of samples outside range, model updates.
Deliverable: a team that stands on its own. A model someone knows how to maintain is still in service long after the first year.
Engineering, modelling, documentation: the three streams of a project
Three streams run in parallel, and we work on all three. Engineering draws on your projects and maintenance: we specify and we follow through. Modelling is the part we most often carry; it needs your reference values. Documentation is written with your quality assurance and lives in your system. What each stream asks for.
Chemometrics runs through all four phases. A model keeps living once the project is finished: your own team can hold it, which is what the Sustain phase is for, and we can also build it and maintain it for you. Both routes, and what each one asks for.
Chemometrics: who builds the model, and who keeps it
Building the model, choosing its pre-processing, bounding its range of validity, and knowing which sign calls for a rebuild. The most outsourceable part of a project, and the one we most often carry.
Integration: what the Deploy phase covers
From mechanical fitting to architecture: who receives the value, on whose clock, and who may act on it. A measurement that acts asks for more than a connection.
Method validation
Specificity, accuracy, precision, range, robustness, detection of out-of-range samples. What has to be established for a multivariate measurement to hold in front of an auditor.
What each phase decides, and what opens the next one
| Phase | The question it answers | Deliverable | What opens the next phase |
|---|---|---|---|
| Structure | Does this project deserve to exist, and on exactly which variable? | A scoping note | A measurement question in words, one or two candidate technologies, a mechanical access point identified, and an estimated gain that stays above the full cost of the measurement point |
| Prove | Does the signal exist on your product, in your conditions, and does it discriminate what needs discriminating? | A report, a decision to commit or stop, and a model built in deployment conditions | Performance reached against criteria agreed before the trials, on batches that represent the real variability, difficult batches included |
| Deploy | Does the measurement hold on production equipment, in routine, with the teams in place? | A qualified system | Batch-by-batch agreement in production, written procedures, and a control strategy that says what to do when the measurement falls outside its range |
| Sustain | What happens when we are no longer there? | A team that stands on its own | Nothing. This is the end of the project. Or the start of another process |
The column that matters is the last one. Satisfying it is what makes the next phase short, and it is agreed in writing before that phase opens.
A documented conclusion is a deliverable, whichever way it goes. Knowing exactly what a variable would take to measure in-line on your process is worth having in writing: the level of the signal, the heterogeneity at the scale of the sampled volume, the scatter of the reference method. The reasons are few and they can be named. That closes the question with its evidence, and it stays closed.
It is also the cheapest step in the method, by a wide margin, and it is the one that makes the deployment hold: the sensor in the right place on qualified equipment, a model with an owner, and operators who know what the curve on their screen means. What makes a measurement hold is known, and it shows early. Naming it is our trade rather than a manufacturer’s.
What runs in parallel, and what waits for the previous deliverable
A gate between two phases protects a decision rather than putting the project in a queue. What waits is the commitment to the next phase. A good deal of the work starts early, because early is when it is easiest to do.
Five pieces of work we deliberately bring forward.
- The proof of concept runs on the pilot or on the process. A model built under deployment conditions stays usable in production, and the Deploy phase becomes qualification and documentation.
- The transfer sample set is acquired during the trials, while the batches are there and the development instrument is in its original state. Carrying the model onto a second instrument is then a matter of days. What a transfer asks for.
- The monitoring plan is built during integration: indicators, frequency, thresholds, recalibration trigger, a named owner. It ships with the model. What keeping a model alive asks for.
- IT and computerised system validation come in at scoping. Consulted while the architecture is still open, they choose alongside you. Where the measurement lands, and who reads it back.
- The validation level aimed for is settled at scoping, because it governs the design of the installation itself: probe position, redundancy, traceability. The three levels of ambition.
The split states itself simply. Whatever depends on a result the previous phase produces waits for that result: a method is frozen once the measurement interface is known, a control loop is encoded once the dynamics are characterised. Everything else gains from starting early, and that is what makes a four-phase sequence shorten the calendar.
Two things experience has taught us
What decides a project is settled ahead of the instrument
The same handful of points decide it every time. How the sample presents itself to the probe. A sampling plan that speaks for the batch. Reference values that hold up. A qualification department brought in at the start. Operators who have been told what the displayed curve means. Laboratory performance has never been the deciding factor. What is prepared on the organisational side weighs more than the choice of spectrometer.
The FDA writes it in its own PAT Guidance: transferring a laboratory method in line is not automatically PAT.
Scoping is what makes everything after it go fast
The request is common and understandable: fit a probe and we will see. A week of scoping first turns that same line time into usable data, because three things are settled before the probe goes on. Which variable is actually followed, where the probe sits so that its own scatter stays small, and which reference values the spectra will be compared against. Trials built on a precise question conclude the first time.
What we supply at filing
An in-line measurement ends up in a dossier. That is where an analytical project becomes a regulatory one as well. The four points below are prepared during the earlier phases; they are handed over at filing.
The analytical elements of the dossier
Description of the method, validation data in the sense of ICH Q2(R2) and ICH Q14, the place of the measurement in the control strategy, the fallback plan. Those deliverables are produced during Prove and Deploy; what filing adds is handing them over in the form the dossier expects.
The scope, and what triggers a variation
This is the decision that governs the whole life of the dossier: a change within the approved scope is subject to GMP only, a change outside it calls for a variation. The EMA guideline on near infrared introduces that notion of scope explicitly. We write it with you before filing, and we handle the variation when one becomes necessary. On the US side the logic is the same with a different grammar: the change is declared in an Annual Report, a CBE-30 or a Prior Approval Supplement according to its impact, and that is the choice being prepared. The text, and what it asks.
Questions from the authority
After filing, questions on the analytical part are handled with us: what the assessor is asking, what the data already acquired answers, and what has to be produced in addition. A trial campaign designed with that moment in mind answers more often without new acquisition.
Exchanges ahead of filing
Scientific advice in Europe; in the United States, a formal Type C meeting with the FDA. The guidance that frames them keeps that category for subjects that do not fit the narrower formats: introducing an analytical technology onto an already registered process is one of them, since it belongs to no development milestone.
These exchanges gain from carrying the measurement explicitly, rather than leaving it inside the technical file. We take part alongside you, on that part.
What the four have in common: they are settled during the trials, not when the writing starts. That is why scoping puts the dossier question before the instrument question. Where does filing sit in your calendar?
QbD and PAT: two halves of the same job
Two trades complement each other on a PAT project: integration, which fits sensors and calibrates models, and consulting, which writes the Quality by Design dossier. We carry both on the same project. A critical quality attribute becomes real once someone can measure it. A measurement becomes a control once it is attached to an attribute.
Holding both ends is why we run a dedicated activity: ProcessControlExpert, an activity of PAT-INDUSTRY devoted to the joint strategy of Quality by Design and process analytical technology.
Why QbD leads mechanically to in-line measurement
This is a consequence, not a preference of method. The reasoning holds in two states.
A fixed process
Settings are decided once and for all. The raw materials vary, by batch, by supplier, by season. Variable input, fixed process: the output varies. The drift shows on the finished product, once the batch is made.
An adaptive process
The input variability is still there, and the process adjusts to what it receives. Variable input, adaptive process: the output becomes steady. That asks for the input and the state of the process to be seen in real time.
This is where measurement becomes structural rather than desirable. A process adapts to what it can see. QbD asks for an adaptive process, an adaptive process asks for continuous observation, and process analytical technology is what makes Quality by Design executable in production.
The regulators put the same chain very plainly. The FDA Guidance for Industry: Process Validation of January 2011, section II.B, sets out four expectations, and each falls on one side or the other:
| What the manufacturer has to do | What makes it possible |
|---|---|
| Understand the sources of variation | QbD |
| Detect the presence and degree of that variation | PAT |
| Understand the impact of variation on the process and on product quality attributes | QbD and PAT |
| Control the variation in a manner commensurate with the risk it represents | QbD and PAT |
The second line has no answer other than process measurement. It is the shortest argument we know, and it does not come from us.
What variability costs, and how to read that figure
The cost of poor quality is not a sector constant. It is a function of how well the process is controlled, and the scale reads step by step, in sigma level. At 3 sigma, cost of poor quality runs at something like 20 to 25% of revenue. At 5 sigma, 4 to 8%. At 6 sigma, 1 to 3%. On a hundred million in revenue, the distance from the first step to the last runs into tens of millions.
The slide carrying that scale is headed “So Why Have Pharmaceuticals Not Achieved 6 Sigma Manufacturing?”, and the next one answers in a single line: “We achieve 6 sigma quality using 3 sigma processes”. That is, by inspection and sorting rather than by process control.
Where that figure comes from, exactly. A PricewaterhouseCoopers presentation to the FDA Science Board of 16 November 2001, reproduced as such by M. VanTrieste (Amgen) in Pharmaceutical Quality System Elements: Continual Improvement of the Process (CAPA), ICH Q10 conference, October–November 2011, slide 11 — a document hosted by the FDA. It is a conference slide, not a study whose method you can read, and it is twenty-five years old. It gives a slope and an order of magnitude, not a measurement of your plant, and the sector average has risen since. We cite it because we can say where it comes from — exactly what the gain percentages circulating in the field lack.
What you can establish on your own data is a capability index — Pp, Cp, Ppk, Cpk, depending on whether you look at global or local spread and whether centring is taken into account. It maps onto sigma level, it is computed against your own specifications, and it tells you whether the process is off-centre, whether it is too widely spread, or whether the limits themselves are the wrong ones. It is what replaces the order of magnitude above with a figure that concerns you.
That is what ties the three trades together. The quality approach names the critical attributes and what governs them. Process measurement watches them in real time. Capability says where you stand, and how far you have moved. The calculation we do on your data.
Depending on your industry, this discipline goes by a different name
In pharma, bioproduction and chemicals it is called Quality by Design, and the vocabulary comes from the ICH guidelines. In food, agriculture, cosmetics and energy the same requirement is called operational excellence. It is measured in yield, in variability and in scrap rate.
The principle is shared: quality is built by understanding and controlling the process, rather than inspected in at the end. The two overlap without being identical, and it is worth saying so. Quality by Design is a design framework that an inspector can hold you to. Operational excellence is a performance discipline with no regulatory obligation. One requires you to document, the other requires you to prove the gain. We work in both registers, with the same measurement tools and a different vocabulary.
What QbD brings to the measurement
It says what to measure, and why. Identification of the critical quality attributes (CQAs) and of the process parameters that govern them, risk assessment, design of experiments (DoE), design space. That work is what puts the instrument on the variable that decides quality.
What the measurement brings to QbD
It makes the reasoning checkable. A design space that is not monitored in production is a document. Backed by an in-line measurement it becomes a control strategy, and sometimes real-time release testing (RTRT). What that means you have to demonstrate.
Our centre of gravity is the measurement, and we say so. In most projects and training sessions, Quality by Design serves as the frame. We name the attributes, place the measurement, and move on to what actually occupies us: making the process observable.
When a strategy has to be built in depth, it calls for a different trade. That is why this activity exists: it was set up jointly with a Quality by Design specialist. The client keeps one point of contact and one project. They get two competences that do not pass responsibility back and forth. Who works on your projects.
It is also what lets us say plainly which projects are worth running and which are already fine as they are. That arbitration takes both trades: one to judge it, one to check it.
Where this activity comes in
- At scoping: mapping the critical quality attributes and process parameters before any talk of technology. That is what puts the instrument in the right place. The Structure phase.
- When designing the trials: a DoE beats a campaign of samples taken as they come, and it often costs less.
- When writing the control strategy: the document that ties the attribute, the measurement, the criterion and the action together. That is the one an inspector reads.
- In training: the principles of Quality by Design, and how they connect to process measurement. The Sustain phase.
Where our four phases meet the QbD ones
Both breakdowns are right, and they do not describe the same object: QbD divides the development of the product, our phases divide the engagement. The correspondence holds in one line each.
| QbD phase | Phase of our method |
|---|---|
| Objective — QTPP, then the critical quality attributes of the product | Upstream of our engagement. Scoping starts from attributes already named. |
| Design — characterisation, risk analysis, control strategy | Structure. Scoping works out the measurement part of that control strategy. |
| Experimental — design of experiments, development and validation of the measurement | Prove, then the chemometrics of Deploy. |
| Production — running, real time release, keeping the model alive | Deploy, then Sustain. |
One practical consequence: where your critical quality attributes are not yet settled, scoping does not start with the measurement. It starts by naming them, and that is the work ProcessControlExpert carries.
Working references: ICH Q8 to Q14, with ICH Q2(R2) for the validation of analytical procedures and ICH Q14 for their development. EU GMP and Annex 11 for computerised systems.
Where to start
If you already know which variable you want to follow, the pages by variable say which technologies answer it and on what conditions. If the question is still open, that is what the scoping and diagnosis phase is for.
Either way, the trade-off turns early on one question worth asking at the start. Do you need an absolute value, and therefore a calibrated model to maintain, or a stabilised state? The calibration burden you take on decides the cost of ownership for the whole life of the installation.
Funding the project: we are CIR and CII approved
An in-line measurement project is research work before it is an equipment purchase. The Structure and Prove phases produce new knowledge about your process — which is exactly what the accreditation covers.
PLM (PAT-INDUSTRY) is accredited for the Crédit d’Impôt Recherche (CIR) and the Crédit d’Impôt Innovation (CII) — the French research and innovation tax credits — by the Ministry of Higher Education, Research and Space for the years 2026 to 2028 (SIREN 983 663 634; art. 244 quater B, French tax code). On that basis, the expenditure corresponding to the R&D work (CIR) or to the design of prototypes and pilot installations for new products (CII) that we carry out for our clients may be included by them in the base of their tax credit, subject to the work actually qualifying and to the conditions set by law.
In practice
Our invoices identify R&D (CIR) and innovation (CII) work as distinct items, so the amounts can be evidenced without reconstruction. The accreditation decisions and a description of the work are sent on request.
What the accreditation does not do
It attests our capacity to carry out this work. It does not settle whether a given project qualifies, and on its own it is not enough to justify that to the tax authorities: only a formal tax ruling (rescrit fiscal, art. L 80 B 3° of the French tax procedure code) gives a binding position. Each client remains responsible for its own filing.
This section is informative and does not constitute tax advice.
Tell us which step you would like to shorten. We will tell you whether a measurement answers it, and which phase to start with.
Forty-five minutes is enough to place your need among the four phases, to know what scoping would look at first, and what you would prepare.