PAT is not a family of instruments. It is a way of running manufacture: measuring during, instead of testing after.
Process Analytical Technology. The acronym comes from pharma; the practice did not stay there. The definition holds in one sentence, and it is the FDA’s, September 2004: a system for designing, analysing and controlling manufacturing through measurements made during the process, on the critical quality and performance attributes of incoming materials, in-process materials and the processes themselves, with the goal of ensuring final product quality.
Everything is in “during”. A release test says whether the batch conforms. A process measurement says what is happening while there is still something to be done about it.
Three received ideas, and what the text actually says
“Analytical” does not mean “spectroscopy”
The founding text states that the word covers chemical, physical, microbiological, mathematical and risk analysis. A tomography, an imaging system or a soft sensor are PAT on the same footing as a near infrared probe.
A relocated instrument is not PAT
The FDA writes it: transferring a laboratory method in line is not automatically PAT. What makes it PAT is that the measurement serves a decision about running the process. A probe producing a curve nobody reads stays a probe.
The goal is not to measure more
It is to understand the process. The same text defines a well understood process in three conditions: critical sources of variability identified and explained, variability managed by the process, and quality attributes reliably predicted over the design space.
Two sentences we would rather quote ourselves
They come from a technical guide published by USP in 2023 on control strategy in continuous manufacturing. They are uncomfortable for anyone selling instrumentation. They are also right, and better heard from us than from a competitor.
The use of PAT decreases as process understanding increases
The text puts it this way: as process understanding grows, the use of PAT tools decreases, allowing a robust control strategy that is not overburdensome for data collection and reporting. And a risk assessment is what decides whether PAT is needed to control a given unit operation.
The same guide goes further: in some processes, especially semi-continuous ones, it may be possible to eliminate the spectroscopic measurement in production runs after using it extensively to build the models — soft sensors and gravimetric control taking over.
What we make of it. You do not replace a model you have not built: the probe is what establishes it, and it has to come back the day the process changes. That is an argument about sequence, not a retreat — and it says in which order to spend.
Closed-loop control is not the default reflex
Same text: although advanced control approaches — feedback, feed forward — are recommended by regulatory agencies, their implementation can be complex and should not be the default approach to a control strategy.
The reason is mechanical, and written elsewhere in the same guide: a loop rests on the response time of the instrument and on knowing the residence time distribution. Without both, the correction arrives too late to be of use. To which add the reliability of the instrument — build-up and calibration drift, which the text asks explicitly to be considered.
USP Technical Guide, Control Strategies for Continuous Manufacturing of Solid Oral Dose Drug Products, 2023, § 2.2.2, § 3.4 and § 4.2.1. That document is published for information: it describes a practice, it does not create a requirement.
Where the measurement sits, and why it changes everything
Four positions, and each name states what becomes of the sample. In-line: the sensor measures in the stream, nothing is removed — in contact, behind a window, or with no contact above the product. On-line: a fraction is diverted to a cell, then most often returned. At-line: a sample removed from the process and measured beside the line. Off-line: in the laboratory, with the delay that goes with it.
This is not a scale of quality, it is a scale of delay. The position follows the rate at which the decision has to be taken: stopping a drying calls for an in-line measurement, checking a material on receipt lives happily at-line. The four positions, defined.
What it changes, in practice
- An operation stops on a state rather than on a duration. The safety margin paid on every batch goes away on the batches that never needed it. Detecting an endpoint.
- A drift shows while it is happening, instead of appearing on a reconciliation afterwards.
- The check moves to where the material arrives, instead of holding a quarantine. Checking identity at goods-in.
- Part of the laboratory work becomes redundant, and the calculation is done on your batches. What a control costs.
None of these gains is automatic. They depend on the matrix, the measurement position and the number of batches a year. That is what scoping establishes on your own data, before anything is bought.
PAT and Quality by Design: two halves of the same job
Quality by Design asks which attributes carry quality and which parameters move them. A critical quality attribute you cannot measure stays an intention. PAT is what makes those attributes observable during manufacture.
The converse holds too: a measurement tied to no attribute stays a sensor. The two are built together, which is why we carry them together. How the two connect.
The acronym is pharmaceutical, the practice is not
Chemicals speak of in-line measurement, food of continuous checking, process industries of analytical instrumentation. They are the same gestures and the same questions: which quantity, at what rate, in what place, and with which model to maintain.
What changes from one industry to the next is not the physics, it is the frame: a regulatory dossier on one side, a margin per tonne on the other. What carries across from one industry to another.
Describe one operation. We will tell you whether a process measurement belongs there, and which one.
Forty-five minutes is enough: describe the operation, say at what rate the decision is taken, and find out whether a process measurement answers it — or whether the subject lies elsewhere.