A request for proposal for an inline process analyzer is one of the few procurement documents where the specification sheet hides more than it reveals. Two probes can quote identical wavelength range and identical signal-to-noise figures, yet differ by a factor of three in calibration maintenance cost across a five-year horizon. The gap rarely appears in the quote. It appears in the validation package, the firmware roadmap, and the small print on calibration transfer.
This guide sets out the practical evaluation axes a PAT or validation engineer can apply when comparing vendor responses for NIR, Raman, FTIR, UV-Vis, or in-line FBRM systems. It assumes the analytical method has been provisionally chosen and the question is now which instrument and which supplier to award.
The framing throughout is regulatory and lifecycle. A probe that meets the specification on day one but cannot survive a calibration transfer from the development unit to the production skid is not fit for purpose, regardless of headline sensitivity.
Verify the specification against a defined acceptance test
Vendor datasheets quote photometric noise, wavelength accuracy, and stray light under laboratory conditions that rarely match the process environment. The first task of RFP evaluation is to translate datasheet figures into a measurable acceptance test the vendor must sign against.
A credible RFP specifies the reference materials, the sample presentation geometry, and the integration time at which each figure of merit must be demonstrated. For NIR, this typically means a NIST-traceable wavelength standard and a defined diffuse reflectance reference. For Raman, a polystyrene or acetaminophen reference and a fixed laser power. For FTIR, a polystyrene film and a defined apodisation.
Without these conditions written into the RFP, the vendor is free to quote best-case figures from a benchtop unit. The proposal evaluation should award no credit for specification numbers that lack a stated test method.
Demand a noise budget, not a single number
Photometric noise at one integration time is uninformative. Ask for noise as a function of integration time across the realistic operating range, and ask separately for short-term (within-shift) and long-term (week-over-week) drift figures. A probe with low short-term noise and high long-term drift will fail a continuous manufacturing application even if its headline number looks competitive.
Calibration transfer is the hidden cost driver
ASTM E1655, the standard practice for infrared multivariate quantitative analysis, sets out the methodology for developing and validating chemometric calibrations. Section 18 of the standard addresses calibration transfer between instruments. A proposal that does not reference E1655 or an equivalent internal procedure is a proposal that has not thought about the problem.
The RFP should ask the vendor three specific questions. First, what is the documented procedure for transferring a calibration from a development unit to a production unit of the same model. Second, what is the documented procedure for transferring a calibration after a major component replacement, such as a detector or a light source. Third, what reference standards and what minimum number of transfer samples are required in each case.
A vendor that answers “the calibration will transfer automatically” is making a claim that ASTM E1655 does not support. A credible answer describes a slope-bias correction, a piecewise direct standardisation, or a generalised least squares approach, with a defined number of standardisation samples.
Ask for evidence from the installed base
The proposal should include at least one anonymised example of a calibration transfer the vendor has executed on a comparable application, with the residual prediction error before and after transfer. If no such example can be supplied, the vendor is asking the buyer to fund the first transfer attempt.
Validation evidence and ICH Q14 alignment
ICH Q14, on analytical procedure development, formalises the lifecycle approach to analytical methods. It expects the analytical target profile, the method performance characteristics, and the control strategy to be defined and documented. An inline analyzer that will support a regulated process must come with vendor documentation that supports each of these elements.
The RFP should require the vendor to supply, at minimum, a design qualification document, an installation qualification template, an operational qualification protocol with defined acceptance criteria, and a performance qualification rationale that ties instrument performance to the analytical target profile. Generic IQ/OQ templates that do not reference the specific analyzer model are a warning sign.
For pharmaceutical applications, USP general chapter 1856 on near-infrared spectroscopy and the FDA’s 2004 PAT guidance set the regulatory baseline. The vendor proposal should map its qualification package to these references explicitly.
Scope the factory and site acceptance tests in writing
The factory acceptance test (FAT) and site acceptance test (SAT) are where ambiguous specifications become expensive change orders. The RFP must define what is tested at the factory, what is tested on site, and what constitutes a pass.
A defensible FAT covers the photometric specifications under the conditions written into the RFP, the communication interfaces against a simulated control system, and the alarm and diagnostic functions. A defensible SAT covers installation against the as-built P&ID, the communication interfaces against the actual control system, and a process-relevant signal check using a representative material.
Vendors that propose to combine FAT and SAT into a single on-site test are shifting risk to the buyer. Each failure mode discovered on site is a failure mode that costs production time to resolve.
Integration constraints decide the long-term tenability
The communication interface is rarely the most interesting part of an analyzer proposal. It is frequently the part that determines whether the system will still be supportable in seven years.
OPC UA has emerged as the preferred interface for new analyzer integrations in regulated industries, because it carries structured data, time stamps, and quality codes natively. Profibus DP and Profibus PA remain common in installed bases. The 4-20 mA loop, while limited to a single scalar value, remains the most robust option for safety-critical interlocks. Ethernet/IP is prevalent in discrete and hybrid plants.
The RFP should specify the required protocol, the required data model, and the required cybersecurity posture. A vendor that quotes only a proprietary protocol with a gateway is quoting a future obsolescence problem.
Maintenance burden over a five-year horizon
Procurement decisions are often made on capital cost. Operating cost over five years frequently exceeds capital cost by a factor of two or more, once consumables, calibration maintenance, and unplanned downtime are included.
The RFP should ask the vendor for a stated mean time between failures for the light source, the detector, and the probe optics, with the assumptions made about duty cycle and process environment. It should ask for the recommended preventive maintenance schedule and the cost of a typical service visit. It should ask for the firmware support commitment, expressed as a number of years of security and bug-fix updates from the date of delivery.
A proposal without these figures is not a proposal that can be evaluated on lifecycle cost.
What a credible quote should and should not say
A credible quote names the model, the firmware version at delivery, the included accessories itemised by part number, the qualification documentation included in the price, the FAT and SAT scope, the warranty terms, and the support response time. It states the calibration transfer methodology and references the relevant standards.
A credible quote does not promise performance figures that exceed the datasheet. It does not bundle qualification documentation as an optional extra without naming a price. It does not refer to “industry-standard” interfaces without naming the protocol and revision.
The proposal that survives this filter is rarely the cheapest. It is the one whose total cost is knowable in advance.