Contract manufacturing (CDMO)
Turning line qualification runs into a permanent process reference
A contract manufacturer qualifying new production lines wants the qualification runs to keep working after the report: as a reference that watches every campaign that follows.
The challenge
Qualification runs produce the cleanest, most densely documented process data a facility ever generates. Typically it goes into the qualification report and stops there, so routine campaigns start with no quantitative reference for what a good run looks like.
A contract manufacturer also runs many client processes with few repeats each, so operators cannot build intuition for a campaign they run twice a year. A deviating campaign is identified from release results, after the batch is finished: a delayed delivery for the client and an investigation that reconstructs events from trends read by eye.
The client
A European contract development and manufacturing organisation running microbial and mammalian processes, qualifying new production lines across two sites with largely identical equipment.
Our approach
We take the qualification and engineering runs of one microbial line and build a golden-batch model from them. During later campaigns the twin:
- compares the live run against the qualified reference
- names the variable responsible for a deviation
- scores each finished batch against the reference, giving the investigation a starting point
Estimated effects
54% vs 12%
deviation score of a bad batch versus a good one
Published golden-batch benchmark on 100 simulated fermentation batches, with pH and dissolved oxygen as the main contributors. Simulated data, not a plant result.
0
extra runs to build the reference
The model is built from qualification and engineering runs the plant performs anyway.
1 model, 2 sites
one reference per line, portable between sites
Applies where the equipment is the same at both sites; to be confirmed in Phase 0.
Estimates to be checked in Phase 0 on the manufacturer’s historical and qualification data, not results.
How the PoC runs
-
01
Phase 0 · 4 weeks
Data assessment, and the choice of line and scope.
-
02
Data buffer · 2–4 weeks
Cleaning, digitising and time-aligning the run data.
-
03
Phase 1 · 8–12 weeks
The golden-batch model, tested on held-out runs.
-
04
Next step
Extending the reference to the next line, the next client process and the second site.
What we asked for
-
Line description
A P&ID or drawing of one fermentation train, with measurement points and ranges.
-
Run trends
Temperature, pH, dissolved oxygen, agitation, feed and off-gas where recorded.
-
Run outcomes
Biomass, titre or yield per run, plus in-process samples if taken.
-
Run history
Run number, start and end, strain, media lot and any recorded deviations.
What you get
-
Time to react
Inside the run instead of after release.
-
Faster investigations
The deviating variable is named, not hunted.
-
A client-facing argument
Quantitative evidence of consistency, campaign after campaign.
-
Reusable
The same method applies to the next line and the next client process.
Data can come in any format — CSV exports, spreadsheets, even scanned batch records — and the work runs on a copy of historical data, outside GxP and without touching production. No new sensors, IT integration, remote access or client-confidential material are needed to start.
Have a similar process?
Tell us about it. We’ll tell you where a model pays off first and what data you’d need.
Book a technical call