Building a CGT Manufacturing Stack That Scales: How Phase 1 Architecture Decisions Impacts Phase 3 Outcomes

The first two posts in this blog series made a case that most of the industry hasn't fully reckoned with. In the first part, we established that roughly 21% of cell and gene therapy (CGT) clinical holds are triggered by Chemistry, Manufacturing, and Controls (CMC) deficiencies — the category that takes the longest to resolve and does the most lasting damage to programs, patients, and capital. Part 2 traced those deficiencies back to their origins on the manufacturing floor: batch records that discover problems too late, process monitoring that can't distinguish biology from drift, change control that can't keep pace with development speed, supplier management that mistakes familiarity for qualification.

The pattern is clear: the programs that get held aren't usually running bad science, they're running inadequate infrastructure. And the infrastructure decisions that determine whether a program can withstand regulatory scrutiny at Phase 3 are almost always made—or not made—in Phase 1.

The Phase 1 Trap

Most CGT sponsors enter Phase 1 in resource-constrained conditions with small teams, limited budgets, academic or hospital-based facilities, and manual processes. The priority is getting into the clinic — generating the early safety and efficacy signal that justifies the next round of funding and the move to Phase 2. In that environment, documentation infrastructure feels like overhead.

The problem doesn't surface until Phase 2 or Phase 3, when the U.S. Food and Drug Administration (FDA) asks a question that the decisions made in Phase 1 either enable you to answer or they don’t: is the product you're manufacturing now the same product that showed clinical benefit then?

That question requires Phase 1 data. Detailed process parameter records, critical quality attribute testing across batches, and in-process data showing how manufacturing behaved across operators and shifts. If those records were captured in real time, consistently, with enough granularity to anchor a comparability argument, you have a Phase 1 baseline that supports every manufacturing change you make on the way to commercial scale.

But if Phase 1 ran on paper, with back-filled records and limited in-process monitoring you don't have a baseline — you have a poorly characterized starting point and a well-characterized current process, with no reliable bridge between them. That's the gap that becomes a CMC hold, and it's a gap that takes more than eight months on average to close because you're rebuilding evidence that should have existed from the beginning.

Put plainly: The cost of building the right infrastructure in Phase 1 is significantly lower than the cost of reconstructing it under a clinical hold in Phase 3. There’s your business case.

Building the Right Infrastructure

The industry default is to cobble together systems incrementally — a lab notebook here, a document management tool there, an electronic batch record (EBR) platform once the process gets serious enough to justify it. The result is a disconnected stack where manufacturing execution lives in one place, quality events live in another, documents and training records live somewhere else, and the data that FDA needs to evaluate your process has to be manually assembled from all of them when a submission is due.

That architecture fails CGT for the same reason paper batch records fail it: it discovers problems after the fact and assembles evidence retroactively. Neither works when your product expires in 24 hours and your comparability story depends on data you generated two years ago.

The infrastructure that works in CGT does something different. It embeds quality controls into manufacturing execution so that documentation happens simultaneously with the work, deviations are caught while there's still time to act, and the evidence base the FDA needs is built continuously as a byproduct of production — not assembled under pressure before a submission deadline.

The Manufacturing Execution Layer: Where Control Lives

Every failure mode identified in the second part of our blog series has a root cause in manufacturing execution. And every one of them is addressable at the execution layer if the system is built correctly.

Electronic batch records are the starting point. When critical process parameters feed directly from equipment into the batch record—temperature probes, timers, flow cytometers, balances—transcription disappears as a failure point. When the system enforces process sequence and won't allow an operator to proceed to the next step until the current one is complete and verified, sequence errors don't propagate through a 14-day manufacturing run. When deviations are flagged in real time, the team can intervene while the batch is still running rather than conducting a post-mortem after it's gone.

For a CAR-T program running 20 to 30 discrete steps across two weeks, with multiple operators across multiple shifts, the difference between paper-based execution and electronic execution isn't a documentation preference — it's the difference between a process you can demonstrate is controlled and one you can only hope was.

Electronic logbooks extend that control beyond the batch record itself: equipment use logs, environmental monitoring records, and cleaning and maintenance histories. In paper form, these are the documentation gaps that surface during FDA inspection or investigational new drug (IND) review and are almost impossible to reconstruct retroactively; when they’re electronic, they're controlled, time-stamped, linked to the relevant batch, and immediately available when a reviewer asks for them.

In-process monitoring connected to execution is what transforms process data from a record of what happened into a tool for controlling what's happening. When critical process parameters—like temperature, pH, or cell density—are captured in real time and continuously compared against defined ranges rather than after the batch is complete, operators can act while the batch is still running.

The cumulative effect of these capabilities is a manufacturing process that generates its own compliance evidence. Every batch executed through an EBR produces a complete, traceable, audit-ready record as a natural output of the work itself — not as a separate documentation exercise that happens afterward.

The Quality System Layer: Where Control Is Sustained

Manufacturing execution controls what happens in a single batch. The quality system controls what happens across the entire program.

The failure modes that accumulate at the program level—undocumented process changes, supplier qualification gaps, training inconsistencies across operators and sites, deviations that recur because corrective action/preventive action (CAPA) was never completed—are quality system failures. And they become CMC deficiencies for the same reason batch record failures do: the evidence that the process is under control doesn't exist because the system that would have generated it wasn't in place.

Document management in CGT is the mechanism that ensures every operator is working from the current approved version of every procedure, that changes to manufacturing processes are reflected in updated standard operating procedures (SOPs) before the next batch runs, and that the document history required for an IND amendment or FDA inspection is immediately accessible rather than assembled from email threads and shared drives. Version control failures are a direct path to the kind of process inconsistency that triggers comparability questions.

Training management connected to manufacturing execution takes document control one step further: it ensures that the operator running a batch today has been trained on the current version of the relevant procedures, and that the training record is automatically linked to the batch record. In multi-operator, multi-shift CGT environments training consistency is a process control, not an HR function. And unexplained, non-biological batch-to-batch variability is often traceable to operator-to-operator execution differences that a connected training system prevents.

Quality event management—deviations, CAPAs, and change control—is where the quality system either closes the loop on manufacturing failures or allows them to recur. The change control failure pattern identified in the second part of our blog series is almost always a workflow problem: change control exists as a separate quality function that manufacturing has to stop and navigate around, so teams find ways to keep moving without it. Change control lives in the quality system. But when that system is connected to manufacturing execution, every approved change is reflected immediately where the work actually happens. The manufacturing record reflects the process as approved, not as it was two iterations ago.

The same integration principle applies to deviations. A deviation captured in real time during batch execution should automatically initiate a quality event record, route for investigation, and link to the relevant batch data, without requiring a separate entry into a standalone quality system. When those steps are disconnected, time between occurrence and investigation grows, data gets reconstructed from memory rather than captured in the moment, and the investigation that results is built on a weaker evidentiary foundation.

The Right Infrastructure in Practice

The clearest way to see whether your infrastructure is working is to imagine an FDA information request arriving during the 30-day IND review window. "Provide all batch records for your Phase 1 manufacturing runs along with associated deviation reports, change control records, and raw material Certificates of Analysis (CoAs)." In a connected system, that's a report generated in hours — batch execution data linked automatically to the deviation records it triggered, change control approvals tied to the document versions operators were trained on, raw material CoAs traceable through to the finished product batches that used them. The manufacturing record, the quality record, and the supplier record exist as a single integrated dataset rather than three separate collections that have to be manually reconciled under time pressure.

The same integration that answers an FDA request in hours is what prevented the deviations from becoming holds in the first place. Deviations flagged during execution rather than discovered at batch record review. Process changes documented before implementation, with comparability data generated at the time of the change. Supplier qualifications enforced at the point of material use, not audited retrospectively when something goes wrong. Training verified against the current approved procedure before the operator runs the batch, not confirmed after the fact when a reviewer asks who was qualified to do what.

That's what connected infrastructure looks like — not a system that helps you respond to FDA, but one where the response is a byproduct of how the system runs.

Building for the Program You're Going to Have, Not the One You Have Now

The argument for building this infrastructure in Phase 1 isn't that Phase 1 requires it — FDA's flexible CMC framework explicitly allows for less complete manufacturing characterization in early-stage programs. The argument is that Phase 3 requires it, and Phase 3 comparability depends on Phase 1 data.

Every architecture decision made in Phase 1 has a Phase 3 consequence:

• Paper batch records in Phase 1 mean a poorly characterized baseline for Phase 3 comparability.
• Disconnected quality systems in Phase 1 mean manually reconstructed evidence packages for Phase 3 submissions.
• Informal change control in Phase 1 means undocumented process evolution that the FDA sees as a moving target when the Biologics License Application (BLA) arrives.

The sponsors who scale successfully aren't simply doing more documentation. They're running manufacturing operations where the documentation is a byproduct of execution, the quality system is connected to the manufacturing workflow, and the evidence base FDA needs is built continuously rather than assembled at the last moment.

That's not a compliance posture. That's a manufacturing strategy.

What You Should Do Next

• Subscribe to the GxP Lifeline newsletter to access other informative articles like this.
• Contact us to get answers for your software, quality, and compliance questions.