00Data Matters · life sciences manufacturing

Why this industry runs on paper longer than any other.

Not because the people are behind. Because every time the industry gained momentum, the regulator tightened how data may be collected, shared and proven, and rightly so. The result is a manufacturing sector where the record is the product.

THE PAGE THEY ASK FORONE BATCH RECORD200+ PAGES
A batch record past two hundred pages, and the one page an inspector will ask for. Finding it is the job. Proving nothing else in the stack contradicts it is the three weeks.Schematic, synthetic. One batch record seen edge on, and the page an inspector asks for.

01The journey

From islands of data to a single source of truth.

Getting to your data should not be this difficult. Every plant we walk into runs on islands of data, and finding the truth takes weeks. Audits stake everything on that truth. There is a way out, and it has four steps.

  1. 01 ⁄ 04 · The real worldYour data lives on islands. A single process generates a tremendous amount of data in every format, from every vendor, and each system keeps it locked in its own controller, database and reporting tool. Local RDBs, flat files, Excel exports, a line historian, hand-written logs, paper reports. Islands everywhere, connected by nothing.
  2. 02 ⁄ 04 · First moveCentralise the truth. One plant-wide historian becomes the single source of truth: every line’s critical process data, collected automatically, landing in one place with its integrity intact. It is the first thing we stand up, and everything else builds on it.
  3. 03 ⁄ 04 · Then structureOne line, one database. Line databases migrate off the islands into centralised SQL: one database per line, never mixed. Queries cannot cross-contaminate, and maintenance on one line never touches another validated system. “Databases are cheap.” Jeff Weiss, Sr. Data Architect.
  4. 04 ⁄ 04 · The payoffNavigable by everyone. Publish it all into a Unified Namespace: one structured, real-time layer where every system, and every person, subscribes to contextualised data in their own language. /Enterprise/SiteA/TabletLine1/Coater/DrumSpeed. Operators see equipment status, quality sees batch data, engineers see process trends, management sees production KPIs. Data, democratised.

02What a plant actually generates

Six kinds of data, six systems, one batch.

01Process & equipmentHistorian time-series from every instrument on the line, at sub-minute resolution.211.188
02Batch & recipeMaster and executed records, phases, setpoints, operator actions.211.186
03LaboratoryIn-process and release testing, stability, certificates of analysis.211.194
04Quality & regulatoryDeviations and CAPA, change control, audit trails, IQ/OQ/PQ validation records.211.192
05Facility & assetBMS room conditions, environmental monitoring, maintenance, calibration.211.46
06MaterialComponent lot genealogy from incoming raw material through to finished product.211.188(b)(3)

Each system is correct. None of them was built to answer a question that spans the others, which is every question a batch investigation asks.

03What’s at stake

Islands don’t just slow you down. They fail audits.

In life sciences, data is only legally valid if it meets ALCOA+, and scattered, un-indexed, hand-copied records cannot. Integrity is what turns thousands of entries into evidence you can stake an FDA audit on. Even a modest database holds thousands of critical electronic records, and more arrive every day, far past what any human reviewer can keep honest. Nine tests, each enforced, each proven.

A

Attributable

Who did it, and when.

HOW User ID on every entry, e-signatures, audit trails of operator actions.

L

Legible

Readable and permanent.

HOW Clean electronic records and clearly labelled charts, for the record’s life.

C

Contemporaneous

Recorded as it happens.

HOW Automated capture from equipment, with no retrospective entry.

O

Original

The first capture, preserved.

HOW Raw PLC and instrument data retained with full lineage.

A

Accurate

Correct and error-free.

HOW Calibrated sources, verified calculations, no unauthorised change.

+

Complete

All data present.

HOW Every record is checked for what is missing, not a sample of them.

+

Consistent

Chronological order held.

HOW Sequenced, time-true records across every source.

+

Enduring

Securely stored for its life.

HOW Hardened, redundant retention with store-and-forward.

+

Available

Accessible for review.

HOW On-demand across the plant throughout the retention period.

Human error at risk: Attributable · Accurate · Complete. Automated capture and validation remove manual entry and its mistakes.211.68
Transfer errors at risk: Legible · Contemporaneous. Reporting happens in place, so records never get copied between systems and lost or corrupted along the way.11.10(b)
Bugs and malware at risk: Enduring · Available. Restricted access, error detection and full logging of every change.11.10(d)
Compromised hardware at risk: Original · Consistent. Redundancy, store-and-forward and health alerts keep collection unbroken.211.68(b)

04Audit trails

Two audit trails, and the review reads both.

Each source keeps its own audit trail: the historian, the MES, the LIMS. encompass360 reads them in place and shows them on the same axis as the trend, so a change to a value is seen beside the value it changed. That is where the FDA puts the work: reviewers “should review the audit trails that capture changes to data associated with the record as they review the rest of the record.”

encompass360 keeps a second, its own: an audit trail throughout the review layer, what was done, by whom, and when. Below is the scope the live demo implements and the one we would take into a validation conversation. It is a design, stated so a CSV lead can argue with it early rather than discover it late. The report the regulator asks for by name sits in report family 10: printouts “indicating if any of the data has been changed since the original entry” (EU GMP Annex 11, §8.2).

ObjectWhat is recordedWhy it is in scope
Discrepancy stateEvery transition, previous state and new state, on the discrepancy key.The disposition of an exception is the reviewer’s act. It is the entry an inspector looks for first.
NoteThe text, the object it is attached to, and whether it is on a discrepancy, a window or the batch.MHRA asks the trail to hold the who, what, when and why. The why lives in the note.
InvestigationOpened, saved with its conclusions, and exported, each with the investigation identifier.§211.192 requires the investigation to be written and extended to other batches. This is that thread.
ExportEvery export, its kind, its scope, and for a signed export the digest. A record that left the system is a record that can be quoted back. The trail says which one left and what it contained.
Pin and true copyThe SHA-256 of the pinned view, and for a true copy the file count and the Merkle root.Annex 11 §8.2 asks whether data changed since original entry. A digest answers it arithmetically.
ViewWhich chart mode and which optional strips were on when a judgement was made.What a reviewer was looking at is part of what they saw. It is cheap to record and impossible to reconstruct later.

Every entry carries the user, an ISO 8601 timestamp, the action, the object and the detail, and the trail is exportable as a flat file with no tooling of ours. Out of scope, deliberately: pointer movement, scrolling, zoom, hover, and any read that changes nothing. A trail that records everything is a trail nobody reads, which is the failure mode EU GMP Annex 11 §9 is pointing at when it asks for audit trails to be reviewed. Source systems keep their own trails and encompass360 neither writes to them nor duplicates them.

Words we use, and why →

05Who takes you there

A partner who owns the outcome.

  1. We own the problemWe know your field and we study the process. We watch how the line really runs, then we stay on it until it is fixed.
  2. 30+ years on the floorDozens of automation and data-infrastructure projects delivered across life sciences and packaging manufacturing.
  3. Open architectureWe pick the strongest technology at every layer. Closed, single-vendor designs only narrow your options over time.
  4. Minimise complexityForcing a tool to do a job it was not built for means you have the wrong tool. The simplest system that does the work is the one that lasts.
  5. We build what does not existWhen off-the-shelf cannot meet the objective, we engineer the system that can, and we stand behind it.

We do not hand you the same off-the-shelf answer everyone else sells. We build the one that moves your numbers and takes cost out.

06Built to validate

Rigour that is fast to stand up.

A standardised, repeatable validation path means the compliance you need does not turn into a multi-year project. Standardised validation, off-the-shelf components, validate in weeks not years, data stays in place, CAPA-remediation ready.

Batch at a Glance, the golden master and the exception report are descriptive statistics: medians, percentiles and ranges over released batches. No model sits between the data and the reviewer, so the draft EU GMP Annex 22 on artificial intelligence does not reach them. Where encompass360 predicts or classifies, in its multivariate analytics, that is a model, and our position is that it must meet Annex 22 as drafted: static, deterministic, a logged confidence score on every outcome, “undecided” below threshold, test data under access control and audit trail.

The same architecture that makes your data provable also pulls recurring cost out of the operation: investigation labour, deviation exposure, reporting effort, downtime. On the reference line the client’s own figures are more than $3M saved per line per year, $1.5 to 2M less batch-deviation cost, full ROI inside three months, and 95% less time aggregating data. One site’s numbers, stated as such; the basis is available under NDA.

07The structural argument

A historian is excellent at numbers and poor at text.

A process historian stores numeric time-series superbly; that is what it was designed for. It handles text badly: audit trails, batch information, user records. Its data structure is long and narrow, one row per tag per change, and that shape is wrong for the records a reviewer needs to read.

A relational database is the exact inverse: strong on text and relationships, without the time-series optimisation a plant floor demands.

Neither is wrong. Neither is sufficient alone. The answer is to layer both, and that architectural decision, made deliberately years ago, is why encompass360 can hold the trend, the audit trail, the recipe and the genealogy on a single axis instead of asking a reviewer to join them by timestamp and hope.

08No more islands of

No more islands of data.

Access critical processes from anywhere, instantly. Bring the data problem that has been costing you; we will show you the path out, and what it has saved teams running lines like yours.