Manufacturing batch analytics that answers a recall or an outlier as a genealogy, in hours instead of months
Batch analysis that took three months now takes hours, with the query behind every answer.
See one case, screen by screen ↓Every batch question crosses four systems. The batch record lives in the MES, the release and in-process results in the LIMS, the lot links in ERP material movements, and the procedures in the document system — while supplier certificates of analysis arrive as PDFs. An MSAT scientist investigating a low-viability outlier joins a dataset of more than 300 columns by hand, walks the lot links one at a time and looks up the limits in the SOPs. A full outlier investigation took about three months by hand.
A supplier recall is harder, and it runs on a clock. The notice names one lot, and the procedure allows 24 hours for the product impact assessment. Someone has to work out which finished batches used it, where each dose shipped, which patients are already infused, which doses are still at the treatment centre — and whether any other lot in the store came from the same failed input. In autologous cell therapy every batch is one patient, so a late answer can mean a dose infused that should have been held.
Estimated split of the roughly 90 days a full outlier investigation took by hand on Line C2, against the 5.5-hour median shown on the solution's dashboard this quarter.
Five agents route, trace both ways, query and cite every batch question; the Qualified Person decides.
Monday, 6 October, Plant 3 — Leiden. Line C2 makes Corventa, an autologous CAR-T therapy. At 08:42 Northgate Bioreagents recalls a cryomedium lot that is in every frozen dose. Here is what happens next, screen by screen, in the working solution.
Lena's Trace screen opens on the line's live questions: 1,284 batches, 9,412 lots and 642 procedures behind a single question bar. This morning's outlier trace is on the graph, and beside it the supplier notice that has just arrived — NBR-FSN-2026-014, cryomedium lot CRYO-5512 recalled because retained bottles failed a container-closure check.
“Cryomedium lot CRYO-5512 recalled — retained bottles failed a container-closure check; sterility cannot be assured.”
Lena clicks “Trace this lot” and asks which finished batches used CRYO-5512 and where they went. In five steps and 2.5 seconds the genealogy specialist walks back to the three inputs on the supplier's certificate — DMSO, human albumin 5% and bottle lot NB-BT-0390 — and forward to 7 finished batches, CRV-26-0420 to CRV-26-0426, shipped to Germany (3), the Netherlands (2) and the United Kingdom (2).
“3 inputs ← CRYO-5512 → 7 finished batches → 3 markets”
Release results and shipments sit on every node. Five doses were infused between 4 and 19 September. Two are in storage at the treatment centre: CRV-26-0426 at St Aldwyn's Hospital, Manchester, infusion due 8 October, and CRV-26-0425 at Klinikum Rheinau, Cologne, due 14 October. All seven passed sterility (no growth) and endotoxin (0.12–0.31 EU/mL against a limit of 1.0). CRV-26-0422 carries a major deviation, DEV-2026-0168.
Looking for the same input elsewhere, the walk finds that bottle lot NB-BT-0390 also filled CRYO-5530 — received 22 September, 12 bottles in Leiden's store, none used. A where-used report on CRYO-5512 alone would have missed it.
Lot record: “Bottle lot NB-BT-0390 — the same as the recalled lot · 12 bottles in stock, none used.”
Every sentence of the answer carries a numbered source: the supplier notice, the material movements, the shipments and chain of custody, the release results and the certificate for CRYO-5530. The Query view shows the read-only walk itself — 31 links walked in 0.6 seconds across genealogy, shipments, lab and supplier certificates.
Links walked: 31 · read time: 0.6 s · marked Read-only.
SOP-QA-0112 §5.2 asks for a product impact assessment within 24 hours of a supplier recall. The drafter reads the trace and the 7 batch records and fills IA-2026-0031: exposure, product status, and four proposed actions — quarantine CRYO-5512 (6 bottles) and CRYO-5530 (12), ask both centres to hold the stored doses, tell the treating physicians of the 5 infused patients through medical affairs, and assess notification in each of the 3 markets. Proposed risk: low, because release sterility passed.
“Drafted by an agent from the cited records. Nothing is sent until you sign; the Qualified Person decides every action.”
Lena signs with the meaning “I confirm the facts in this impact assessment are correct”. IA-2026-0031 goes to Omar Haddad, QA · Qualified Person, who decides on the two doses at the centres and on notifying the authorities. On the graph both cryomedium lots now read “To quarantine” and the two stored doses “hold requested”. The 24 hours run to 08:42 on 7 October.
“IA-2026-0031 sent to QA. Signed by Lena Ortiz.”
Lena's other open question is a trend. Asked for viability at harvest over the last 20 batches, the batch data specialist finds the metric (lab test VIAB-HARV, 7-AAD flow) among 312 columns, runs a read-only query and draws the chart: an 87.6% average, with CRV-26-0431 at 68.9% — under the 70% release limit — and CRV-26-0418 at 72.4%, under the 75% action limit.
Outlier rule shown with the query: below mean − 3σ of the 32 batches before 21 July (83.4%).
On CRV-26-0431's genealogy, from apheresis to drug product, the specialist compares every raw material lot with the other outlier. One is shared: culture medium MED-24-0917 from Halvern, received at 318 mOsm/kg — the top of its 280–320 range, where the other lots ran 296–302. The same lot is in CRV-26-0436, now in expansion at 81.2% on day 6, with its next feed at 14:00.
“Same supplier, same medium. Lot MED-24-0917 came in at 318 mOsm/kg osmolality — the top of its 280–320 range.”
Below 75% at harvest is an action-limit excursion. From effective versions only, the procedure specialist cites SOP-MFG-0214 §8.3: tell MSAT and QA within 4 hours, open a deviation the same working day, keep two 1 mL samples at ≤ −150 °C. SOP-MFG-0211 §6.4 asks for a raw material lot review before the next feed. The drafter prepares DEV-2026-0187 with a proposed class of major; Dana Okafor decides the class.
The v8.0 draft of SOP-MFG-0214 is in review — the specialist does not cite it.
The same recall and the same outlier, seen by the five people on Line C2 who carry them — what the work looked like, and what it looks like now.
Five agents route, trace both ways, query and cite every batch question; the Qualified Person decides.
Reads each question from MSAT, QA or manufacturing and routes it to the right specialist — batch data, genealogy or procedures — asking one clarifying question when the batch or time range is unclear, then composes the answer with its evidence.
Traces a lot backward to its inputs and forward to finished batches and markets in one walk, across material movements, shipments and supplier certificates, and finds lots shared between batches.
Turns a plain question into one read-only query over 312 batch columns, using the approved metric definitions, flags outliers against the control limits and picks the chart.
Answers process questions from SOPs, master batch records and test methods, quoting limits exactly as written with a section and page for every statement.
Drafts a deviation under SOP-QA-0102 or a supplier-recall impact assessment under SOP-QA-0112 from the genealogy trace and the cited records, with the classification and actions marked as proposals.
Decides holds, quarantine and notifications. The agents propose; a named person decides.
Ask about a lot, a batch, a result or a procedure. The right specialist answers with the chart, the query or the page.
Which finished batches used cryomedium lot CRYO-5512?
Seven: CRV-26-0420 to CRV-26-0426, shipped to Germany, the Netherlands and the United Kingdom. Five have been infused; two doses are in storage at the centre. The trace opens in both directions, with the supplier inputs.
Show viability at harvest for the last 20 batches and highlight outliers
Viability at harvest averaged 87.6% over the last 20 batches (21 July – 2 October). Two fall below the lower control limit of 83.4%: CRV-26-0431 at 68.9%, under the 70% release limit, and CRV-26-0418 at 72.4%, under the 75% action limit. The other 18 sit between 86.2% and 92.6%.
What does the procedure say when harvest viability is below 75%?
It is an action-limit excursion. SOP-MFG-0214 asks you to tell MSAT and QA within 4 hours of the result, open a deviation the same working day and keep two 1 mL post-harvest samples at ≤ −150 °C. Below 70% the batch cannot go to release without a QA disposition decision.
Which batches used vector lot LVV-2238?
Thirteen batches, CRV-26-0412 to CRV-26-0424. Vector copy number stayed between 1.8 and 2.9 copies per cell, inside the 5.0 limit on every batch. The lot is fully used; the next lot, LVV-2241, started with CRV-26-0425.
13 screens from the working solution, on its sample data. Pick one to see it large.
What each specialist may read, what is watched and who decides are settings, set by the head of MSAT.
| Setting | Default | Choose from |
|---|---|---|
| Batch data may read | Line C2 only | Line C2 only · All Corventa lines |
| Lot genealogy may walk | Plant 3 — Leiden | Plant 3 — Leiden · All plants |
| Cite effective versions only | On | On or off · superseded versions on request |
| Show the query with every answer | On | On or off |
| Watch rules | Harvest viability below 80% · in-process viability below 82% from day 6 · vector copy number above 4.0 | Add a rule · switch each on or off |
| Raw material lot hold — e-signature by | Omar Haddad | Omar Haddad · Dana Okafor |
| Deviation classification decided by | Dana Okafor | Dana Okafor · Omar Haddad |
| Answers saved to dashboards confirmed by | Sam Patel | Sam Patel · Lena Ortiz · Dr. Maya Chen |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample plant data · “proven” = in user acceptance testing on a commercial cell-therapy line at a global medtech/pharma manufacturer (80%+ accuracy) · “target” = the design goal, measured in the live solution · “estimated” = our estimate · “typical” = published figures (The Medicine Maker, “The quest for the golden batch”, and Frontiers in Manufacturing Technology, 2024 — $20,000 to $1 million per batch deviation, 2–8 weeks of root cause analysis for an out-of-spec or out-of-trend anomaly; Pharmaceutical Technology and L.E.K. Consulting — over $500,000 to produce a cell or gene therapy batch). People, companies, products, lots, sites and hospitals named on this page are fictional — characters and sample data in the working solution.
Analysis of batch records, lab results and lot links to explain yield, viability and deviation questions and to scope recalls. Manufacturing Batch Analytics lets plant teams ask in plain words and get the answer as a batch genealogy, a chart or a cited procedure, with the query or page behind it.
The genealogy specialist walks the recalled lot backward to the inputs on the supplier's certificate and forward through material movements to finished batches, shipments and markets in one walk. In the working solution, CRYO-5512 traced to 7 finished batches in 3 markets in 2.5 seconds, showing which doses were infused and which were still at the treatment centre.
Yes. The walk looks for the same input elsewhere. In the sample recall, the failed bottle lot NB-BT-0390 had also filled a second cryomedium lot, CRYO-5530, with 12 unused bottles in the store — a lot the supplier notice did not name.
No. Every connection is read-only and data answers come from SELECT-only queries with a 10,000-row limit. Deviations and impact assessments go to the quality system as drafts, held until a named person e-signs.
Every answer shows its query or its page: the SQL and tables behind a chart, the read-only walk behind a trace, the section and page behind a procedure answer. A person confirms an answer before it is saved or shared, and corrections go back to the specialist as examples.
Yes. The agents trace, query, cite and draft. The author signs the facts; QA decides the deviation class; the Qualified Person decides holds, quarantine, disposition and notification of the authorities, and lot quarantine needs that person's e-signature.
No. Only chain-of-identity codes are shown — never names or dates of birth — and each specialist reads only the data, line and plant it is allowed to.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your SOPs, master batch records and a sample of batch data and lot links — and it goes live once every quality gate has passed. We will walk you through it on your own batches first.
Any batch-record question answered in seconds, cited to the page and step.
1-min demoValidation Document AutomationProtocols drafted in minutes, every value traced to its source.
1-min demoDeviation & CAPA ManagementEvery deviation closed inside the 30-day target, with a CAPA aimed at the cause.
1-min demoWe'll run Manufacturing Batch Analytics on a sample of your own batch records, lot links and procedures.