Ask a plant what happens when a deviation is opened, and you often hear about three places. The batch record is in one system. The equipment and area logbooks are in another. The deviation itself is somewhere else. Someone has to reconcile them before anyone can say what happened.
Now add AI. If an assistant has to read from three systems, it inherits the reconciliation problem, and it may not be clear which version of an event it is describing.
A recent CIO.inc interview, Pharma's AI ambitions run into disconnected systems, points at this gap. This article gives our view as the makers of Atto Pharma eBMR and Atto eLogbooks. It rests on how our own software is documented, and we have kept to that.
The problem with three systems
A deviation question is rarely about one record. "Which readings were out of limit before this step?" "Who corrected this value, and why?" "What else ran in that area that week?" The answers sit across the batch, the logbooks and the quality system.
When those are separate systems, three things tend to happen:
- Each has its own timestamps, user names and edit history, so they have to be matched by hand.
- A correction in one place may not show in another.
- Any AI layered on top produces an answer that is only as consistent as the weakest link underneath.
The AI is not the main risk here. The reconciliation is. You can't investigate quickly, or let a tool help you, if the facts first need assembling.
One database, one audit trail
Atto Pharma eBMR and Atto eLogbooks are delivered as one hosted web application with one database. The batch record and the logbooks sit together. The documented behavior that matters most for deviations is this:
- Results are entered once, at the step, against limits held in the record, and attributed to a person and time.
- A result outside the internal limit needs a remark or a deviation number. A result out of the buyer's specification needs a deviation number before work continues.
- Each change to a value adds an audit-trail entry in the editor's own name. Earlier values are never overwritten or removed, and a correction needs a reason.
- The audit trail is computer-generated and time-stamped by the server. The audit-trail screen shows the original and the new value, who, when and why.
- Records can't be deleted. Every saved version is kept in an append-only history, chained by SHA-256, so a change made outside the application is detected.
- The batch record can be produced as a PDF at any time, with every value, limit, signer, time, exception, correction with its reason, review and release. Logbooks export to Excel and PDF, including corrections and cancelled entries.
There is one record and one history. When someone, or something, asks "what happened at this step," there is one place to look and nothing to reconcile.
Where AI fits, and where it doesn't
Atto Pharma has AI agents, introduced from release 1.22.0. It helps to be exact about what they are and what they do.
The Help agents (Sales, Implementation, Support and, from release 1.38.0, IT & Security) answer questions from Atto Pharma's published documentation. Each answer lists the manual parts, guides or documents it is based on, so you can check anything important. These agents can't see your company's batch records, products, people or files. They see your question, with email addresses, phone numbers and record numbers masked, plus your role and site.
Separately, a preview feature, agents on this device, answers questions from the company's own records. The Quality Head agent, for example, answers questions such as "Which batches were made for each buyer?" from those records. The charts it can draw include:
- Out-of-limit values by stage
- Corrections by stage
- Deviations by step
- Batches waiting for QA
Those charts run across every batch of the product, for the Quality Head. The eLogbook charts, such as temperature and RH trends by area, with out-of-limit days shown in red, come from the same system.
The documentation is clear on one point. The answers are not part of the batch record. The batch's own screens and audit trail are the record. Atto Pharma records that an agent was used on a batch (who, when, which agent) but never the question or the answer.
The documentation describes agents that answer questions. It does not describe any agent that signs, approves or edits a batch record. Signatures stay with people. The database decides which role may give which signature: QA Head only for issue and release, QA or IPQA for section review, and a checker must be a different person from the doer. Each signature needs the signer's password.
For deviations, this means an AI answer can point you to where to look. The investigation, its conclusions and its closure stay in your quality management system, under your procedures.
What this means for your plant
If you are thinking about AI for deviation work, a few questions are worth asking of any system, ours included:
- Where is the record of truth? If it is spread across systems, an AI answer needs a reconciled source first.
- Is the audit trail one chain or several? Ask whether an edit made outside the application would be detected.
- Is the AI's output kept apart from the record? An answer should never be mistaken for the entry itself.
- What can the AI see? Know whether it reads your records, or only documentation.
- Who still signs? Confirm that approvals stay with named people and roles.
What stays with you
Software records and enforces decisions. It doesn't make them. For deviations, you still:
- investigate and close them in your quality management system;
- set internal limits that make sense for your process;
- train your people and approve who may draft, review and approve;
- validate each release and each product under your own procedures.
Atto Pharma eBMR supports 21 CFR Part 11, EU GMP Annex 11 and ALCOA+ expectations through its controls. Compliance depends on your procedures and your validation. The 21 CFR Part 11, cGMP and GAMP 5 Whitepaper and the One Approved BMR, Every Process Area whitepaper set out the controls in more detail. You can find the first in the app under Help → Documents. For IT Heads, the Application Security Whitepaper and the Trust and Security Overview cover hosting, access and records isolation.
Try it in a free 14-day pilot
If you would like to see one batch record, its logbooks and its audit trail together, you can try Atto Pharma eBMR or Atto eLogbooks in a free 14-day pilot. The pilot runs on the development site, with fictional data or your own non-GMP data. It gives your QA, manufacturing and IT teams a way to check the questions above against real screens before deciding anything.
See it on your own batch records
Atto Pharma eBMR and Atto eLogbooks turn your Word or PDF batch records and logbooks into electronic records, with electronic signatures, audit trails and limits checked as values are entered. Every new company starts with a free 14-day pilot.