Case study · Research and innovation (R&D&I)

Document AI and Salesforce integration for regulated environments

Translates without losing the format. Extracts handwritten notes. Creates prices and opportunities in the CRM.

I translation that keeps the format

Handwriting extraction (vision + OCR)

Teams agent connected to Salesforce

Where it all started

Translating regulated documents by hand and feeding the CRM one by one

An international CRO in a regulated environment carried two manual burdens. Technical and quality documentation arrived in another language, on templates with watermarks that covered almost every page, and with handwritten sections.

It had to be translated keeping the exact format, and the handwriting transcribed by hand.

In parallel, the commercial team created price lists, opportunities and quotes one by one in the CRM.

The challenge

Translate regulated documents keeping format and structure, with no margin for error.

Watermarks covering almost every page, complicating any automation.

Handwritten text inside the documents, transcribed by hand: slow and error-prone.

Two templates per document and the handwriting overlaid, which had to be reassembled piece by piece.

Price lists, opportunities and quotes created one by one in the CRM.

Traceability and access control to respond to audits.

What we proposed

Two AI layers on top of the platform the CRO already had on Azure Databricks. One cleans the watermarks, translates keeping the template format and, with computer vision and OCR, extracts the handwriting and places it back where it belongs. The other is a multi-agent AI system connected to Salesforce that creates prices, opportunities and quotes applying business rules, with validation before it reaches the CRM. A single point of entry: a chatbot in Microsoft Teams.

The shift

Day and night

The repetitive work (translating, laying out, transcribing and creating records) moves to the AI. Judgment and validation stay with people.

Before

Manual translation and layout, document by document

Watermarks and handwriting that blocked any automation

Prices, opportunities and quotes created one by one in the CRM

No clear trail for audit

Now

The AI cleans, translates and keeps the format; it stores the original and the final version

The handwriting is extracted with vision + OCR and placed in its position

An agent creates prices, opportunities and quotes in Salesforce, with business rules

Everything is stored and traceable in a governed Data Lake

document AI flows (translation + handwriting extraction)

artifacts the AI creates in Salesforce (prices, opportunities, quotes)

chatbot in Teams as the single entry point

The value generated

What changes in a regulated operation

Translation that respects the document

It keeps format, structure and layout; nothing to redo by hand.

Handwriting too

Computer vision and OCR extract the handwriting and place it in its original position.

An agent that works in the CRM

A multi-agent AI system creates prices, opportunities and quotes in Salesforce from the data.

Business rules that are respected

Every request is validated before it reaches the CRM.

Traceable for audit

Originals and final versions are stored, with access control and retention.

On top of what you already have

It builds on the existing Azure Databricks platform, without duplicating infrastructure.

What the client says

What used to be translating, laying out and transcribing by hand now arrives resolved and traceable. And the commercial team creates prices and opportunities in the CRM by talking to a chatbot. We gain time without giving up the control a regulated environment demands.

Operations Management

CRO / Life sciences

F&Q

Frequently asked questions

What is document AI in a regulated environment?

It is the use of artificial intelligence to translate, read and structure technical and quality documentation while keeping its original format, with end-to-end traceability and access control. At BertIA, an AI consultancy based in Barcelona, we apply it in CROs and life sciences environments where every document must be auditable.

How is it different from a machine translator or a standard OCR?

A standard translator returns plain text and loses the layout; a generic OCR does not distinguish which part of the document matters. Here the AI first cleans the watermarks, then translates keeping the template format and, with computer vision, isolates the handwritten text and places it back in its exact position on the translated version.

Does the conversational AI create data in the CRM on its own? Is it reliable and secure?

It does not act without control. The chatbot receives the request, validates it against the business rules and only then generates or updates price lists, opportunities and quotes in Salesforce. The person keeps the decision; the AI performs the repetitive work in a controlled way.

How are traceability and compliance guaranteed in a CRO?

Original documents and their final versions are kept in a governed Data Lake, with retention policies and access control. Every step is recorded, so the process is reconstructable and auditable, a basic requirement in a regulated environment.

What does it take to implement it on an existing data platform?

If there is already a platform on Azure with Azure Databricks, at BertIA we reuse it as the processing core and add the document AI and Salesforce integration layers on top. That lowers the cost and time of implementation and avoids duplicating infrastructure.