Case Study: Intelligent CRM with OCR and Automatic Synchronisation

In funeral services, a client file is not simply a contact record and a commercial opportunity. It aggregates contractual documents, documents transmitted by email, scans retrieved in the field, billing items, sometimes content shared by multiple parties, and often urgent deadlines that require immediate action. In this context, a document is not just an attachment. It sits at the core of the business process.

The problem emerges as soon as these documents live in a fragmented environment. Useful information may be visible in a PDF but absent from the CRM. A document may be stored somewhere without being linked to the right case. An update may be made in a commercial tool without being reflected in the document history. The cumulative effect is well known: teams lose time, data degrades, and operational continuity depends on individuals who know where to look rather than on a system that knows what to do.

This is precisely the terrain on which the project was framed. The objective was not to create yet another storage space, nor to add decorative AI functionality. The goal was to transform a document-heavy environment into a genuinely operational CRM — one capable of absorbing incoming documents, structuring their content, and making them immediately useful for commercial and administrative tracking.

The first challenge was very concrete: too much time was disappearing into re-entry. When a document arrived, it had to be opened, read, matched to a case, and its key information extracted and re-typed into the CRM or management tool. This kind of operation seems simple at small scale. It becomes costly as soon as it repeats dozens of times a week, especially in a sector where urgency and precision must coexist.

The second challenge was structural. The tools did not always speak the same language. A contact might exist in the CRM, a document in a shared folder, and billing information in Sellsy — each block of truth becoming partial. When an organisation lives with multiple versions of the same information, it suffers not only from inefficiency. It suffers from a loss of confidence in its own system.

The third challenge, often underestimated, concerned documents disappearing in the mass. A document that exists but is not linked to the right case is practically a lost document. The team must then go back through emails, exports, local scans, or shared directories. This document hunt slows everything down, exhausts teams, and increases the risk of error at the most sensitive moments.

The implemented response is grounded first in a multimodal OCR engine powered by Gemini. The challenge is not simply to convert an image into text, but to reconstruct an actionable understanding of the document. This means recognising the nature of the document, identifying important zones, extracting structured fields, and preparing data ready to enter a business workflow. This approach handles real documents — imperfect, varied, and sometimes incomplete — which is exactly the kind of material teams encounter daily.

Next, the system introduces a validation layer before any automatic creation. This is a decisive point. In a serious CRM, automation only has value if it reduces cognitive load without surrendering control. Extracted data can therefore be verified, completed, or corrected before feeding business objects. The AI prepares the work, humans retain control over sensitive cases, and the system learns from genuinely useful patterns.

Once that step is passed, synchronisation to Sellsy becomes the natural extension of the flow. Validated information can update the commercial CRM without going through a manual re-entry phase. This immediately changes the quality of tracking. Teams no longer work with a system lagging behind the documentary reality. They work with data that follows the document almost in real time, in a structured and reusable format.

Finally, the project addresses a topic often neglected in AI deployments: document routing. A good OCR system is not enough if the document ends up in the wrong place. Each document must be linked to the right case, the right client, and the right point in the cycle. It is this articulation between reading, validation, synchronisation, and filing that transforms a collection of automations into a truly operational CRM agent.

The first visible result is the reduction in processing time. When a team no longer has to read an entire document to re-type its content into multiple tools, a case advances faster and with less cognitive fatigue. This gain is not limited to a few minutes saved. It changes the speed of transition between receipt, qualification, and exploitation of the data.

The second result concerns information quality. By structuring extraction, enforcing validation, and synchronising to Sellsy from clarified data, the CRM gains coherence. Duplicates decrease, fields are better normalised, and teams regain confidence in what they consult. This improvement in precision is fundamental, because a CRM is only useful if it is considered reliable by the people who use it every day.

The third result is the elimination of manual re-entry on the automated scope. For correctly recognised and validated documents, information flows without being copied field by field. This is a key point for industrialisation. An organisation can accept human validation. It cannot sustainably accept transforming every incoming document into a repetitive administrative task.

The main lesson is straightforward: an AI CRM agent is not a chatbot plugged into a database. It is a system that understands flows, respects business constraints, handles ambiguous cases, and integrates into a real environment composed of emails, documents, validation rules, and external synchronisations.

This case also shows that AI does not replace process design. It makes it more powerful, provided the framing is clean. When the data model, validation rules, and filing logic are clear, OCR and automation produce genuine value. If those foundations are blurry, AI primarily accelerates the disorder.