Process Automation in Mortgage Advisory

The Challenge

The client had already automated - that was not the problem. The problem was that the automation did not run reliably. Agents failed to find CRM deals they had created, did not set up folders correctly, aborted API calls. In roughly eight out of ten cases somebody had to step in manually.

That is worse than no automation at all. A system you cannot trust still has to be checked - the effort remains, the confidence is gone. Over the years a tool zoo had also accumulated: several AI platforms running in parallel, plus CRM, document storage, automation layer and banking platform.

Our Approach

Reliability first, scope second. We built no new use cases while the existing ones were unstable. Root-cause analysis on the actual failures, more robust deal matching, proper error handling on API calls.

Separating sorting from evaluation. The decisive architectural cut: documents are first filed reliably - with upstream text recognition and to a fixed naming convention - and only then evaluated for content. Previously both steps ran intermingled, and a filing error poisoned the evaluation.

One lever at a time. Document filing, completeness checks per customer type against checklists, generation of the bank enquiry, automatic follow-up activities in the CRM. Each lever was built, quality-assured and stabilised individually until it ran on autopilot - no big bang, no proactive tool replacement.

Consolidation onto one platform. The AI tools running in parallel were consolidated onto a single EU-hosted, GDPR-compliant platform. For financing documents that is not a preference, it is a precondition.

The Results

The automations now run in a way the team trusts. Intervention has become the exception rather than the rule - and only at that point does the automation actually pay off.

Advisors work on the loan case again instead of on filing. That was the real goal; the clicks saved are merely how it is measured.