How a gap analysis found 40% faster document processing

← Back to all articles

The details in this case are anonymised, but the shape of the engagement is typical: a professional services firm with a small back-office team that was quietly drowning in inbound paperwork: contracts, invoices and intake forms, all arriving by email and requiring manual review before anything could move forward.

The starting point

During the discovery phase, we sat with the two people responsible for document intake and timed a representative sample of their work. Each incoming document required roughly 8 to 12 minutes: opening it, identifying the type, extracting the key details, and entering them into the firm's case management system. With dozens of documents arriving daily, this was consuming the better part of two full-time roles.

What the audit found

The firm was already paying for a Microsoft 365 plan that included AI-assisted document processing features, but they were unconfigured and unused. The case management system also had an existing API that nobody on the team knew could be connected to anything else.

The roadmap

Rather than proposing a new platform, the roadmap focused on three changes: configuring the existing document intelligence tools to extract structured data from the most common document types, building a simple routing step that flagged anything the tool wasn't confident about for human review, and connecting the output directly to the case management system via its existing API.

The result

Over a phased six-week rollout, average processing time per document dropped from roughly 10 minutes to about 6, a reduction of around 40%. Critically, the two staff members weren't replaced; their roles shifted toward reviewing the documents the system flagged as uncertain and handling exceptions, which they reported as more engaging work than the repetitive data entry it replaced.

Why this is typical, not exceptional

What made this engagement work wasn't a clever piece of technology, but the audit. The tools already existed inside software the client already paid for. The gap was configuration, integration, and a clear view of which document types were worth automating first. That's the pattern we see again and again: the highest-value AI work is rarely about acquiring something new.