Senior consultants were each spending around ninety minutes a day reading and routing client email. The team had grown faster than its systems and the triage rota was eating partner time.

We chose to keep the existing inbox and layer a quiet auto-sort and draft-reply system on top. No new app for the team to learn. No platform migration. Reversible inside an hour.

A grounded retrieval system over the firm's tone and prior replies, with intent detection wired to inbox labels. Draft replies prepared and waiting in the queue for approval and send.

Triage time fell from ninety minutes to eight. The drafts are accepted more than seventy percent of the time. Partner time returned to the work that actually pays.

The bookkeeping team was reading two hundred supplier invoices a month line by line. Errors slipped through. The accounts team was unhappy. The supplier reconciliation was always two weeks behind.

We chose extraction over OCR-and-template. The supplier list was long enough that templating would have collapsed, and short enough that a small grounded model could carry the work.

A PDF intake pipeline reading invoices into a clean spreadsheet schema, flagging any row outside expected range for human review, and posting the clean rows directly into the accounting system.

Reading time dropped from twenty hours a week to ninety minutes of review. Reconciliation moved from two weeks behind to current. The bookkeepers got their evenings back.

Every shipment moved through six spreadsheets and three shared inboxes. People were the integration layer. The team carried the system in their head and never quite went off-rota.

We chose an event-driven dashboard built on the data already in the spreadsheets. Migration in stages. No big-bang cut-over. Every stage proven before the next began.

A single operations dashboard with event-driven updates, integrated with the carrier APIs and the firm's accounting system. Notifications only when human attention was actually needed.

Shipment-touching time fell by sixty percent. The operations team got their evenings back. The shipments still moved.

The team was shipping twice a week through a deploy chain that broke on roughly one in three releases. The on-call phone rang almost every weekend. Engineering morale was poor.

Rebuild the deploy chain in stages while continuing to ship. Add observability before changing behaviour. Decommission the brittle pieces last, after the new path was carrying real traffic.

Continuous integration on every pull request, atomic deployments with automatic rollback, observability across the request path, and a structured on-call rota with runbooks.

Deploy frequency moved from two a week to fourteen. The on-call phone went quiet. Engineering retention improved over the next two quarters.

Brought us in for a single strategy engagement. The leadership team valued the way we worked and asked us to stay alongside them as the business grew.

Move from project-by-project to a small embedded retainer. One advisory thread, one engineering thread, one AI thread. Three integrated functions running on one shared rhythm.

Eighteen months of embedded partnership covering strategy, the technology stack the firm relies on, and AI integration across acquisitions, lettings, and back-office. Monthly cadence with leadership.

We now sit alongside leadership for every major decision. The firm has shipped systems that survived two reorganisations and three new hires. The relationship is the proof.