Case Studies — ALMO Nexus

Government · Production AI

An assistant the public service could actually put in production.

The problem: Staff answered the same policy questions repeatedly, reading from documents that changed regularly. Generative AI was the obvious fit and the obvious risk: no agency can ship an assistant that invents an answer or leaks personal information.
What we did: Built a Copilot Studio assistant that splits the problem in two. Deterministic flows handle anything with a correct answer — entitlements, process steps, referrals — so those responses cannot drift. Generative answers are grounded strictly in approved source documents, with enterprise PII guardrails and handling rules applied before anything reaches the user.
The result: Running in production inside a government environment, against the security and privacy review that implies. The split design is the reason it passed: the agency can point at exactly which answers are generated and which are not.

Wholesale · Process automation

The problem: A national wholesale group ran order fulfilment, rebate calculation and inventory across systems of very different ages, some with modern APIs and some with none. The gaps were filled by people re-keying data and reconciling spreadsheets — slow, expensive, and wrong often enough to matter.
What we did: Automated the processes end to end across the legacy systems and the APIs, rather than replacing anything. Built the reconciliation logic that proves the numbers agree at each handoff, with exception queues so the cases that genuinely need human judgement still get it.
The result: Substantial and sustained operating cost reduction for the group, and a set of numbers finance could trust without a manual check. The exception queues stayed — by design.

Banking · API orchestration

The problem: Opening an account touches a long line of systems that were never designed to be used in sequence — identity and document verification, fraud and sanctions screening, credit checks, the core banking platform, the card issuer. Any one of them can be slow, refuse the request, or go down entirely, and each failure used to land on a person to chase.
What we did: Built the orchestration that runs the whole sequence over APIs: each step called in the right order, results carried forward, and the awkward cases handled explicitly rather than by hand. A step that times out is retried safely rather than re-run blindly — critical when the downstream call creates a real account or orders a real card. Anything genuinely ambiguous routes to an operator with the full context already assembled, instead of a blank screen.
The result: Applications that used to wait on manual intervention complete on their own, and the ones that still need a human arrive with the work already done. The state of every application is visible at each step — which is what made the process auditable rather than merely fast.

Banking · Appointment management

The problem: Customers booking time with the bank were matched to staff through a mix of phone calls, local calendars and branch knowledge. Availability was frequently wrong, appointments were double-booked or quietly lost, and nobody could see the true picture across locations.
What we did: Built appointment management over the calendar and staffing systems already in use, so availability reflects who is genuinely free and qualified for that kind of appointment — not just an open slot. Customers book, reschedule and cancel themselves, with confirmations and reminders handled automatically, and every change reconciled back to the staff member's real calendar.
The result: Appointments stopped being a coordination task. Staff see an accurate day, customers get a time that holds, and the no-shows that reminders can prevent are prevented.

Retail · Platform re-architecture

The problem: Several online brands sat on a platform that had stopped keeping up — with the order volume, with the change the business wanted to make, and with the cost of running it.
What we did: Re-architected and rebuilt the platform end to end, including the order-management integration that connects the storefronts to what happens in the warehouse. Ran the program as well as the architecture.
The result: Delivered on time and on budget — which, on a rebuild of that size, is the result worth reporting.