Skip to content

Delivered work

What we have actually shipped.

Anonymised to industry level. We do not publish names and have not asked permission to, so there is no logo wall here. What you will find is the work, the system we built, and the number it moved.

Each one leads with the shape of the problem rather than the technology, because the shape is the part that transfers between industries. Four of the ten are the same client — fourteen engagements deep, and still running.

Benefits administration

Fourteen engagements, and we still run it

Data arrives from parties who will never agree on a format, and someone reconciles it by hand before anything downstream can move.

What we built
A medallion-architecture data platform on Microsoft Fabric, third-party administrator and pharmacy-benefit integrations, pre-certification workflows, invoice automation built three separate times as the requirement changed, a data-quality programme, and robotic automation over the ticketing system.
Outcome
Fourteen sequential engagements since 2020. We still operate the platform, with weekly operational alerting. Four of those fourteen are below.

If billing, reporting or fulfilment waits on someone finishing a reconciliation, this is the same build.

Benefits administration · claims data

Every vendor sent a different file, on a different day

Nobody can see across the whole picture, because the picture is assembled by hand each cycle and is stale by the time it exists.

What we built
Automated ingestion across formats, a rule-based validation engine, exception flagging into a human review queue, one reconciled claims database as the single source of truth, and dashboards tracking the health of every vendor feed.
Outcome
Visibility into high-cost claims, anomalies, duplicates and billing errors that had not been visible before — driving cost savings of more than $1.5M. Standing up a new point solution on vendor data fell from 12+ months to one or two.

If onboarding a new vendor, carrier or supplier takes months because nothing is standardised, this is the same build.

Benefits administration · reconciliation

Finding the real variances took longer than fixing them

Two records disagree. Most of the disagreements are fine. Finding the few that are not is the entire job, and it is done by eye.

What we built
A rule-based matching engine comparing census files against carrier invoices, exception categorisation by cause, a resolution workflow with audit logging, and reporting that surfaced which variance types kept recurring.
Outcome
80% reduction in manual effort per reconciliation cycle — 3,200+ hours a year. Only true exceptions reach a person.

If your team compares two systems every month and most of what they find turns out not to matter, this is the same build.

Insurance operations · invoicing

Hundreds of invoices a month, held together by macros

A process that has to be exactly right is running on spreadsheets, which means it can neither scale nor be audited.

What we built
A cloud platform ingesting third-party data, configurable inputs and templates, validation against internal sources, exception flagging for review only where needed, and a full audit trail from raw input to final invoice.
Outcome
90% reduction in manual invoice handling, and a hundred invoices generated in under three minutes. Full audit readiness, with logged activity and access control.

If growing volume means hiring, and nobody can reconstruct how a number was produced, this is the same build.

Benefits administration · AI

The answer existed. Finding it required knowing where to look

The organisation holds the answer somewhere across its tables, its documents and its complaint history, and only a handful of people know how to retrieve it.

What we built
Natural-language query over structured data, AI-assisted search across policy documents, theme detection over free-text complaints, and context delivered where people already work — an underwriter gets the group history at renewal, a case manager gets the claims picture while the member is still on the line.
Outcome
Live in 2026. Judgement-based workflows now arrive pre-assembled for a person to finalise, rather than being researched from scratch each time.

If the useful answer is technically available but practically out of reach, this is the same build.

Freight operations

Dispatchers spent the day answering a question the system knew

Status lives in four places that do not talk to each other, so a person becomes the integration layer.

What we built
Automated status capture and distribution, so information reached whoever was asking without a human relaying it — a web dispatch interface for planners, a mobile app for drivers accepting and confirming loads.
Outcome
85% reduction in phone-based coordination, digitised proof of delivery, and on-time performance improved by adjusting in real time.

If somebody on your team is the integration layer between systems you already pay for, this is the same build.

Freight operations

Every customer question interrupted the one person who could answer it

The people who need the answer cannot reach it, so they interrupt the people who can. Both sides lose the day.

What we built
GPS feeds integrated with the shipment system: continuous asset tracking, geofence-triggered status updates, delivery windows calculated from actual movement, and a fleet view for operations.
Outcome
70% fewer "where is my shipment" calls, proactive alerts on delays, and idle-time visibility that improved asset utilisation.

If your team is effectively a lookup service for customers who could serve themselves, this is the same build.

Freight forwarding

Approvals lost in email threads, and the delay compounding

A process with a defined sequence is being run inside an inbox, which has no sequence.

What we built
Load confirmations generated, routed for signature, countersigned and filed electronically, with a central repository and live status on what was pending versus signed.
Outcome
3x faster load-confirmation cycles, version history on every signed document, and materially less admin chasing.

If something with real steps and real deadlines is being tracked in email, this is the same build.

Manufacturing and automotive

Answering one traceability question meant going to find a binder

The record exists. It is simply not queryable, so retrieving it costs a person and an afternoon.

What we built
Full production data capture at source, with genealogy queryable directly instead of reconstructed from paper.
Outcome
Traceability lookup from hours to seconds, operator paperwork eliminated, and hidden inefficiencies surfaced that improved machine utilisation and quality control.

If your answers exist but live in binders, shared drives or somebody's memory, this is the same build.

Investment operations

An audit rebuilt by hand, every single cycle

Work that recurs on a known schedule is treated as brand-new work every time it recurs.

What we built
Automated assembly of the audit pack from the systems of record, with exceptions surfaced rather than hunted, across complex instrument positions.
Outcome
Over four hours down to under seven minutes. Operational capacity expanded without adding headcount.

If your team rebuilds the same pack every quarter under a deadline, this is the same build.

Nothing here is aggregated into a success rate or a total-value figure, because we have not measured one. Each number above belongs to the engagement it came from.

Is yours one of these shapes?

Most operational problems rhyme across industries. Twenty minutes is usually enough to tell whether yours does.