orangeAI
Discover · phase 1
AI Readiness Scorecard FreeA three-minute self-check.AI Opportunity Call FreeA straight steer on your best next step.AI Activation $2,499Switched on as a client, map in hand.AI Discovery & Roadmap AnchorKnow exactly where AI earns its place.Process CaptureHow the work happens, documented as an asset.Reasoning CaptureKey-person judgement, made an asset.
Design & Ignite · phases 2–3
Design & BuildThe engine that constructs it all, including skill design.
Amplify · phase 4
Chief AI OfficerAn accountable AI executive on retainer.Managed AgentsHigh-stakes work, run as a governed service.Business IntelligenceLive cockpits, assembled and analysed overnight.Governed AI RolloutYour whole workforce, switched on safely.
One methodology: Discover → Design → Ignite → Amplify. Every product sits on it.How I work →The Client Portal →Full map →
MethodologyCase studiesInsightsAboutClient Portal ↗
Free AI Scorecard
Home
Discover
AI Readiness Scorecard · FreeAI Opportunity Call · FreeAI Activation · $2,499AI Discovery & RoadmapProcess CaptureReasoning Capture
Design & Ignite
Design & Build
Amplify
Chief AI OfficerManaged AgentsBusiness IntelligenceGoverned AI Rollout
Company
MethodologyHow I workCase studiesThe Client PortalInsightsAboutClient Portal loginContact us · hello@orangeai.com.auStart here: free AI Scorecard or a call
orangeAI

We map how your business actually runs, then put AI where it earns its place. Brisbane, Australia.

What we do
What we doMethodologyCase studiesInsights
Company
AboutContactPortal login
Follow

Content lands on LinkedIn first.

The Lens, weekly
PartnersMicrosoft PartnerAnthropic · Claude Partner Network
Orange AI Pty Ltd (ABN 55 697 856 447) · Privacy · Terms · Contact© 2026 Orange AI Pty Ltd
← Home

Case studies

The systems we run. And what changed when we did.

We do not have a logo wall yet, and we will not borrow one. What we have is real systems, running today: one built for a client, the rest built for ourselves and in daily use. Here is what they do, in plain terms, with the numbers left unrounded.

An Australian uniform manufacturer · client work, published unnamed

Six weeks from first conversation to a working forecasting agent

Validated

Get demand forecasting wrong in this business and it costs you either way: millions of dollars of stock you cannot sell, or customers you cannot supply. We encoded the ordering team's own method into an agent that runs daily on live company data.

The problem

A family-owned Australian workwear and uniform manufacturer, second generation, vertically integrated from garment design through its own manufacturing to wholesale and retail. Its hardest decision is demand forecasting, and the money runs in both directions.

  • Order too little and you cannot supply your customers. In the worst case you lose them.
  • Order too much and millions of dollars sit in stock you cannot sell, locking up working capital and paying interest on the facility that funded it.
  • Orders are committed six months ahead of the season, a large share of annual sales lands in a single month, and the catalogue runs to thousands of style-and-size combinations.

What we built

DiscoverFour weeks. The operating model mapped end to end, and the reasoning behind the key expert decisions written down so it stopped living in one person's head. Out of it came nineteen scored opportunities, sequenced into waves.
Build and DeployTwo weeks later the agent was live on company data: the ordering team's own method encoded into a governed engine, reading their sales and inventory records and producing proposals in the exact workbook they already use. It reads 136,000 sales records in about a minute.
IterateThe client came back with corrections on which data source was canonical, how work in progress should be treated, and his own template as the binding format. We rebuilt to his rulings and cut over inside 48 hours.
GovernEvery run is reconciled to source and scanned for formula errors before release. It runs in supervised validation with the ordering team ahead of the next major ordering round. The agent proposes. The experts decide.

Value delivered

Running the full portfolio through the data-capture step took the best part of a week. It now takes minutes, overnight on a schedule, which gives the business effectively real-time access to its own numbers every morning.

That time goes back to where it actually matters: the judgement on what to produce, how much of it, and when. The prize is the working capital, and the roadmap put a number on it, agreed with the client. Between $400,000 and $500,000 released off the back of a 10 per cent improvement in forecast accuracy, which is exactly what better data in front of the ordering team every morning is there to buy.

Minutes
to run the full portfolio, down from the best part of a week, now overnight on a schedule
$400-500k
working capital identified, at a 10% improvement in forecast accuracy
60 of 60
workbooks reconciled to source, zero variance, no formula errors
6 weeks
from first conversation to a working agent on live company data

Our own business · client zero

The assistant that runs our own inbox

Running

The first system we built was for ourselves. Before we sold anyone an agent, we put one to work on the job that was slowing us down most: the inbox.

The problem

We are a founder-led advisory with no admin team. The email still has to be read, sorted, filed and answered every day, and for a long time the founder was the only person who could do it. The inbox was the bottleneck.

What we built

An assistant that watches our company inbox around the clock. It sorts every message twice over: by what it is about, and by whether it actually needs a reply. Anything that is only for information is filed automatically. Anything that needs an answer gets a first-draft reply written in the founder's own voice, saved ready to read. It never sends. Every draft waits for one read and one click. The same daily brief surfaces new mail sitting in the shared company mailboxes we do not otherwise watch, and lays out the day's meetings.

What changed

The inbox now sorts itself, and we open a short daily brief instead of a full inbox. The safeguards are the whole point. The assistant files freely but never sends on its own, and everything it moves is listed back so nothing disappears quietly. When it is not sure, it leaves the message alone and flags it rather than guessing. The system is held to 628 automated tests, run before any change ships. It has been in daily use on our own business since June.

A Brisbane family investment office · in daily use

One view of a family's whole portfolio

Running

A live portfolio system, built and run on the same stack we sell. It answers a basic question that used to take an afternoon: what do we hold, what is it worth, and where is the risk?

The problem

A family's investments were spread across roughly ten custodians, each with its own login and its own statement. There was no single, current picture of what was held, what it was worth, or where the risk sat. Answering a basic question meant opening ten places and adding it up by hand.

What we built

A private portfolio application that pulls every holding into one view: allocation, performance, income, borrowing, unrealised position, and the gaps in the record that still need attention. It runs without a database. The underlying data lives in the family's own document store, so control never leaves the family. A market-news feed reads the day's financial press each morning and afternoon, scores every article for relevance, and keeps only the ones worth reading. And the whole book is exposed to the family's own assistant through a single connector carrying twenty-one query and update tools, so the office can ask questions of the portfolio in plain language instead of opening ten logins.

What changed

Ten logins became one view. The daily read went from a manual scrape to a scored shortlist. Price updates that once meant editing files by hand now run through a preview-first tool that shows the change before it is written. The record's own blind spots are surfaced, not buried.

◎

One more is on the way. Our estimating agent cleared its evaluation gate against real client data, line by line, before we would quote the work. It joins this page once that client has signed off. The forecasting study above is published with the client’s agreement and their name held back by us, not by them. We would rather wait for consent than borrow a logo.

We are Customer Zero

Everything we’ve built. And we run on all of it.

Before Orange AI sold anything, we productised our own operations: the consultancy, a family office, and our founder’s working day. This is the inventory. Numbers only mean something next to what they measure; every line here is a real, countable artefact. Validated means proven against real data behind our evaluation gate before any client relied on it.

SystemWhat it doesMeasured resultStatus
Ada, our own assistantA production email and calendar chief-of-staff on our company tenant: triaging, filing and drafting on four live scheduled jobs.628 testspassing, 4 live jobsRunning
Estimating agentPriced a 500-staff contractor's live cable scope from their raw take-off with no answer key, before we would quote the work. Where it was unsure it showed its working rather than guessing.35 of 37codes found, 34 priced to the centValidated
Forecasting agentA manufacturer's demand-forecasting workbooks, machine-checked against source data on every run before any output is released.60 of 60reconciled, zero varianceValidated
The signal feedThe AI-intelligence signals on this site are researched and written by our agents overnight. You can read them right now.Publicon this siteLive
Research deskConsecutive dated intelligence reports generated by our autonomous researcher, a countable trail of continuous agent work.92 reportsand countingRunning
Governance, at homeOur own review pipeline found and purged a real data-privacy leak before a human saw it. We hold ourselves to the bar we sell.Caught firstby the pipelineStanding
100+
bespoke skills, built and governed (plus 100+ installed automation recipes)
10+
named agents across research, sales, delivery, governance and operations
2
production client apps (portal + intake), plus the public site
12+
live system integrations
5
named proprietary frameworks in every diagnostic
4
phases, one methodology: Discover, Design, Ignite, Amplify. Every product sits on them

The agents

10+ named
AdaExecutive assistant on our own M365 tenant: email triage, filing, drafting. 628 passing tests, four live scheduled jobs.
Discovery DirectorOrchestrates the client engagement pipeline end to end, with human approval gates at every stage.
ResearcherDaily scheduled scans, deep investigations, competitive intelligence. 92 consecutive dated reports.
CRO agentOwns the sales pipeline system of record; writes a revenue brief from the live CRM every day.
Demand agentSources and scores leads and drafts outreach. Drafts only: a human sends, always.
Vault LibrarianGovernance QA: audits our whole knowledge base for structural integrity and rule conformance, weekly.
WilsonChief-of-staff agent orchestrating the personal fleet: daily logs, memory, routing.
The scheduled fleetOvernight transcript reviewer, nightly technical researcher, weekly knowledge-base auditor, morning amplifier, nightly task triage: the automations behind the 24-hour feed on our home page.
Health & travel & moreA personal health-intelligence officer, a family travel agent, nightly task triage. We run our own lives on the same stack.

The software

Production
Client portalLive production app: engagement dashboards, deliverables hub, structured intake, governance cockpit, agent workspace. Tour it →
Intake appGuided multi-stakeholder structured intake: large question bank, AI-assisted follow-ups, voice capture, e-signature.
This websiteIncluding a public research feed refreshed automatically from our internal daily intelligence work.
Family-office appWe run our founder’s family office on the same stack: portfolio tracking, pricing, performance. Personal data stays personal.

The integrations

12+ live
CRMAccountingTask managementProspecting & enrichmentWeb research & monitoringGoogle WorkspaceMicrosoft 365Meeting captureBrowser automation & QAResearch synthesisInternal API + operator connectorNightly intelligence pipelines

Named vendors and connection detail shared in an engagement, where they’re relevant to your stack.

The frameworks & IP

5 named proprietary + methodology
Operating Stack™Five layers plus governance: how AI-ready an organisation is.
Opportunity Register™The ranked, costed register that anchors every diagnostic.
Tri-score™Return, risk reduction, strategic upside: one score per opportunity.
Two-Tier Listening™Leadership’s account vs the frontline’s; the gap is the signal.
Diagnosis–Prescription Separation™What to do and why it pays, kept strictly apart from how it gets built.
4-phase methodologyDiscover, Design, Ignite, Amplify. See the methodology →
◎

A note on honesty. Some of what’s listed is running in production today; some is validated and staff-gated behind our own activation discipline. Every skill and agent carries a formal status, and nothing goes client-facing without clearing its gate. We’d rather show you a governed pipeline than an inflated logo wall.

Book a callWhat we do with all this