The challenge

The client’s accreditation process was slow, inconsistent, and reliant on disconnected tools and manual workarounds. Key issues included:

• 6-month average assessment cycle
• Unstructured data buried in documents and email
• Frequent scope creep and duplicated effort
• No single source of truth for key decisions
• High friction between teams and outdated workflows.

The solution

From first principles, we reimagined the process to be digital, intelligent, and outcomes-focused; supported by a Power Platform prototype and high-fidelity designs. Key features of the redesign included:

• Shifting scope reviews to the start of the process to prevent late-stage issues
• Digital-first assessments using structured data instead of Word documents
• Risk-based routing to assign assessments by complexity
• A streamlined 8-stage workflow with automation built into each step.

The challenge

One of Australia’s leading energy providers had a B2B sales support process bottlenecked by repetitive manual work. Every broker request required sales support team members to read an email, create a CRM opportunity, extract and clean usage data, and upload documents; often taking hours and relying heavily on a few key staff. With volumes growing and margins tightening, the process wasn’t scalable.

The solution

We designed and deployed an AI-powered sales agent to automate the full sales support flow for single-NMI acquisition requests. This included:

• AI email triage to classify broker emails and trigger the right workflow
• Automated CRM and SharePoint setup with the correct structure and metadata
• Usage data cutting and remediation, generating pricing-ready load files, with a lightweight front end portal to handle semi-automated scenarios
• Built-in exception handling, ensuring failed or incomplete requests were flagged early and quickly actioned.

The technology

The solution combines Azure OpenAI for natural language processing, rule and logic-based usage data transformation, and a lightweight React frontend portal to support semi-automated workflows. It also integrates with existing CRM and document tools to deliver automation without disruption, thereby accelerating delivery and adoption.

The challenge

The organisation manually created custom meeting minutes after each board meeting, requiring multiple people over many hours. Staff had to review transcripts and format them to specific agenda structures with custom templates, and existing meeting minutes couldn't match their formatting requirements.

The solution

Utilising Azure OpenAI, we built a custom meeting minutes automation platform to transcribe 2-8 hour board meetings with up to 12 participants located in the same place. Once users upload a meeting agenda and transcript, AI processes the content against the agenda structure and generates properly formatted minutes in Word. The solution also removes verbose explanations of areas that should just be noted, as well as summarisation and style to comply with regulation. Extra features include custom branding and manual edit option prior to final circulation.

The technology

Built with ReactJS frontend and powered by GPT-4o Turbo via Azure OpenAI, with real-time WebSocket communication.

The challenge

Non-English speaking members of the community and those with English as a second language (ESL) faced issues with interpreter access, navigating online info, and communicating confidently via chat or phone. In parallel, contact centre staff were facing high cognitive load, inconsistent service outcomes, and process inefficiencies when supporting this community cohort.

The solution

A practical, people-centred prototype across chat and voice channels:

• Live chat translation: 46 conversations tested in 15 languages using Crisp AI (an off-the-shelf omnichannel engagement platform with native AI-driven language translation capabilities)
• Voice agent: custom AI solution for 1:1 translated conversations in Hindi, Vietnamese, Urdu and Mandarin with a simple user interface
• Focus areas: translation accuracy, ease of use and handling government jargon.

The technology

Crisp AI used to test out the live chat experience, and a custom OpenAI voice solution (using the real-time voice API and a simple React web interface) used for prototyping the voice agent.

The challenge

A professional accreditation body faced significant bottlenecks in its accreditation content review process. Key issues included human error when processing lengthy content (2-3 hour videos), staff only reviewing summaries instead of complete materials, and slow document processing speeds of 1-2 days. The manual approach wasn’t scalable in line with increasing submission volumes.

The solution

We built an intelligent system to process multiple content formats against established accreditation criteria:

• Content creators upload materials (PDF, video, presentation, text)
• AI analyses complete content against the organisation's knowledge base and rules
• System verifies claimed CPD areas and calculates appropriate hours based on content complexity
• Generates detailed accreditation reports with recommendations
• Handles multi-format processing with automated transcription for video/audio content.

The technology

React frontend dashboard, supported by GPT-3.5 Turbo for content analysis and rule-based evaluation and DeepGram for audio/video transcription and analysis.

The challenge

One of Australia’s largest fully integrated poultry producers was struggling with a common agricultural problem: unpredictable demand leading to high waste volumes and missed opportunities.

The producer’s existing forecasting followed traditional statistical models which were functional, but significantly less accurate than modern AI approaches. Forecasting inaccuracy resulted in two costly scenarios:

• An over-production of premium products that would spoil
• An under-estimation of demand with disappointed customers.

The solution

An AI-powered demand forecasting system and analytics dashboard which drew from two years of historical sales data. The dashboard now offers teams a real-time view of demand and production planning: no spreadsheets, no guesswork.

Under the hood, multi-agent AI architecture does the heavy lifting across the producer’s supply chain:

• One agent identifies demand patterns
• Another optimises production across facilities
• A third spots sales opportunities
• A fourth handles scheduling.

Built with Azure Data Factory and Databricks, the new automated forecasting pipeline updates models in real time. Forecast outputs are also exposed via APIs to interface seamlessly with existing systems and minimal disruption to operations.

The technology

The platform runs on a React dashboard to facilitate intuitive demand planning. Behind the scenes, it’s powered by Python and proven ML libraries like NIXTLA, Prophet, and scikit-learn; enabling multi-model forecasting with automated selection and fine-tuned hyperparameters. Azure infrastructure features end-to-end, including Data Factory, Databricks, and Data Lake Storage. The system connects to existing production environments via REST APIs.

The challenge

A national charity was struggling with manual cataloguing of clothing donations. Staff spent hours sorting donations, identifying types, sizes, and demographics, creating inconsistent inventory descriptions. The sluggish donation processing made it harder to match donations with those in need.

The solution

We developed an AI system to automatically analyse and categorise clothing from photographs:

• Smart classification: automatically identifies clothing type, size, and target demographic from photos
• Tag recognition: extracts information from visible clothing tags when available
• Voice and text correction: staff can easily correct classifications using voice commands or text inputs.

The technology

React application for an intuitive user interface, enabled by GPT-4 Vision for image analysis and clothing classification and DeepGram for voice-to-text conversion.

The challenge

While the client’s phone-based sales were converting strongly, the same couldn’t be said for online leads. 75% of users who commenced their journey online opted out before completing or speaking to an advisor. Key challenges included:

• Friction-heavy online forms that felt static and impersonal
• Trust gaps that made users hesitant to share details
• A disconnect between digital and phone experiences
• Legacy constraints that limited experimentation.

The project goal was to prototype a new kind of digital experience: one that could deliver clarity, warmth, and tailored guidance, without a human agent.

The solution

We developed a voice-first, AI-led advisor; designed to test whether online quotation journeys could feel more like expert conversations. Key features:

• Voice-led conversation agent with dynamic questions, objection handling and emotional intelligence cues to mimic the experience of high-performing human advisors
• Multi-AI agent orchestration with skills to handle conversation, policy analysis, customer classification, product comparison, product recommendations, transcription and sentiment, plus document analysis
• Smart handover logic to enable seamless transition to a human agent
• Omnichannel by design: voice-first, with visual flows and mobile/web parity.

The technology

• Voice tech: ElevenLabs (for stability and production readiness)
• AI model strategy: Google Gemini (documents), GPT-4 (conversation), custom models (recommendations)
• Frontend: React web app with real-time voice and mobile-first design
• Architecture: event-driven, scalable, omnichannel-ready
• Compliance: built-in explainability and audit trails aligned to ACCC expectations.

The challenge

A construction company was drowning in project documentation stored across SharePoint. With each project folder containing 70-80GB of data, staff had to trawl through large volumes of documentation for answers to their questions.

The solution

We developed a custom Teams-integrated knowledge assistant that enables natural language queries across massive project repositories. On a simple voice prompt in Teams, the assistant handles intelligent project context switching, implements permission-based access, and provides accurate project-specific responses.

The technology

Built using the Microsoft Teams interface with a Python backend, powered by OpenAI GPT-4o via Azure with SharePoint integration.

The challenge

A leading healthcare staffing agency was losing time, revenue, and clients to a broken shift processing workflow.

The process was manual whereby every shift request was raised via phone call. Staff would listen to the call, transcribe details, and enter them into a workforce platform. On average, it would then take 24 hours for a request to translate into an available shift. The effects of the sluggish workflow included:

• Revenue leakage: shifts went unfilled, and commission opportunities disappeared
• Frustrated clients: urgent cover requests were left waiting
• Heavy admin load: staff spent hours resolving jargon-filled transcripts
• Lost deals: faster competitors picked up time-sensitive opportunities first.

To make things harder, the language itself was complex. Everyday calls were packed with nuanced terminology (“scrub tech”, “ortho”, “double cert nurse”) requiring precise categorisation to meet compliance and staffing standards.

The solution

Rather than optimising the existing workflow, we reimagined it via a domain-smart AI agent tailored for healthcare staffing. The new agent could listen like a recruiter, understand like a clinician, and act like an ops assistant.

• Understands the language: training the custom language model on healthcare staffing terminology means it can decode shorthand, match roles to workforce codes, and distinguish between similar terms, all with compliance-level accuracy.
• Handles complexity: with single calls often including multiple shift requests, the AI can untangle them, log each separately, and match roles to the right requirements in an instant.
• Acts in real time: the agent monitors call transcripts in SharePoint, triggers the right workflow, and syncs directly with the workforce management platform with no lag or re-keying.

The technology

• Real-time processing pipeline: built for speed and scale, the system watches for new call transcripts in SharePoint (automatically placed there by the recording provider), processes multiple shift requests in parallel, and pushes structured data straight into the platform with no human delay or bottlenecks.
• Smart data standardisation: a translation layer makes often messy human input usable by machines. The AI normalises terminology, converts language into compliant shift codes, and validates every entry against industry regulations, all while generating a full audit trail for quality control.

The challenge

As customer expectations shift and contact volumes grow, insurers are looking for new ways to deflect calls, increase self-service, and streamline simple interactions, without compromising compliance or customer experience. This project set out to explore whether quoting (traditionally a human-led conversation) could be handled autonomously by an AI agent.

The solution

An AI voice agent capable of providing real-time quotes to new customers, directly within the call flow, involving:

• Conversational voice agent via ElevenLabs, designed to sound natural and confident
• Quote capture workflow to collect key details (e.g. property, location, cover level) in free-form conversation
• Dynamic pricing integration with the insurer’s pricing engine to retrieve and present accurate quotes instantly
• Edge case handling for issues like unclear responses, out-of-scope requests, or regulatory boundaries (e.g. not giving advice or steering)
• Compliance-safe responses, designed to inform without influence, and hand off when necessary.

The technology

Combined ElevenLabs for a high-quality AI voice agent, with direct integration with the client’s existing pricing engine.

The challenge

A national supermarket retailer had rolled out ChatGPT Enterprise to 5,000 corporate employees and built a first network of about 120 AI champions. The next step was to move beyond early adopters to practical, confident use across the whole corporate workforce. That meant three things:

• Equipping senior leaders to model AI use and champion it in their teams
• Training thousands of people at scale, across very different business functions
• Building a cohort of ambassadors to sustain adoption after the engagement ends

The solution

A six-month accelerator (April to October 2026) built around ChatGPT Enterprise, in three streams:

• GM Leadership Series: 10 in-person working sessions of 90 minutes, where general managers use ChatGPT on real problems from their own week
• Practical AI Tool Training: 26 webinars across six progressive modules, each module run four times so the whole workforce can attend, with every session recorded
• Ambassador Cohort: 60 nominated ambassadors in three cohorts of 20, in a four-week sprint working with ChatGPT as an assistant, a thinker and a creator, and ending in a showcase of their own work
• Supporting all three streams: drop-in support sessions, business unit deep dives, weekly working sessions and monthly programme reporting

The challenge

A national airline had rolled out ChatGPT Enterprise and set up an internal network of AI champions. The next step was to move 2,000 corporate staff, across very different functions, from general awareness to confident, practical use. That meant delivering relevant training at scale, in a short window, without losing quality or engagement.

The solution

AI Take-Off Week: four days on site across the airline’s corporate offices.

• Eight 45-minute masterclass tracks, from ChatGPT 101 and 102 to deep dives on data analysis, content and communications, leading teams, transformation, business operations, and technology and engineering
• Every session built around real airline scenarios sourced from its AI champions, and run inside ChatGPT Enterprise so people worked in their own environment from the start
• A daily AI Concierge booth for walk-up, one-to-one help, plus a Teams channel for remote staff

The challenge

With six brands, thousands of employees and global operations, a global fashion retailer needed its leadership team to do more than experiment with AI on their own. They needed to set strategy, govern adoption and lead transformation at scale. At the start:

• Leaders were using AI individually, without a shared way to judge tools, use cases or risk
• It wasn't clear which tasks Copilot suited best, and where other models would serve better
• There was no AI champions structure, no model for governing agents and no shared language for what AI transformation means
• A major decision on enterprise AI licences was approaching, with no framework for making it

The solution

A three-session leadership accelerator designed for the retailer and run on site, with office hours, an executive check-in on strategy and monthly follow-ups:

• AI Leadership Foundations: a shared mental model for adoption, the assistant, thinker and creator modes of working with AI, and what makes this moment different
• Executive Advantage: from using AI to building with it, covering agent governance principles, a live agent build and leaders designing their own agents in pairs
• Reimagine Workshop: governance, an AI champions structure, a retailer-wide AI action plan and the path from capable individuals to organisational change