I take AI from a blank agenda to running software.

The organisation had genuine AI ambition and no plan yet. I ran discovery across 17 stakeholders, mapped six delivery phases, and surfaced 27 business problems. Each candidate was scored through my investment-readiness framework: six made the shortlist, and two were scrapped at that stage, with the reasoning shared: one on economics, the other on data readiness once it was clear the vendor's own roadmap was heading at the same capability. Cutting them early is the gate doing its job.

I wrote the business cases for the four that remained (Playbook, HSE Risk Planning, Lookahead and Design Review), each with ethics and responsible use assessed, and built a working proof of concept on Microsoft AI Foundry and Databricks, because executives fund what they can see working. All four were approved and funded by the Australian Executive Committee, with an initial ~$300K PoC and pilot budget. The PoC, Project Delivery AI, is one assistant with two live agents, Playbook and HSE Risk Planning, now trialling on a live alliance project. Lookahead and Design Review are next on the funded roadmap, with production targeting 2027.

Teams were losing hours to manual, rules-based work. Everyone knew automation was possible; no one had determined which processes were worth the investment. I ran workshops across three business areas, applied the same filtering discipline I use for AI investments, and selected five: Model Federation and Indexing, Batch Drawing Stamps, Document Management, CADConform, and Report Generation.

I then directed two matrix-assigned developers to build them: Python with React front ends, built to fit the workflows teams already had. Three are live in production today: Document Management, Report Generation and Batch Drawing Stamps. The other two are built and on the roadmap for the rest of this year.

The $700K is built bottom-up: each automation saves a project an estimated $20–40K a year in manual hours: about $100K per project, roughly $700K across seven, at a company-standard loaded labour rate validated by our Technical Leaders. The build took ~1.5 weeks of the three of us and repaid itself within its first fortnight live.

Universities had the data (enrolments, HEIMS, Burning Glass labour-market signals) but no way to make decisions with it. The answer was an R Shiny analytics platform integrating all three sources: a multi-person build in which I managed a team of up to five consultants, acted as a senior developer, and ran the client workshops. I also built the component the sale hinged on: the AI layer that turned findings into a ready-to-read executive report, cutting turnaround from two weeks to five minutes.

The deliverable was productised into a standalone offering outside consulting: a new product line Nous Group now sells in its own right, beyond the three universities it launched with. The technology was deliberately unexotic; it sold because it fit how decisions get made.