QuantAI · Fractional AI Operations

One governed operating model, three under-used tools, engineers still in charge

A 52-person Melbourne engineering consultancy had bought three capable AI platforms but adoption was low and fragmented, and its technical staff were rightly wary of professional liability. This is the firm-wide operating model that fixes it, with a signature "which AI for which task" framework and an engineering-grade assurance line, proven on three workflows end-to-end.
Operating model & governance Tool-selection framework Engineering assurance line Claude Projects & prompts Adoption & AI Champions
Vertex Engineering · Prepared by Dave Richardson, QuantAI

Vertex had done the hard part most firms avoid: it had invested in capable tools (Microsoft Copilot, Claude Team, ChatGPT Enterprise). The problem was the opposite of scarcity, three overlapping platforms with no decision rules, no shared workflows, no governance, and no way to connect licence spend to benefit. Adoption was bimodal: a few enthusiasts, many cautious avoiders. The task was not more technology; it was an operating model, and, because this is engineering, one built so that a qualified professional owns and signs every technical output.

Scenario results across three pilots

Modelled under stated assumptions. The brief's bar is at least a 15% cycle-time improvement per pilot without unacceptable quality loss. All three clear it.

27%
Proposals & bids ↓
30%
Technical report drafting ↓
33%
Project status reporting ↓
~1,410
Hours / yr (scenario)

About AUD $134,000 of professional time at a blended rate, roughly 3.8× the engagement fee on the modelled assumptions. The percentages depend only on per-task time (directly measurable in the pilot), so they hold even if the volumes are optimistic; the dollar figure sizes the prize, it is not the headline. On technical reports, none of the saving comes from, or is taken from, the engineering work itself.

The signature deliverable: which AI for which task

The piece Vertex was missing most was not another tool, it was a rule for the three it already owns. A one-line rule of thumb plus a ten-second decision path turns three overlapping licences into a deliberate toolkit: Copilot for work inside your Microsoft files and meetings, Claude for careful long-form drafting and review, ChatGPT for open-ended thinking, with a confidentiality rule that overrides every choice. Three tools · one deliberate toolkit

The three workflows

Proposals and bids

Highest value. A consistent, on-template draft from a structured bid brief and the firm's reuse library. The author owns the win strategy and every claim; the firm's best material works for every bid.

Technical report drafting and QA

Highest stakes. AI drafts structure and prose around fixed, human-owned technical content and never touches the numbers. The engineer confirms and signs. This is the pilot that earns the technical staff's trust.

Project status reporting

Highest priority, lowest risk. A clean, consistent update from project data in-tenant. The PM owns the honest RAG status. The fast, safe, visible first win.

The line that makes it safe to adopt

Engineering design, calculation, verification and certification are where AI does the most harm in an engineering context, so they are excluded by design, on every platform. The AI drafts and documents; it never performs, checks or certifies engineering work, and it never states that anything meets or complies with a code or standard. A qualified, registered engineer owns every technical output, and reviews and signs everything a client sees. AI assists · a qualified professional decides and signs

Why this wins the room. Framed well, the assurance line is the reason the most cautious engineers say yes. It tells them, in writing, that the firm is not asking them to trust a tool with their signature. The governance here is built to be shown, not bolted on.

Designed for adoption, not just capability

Because adoption was the real gap, the model is built to stick: a role-family capability matrix that says what "good AI use" means for each kind of role, an AI Champions network that spreads it, a light governance forum that steers without slowing work, and an executive dashboard on four honest measures (adoption, cycle-time, licence value, assurance health). The 90-day plan sequences a safe quick win before the high-stakes pilot, and resources the Champions explicitly.

What was delivered