AI discovery · via GCD
QuayStreet Asset Management
AI discovery · via GCD
QuayStreet Asset Management
Which AI, in what sequence, inside which rules.
GCD's three-week AI discovery for a New Zealand active investment manager — with Wild embedded as the AI engineer alongside their strategy lead, shaping architecture, use cases and regulatory guardrails into a roadmap the board could fund.
Discovery engagement — embedded alongside the strategy lead.
Roadmap phases — Foundation, Growth, Intelligence.
Regulatory constraints every option was designed within.
Architecture decision points — each with one clear recommendation.
Tiers of AI capability mapped, from grounded answers to agentic workflows.
Vendor lock-in — the model provider is a configuration setting, not a dependency.
Not whether to use AI — which, in what order, within which boundaries.
QuayStreet is an active investment manager entering a growth phase — with an essentially greenfield marketing and communications stack, and AI ambitions that had to survive contact with financial-services regulation.
In a regulated environment, every AI output that reaches a client is a regulated communication. The discovery’s job was to make sure the content and data decisions being made now wouldn’t foreclose the AI capability the firm will want in twelve to eighteen months — and to lay out the options honestly, with a recommendation at every decision point.
The defensible asset is the knowledge corpus, the prompts, the guardrails and the evaluation suite — and the client owns all of them. Swap the provider in a day; the intelligence stays.
Every AI output reaching a client is a regulated communication under the FMC Act — designed to stay within class advice, consistent with current disclosures, always.
In the client's own words: extending investment rigour into a service experience. The AI has to meet the same bar as the funds.
Options laid out. Trade-offs named. One recommendation each.
The architecture & options paper walks every decision — approach, retrieval, model, orchestration — to a clear answer.
Foundation — content architecture and data structures that don't foreclose AI capability wanted 12–18 months out.
Growth — the marcomms stack matures on those foundations, with every content decision made AI-ready.
Intelligence — grounded, cited AI capability arrives on foundations built for it, not retrofitted around it.
The reason AI sits in discovery rather than at the end: content architecture and data structure decisions made in the Foundation phase directly determine what AI-assisted guidance and personalisation are possible later. Getting them right costs little during the build — retrofitting them is expensive.
A roadmap the board can fund
AI-relevant sections of the discovery roadmap and board pack co-authored with the strategy lead — reviewed, signed off, and grounded in what's actually feasible.
Every decision, one recommendation
An architecture & options paper covering the core AI approach, retrieval, model provider and orchestration — trade-offs laid out, a clear recommendation at every decision point.
Regulation designed in
FMC Act, licence conditions and the Privacy Act treated as design constraints from day one — including a supervised regulatory-sandbox pathway for future guidance features.
Freedom to change its mind
No vendor lock-in anywhere in the design — the model provider is a configuration setting, and the corpus, prompts and evals belong to the client.
The deliverable was the advice itself: architecture, sequence and rules the board could act on.