Services · retained
Fractional enterprise architecture for teams building an AI practice.
A retained relationship for organisations that need senior architectural judgement more often than they need a full-time hire — and want someone with no stake in the answer being complicated.
The expensive decisions are made early, quietly, by people with no time to research them.
Whether to build or buy. Whether a vendor’s platform is a genuine accelerator or a data-gravity trap. Where inference is allowed to run given your regulatory position. Whether the retrieval architecture being proposed will still work at ten times the corpus size. These decisions get made in a week, under delivery pressure, and are expensive to reverse eighteen months later.
Most teams do not need a permanent enterprise architect to answer them. They need access to someone who has made the decision before, is available when it arises, and has no commercial interest in the more elaborate option.
This is deliberately structured to avoid that conflict. If the right answer is a product you should buy, or work your own team should do, we say so — and lose the revenue.
What we build
What a retained engagement covers
Design review
Standing review of AI and data architecture decisions before they are committed, with the failure modes named specifically rather than generically.
Build, buy or wait
Honest assessment of vendor platforms against what you would be building — including the cases where waiting two quarters is the correct answer because the market is about to resolve the question for you.
Governance posture
How AI systems will be assessed, approved and evidenced inside your own governance process. Written by someone who has taken AI systems through an enterprise approval process rather than theorising about it.
Data residency and model placement
Where inference can legitimately run given your regulatory position, what changes if it must run inside your tenancy or on your own hardware, and what that costs in practice.
Standards and reference patterns
Reusable patterns for grounding, evaluation, cost control and provider abstraction, so each team is not solving the same problems independently and differently.
Team enablement
Working sessions with your engineers on the parts that are genuinely unfamiliar — evaluation design, prompt contracts, retrieval tuning — aimed at removing the dependency on us rather than deepening it.
What changes
What this changes
Fewer expensive reversals
The decisions that are costly to undo get an hour of experienced scrutiny at the point they are made, which is when scrutiny is cheap.
A consistent house style
Teams converge on shared patterns for grounding, evaluation and cost control instead of each inventing their own and discovering the same failure modes separately.
Governance stops being the blocker
When approval requirements are designed for from the start, review becomes a checkpoint rather than the place delivery goes to wait.
A path to not needing us
The explicit aim is that your team can make these calls without external help. Retainers that quietly optimise for their own renewal are not a service, they are a subscription.
Have a decision in front of you right now?
Bring it to a first conversation. It is a better test of whether this is useful than any description of the service.