AI Architecture Review
A specialist follow-on for one or two workflows where model cost, data sensitivity or infrastructure ownership deserves deeper analysis. The review compares closed APIs, hybrid, private and self-run paths, including whether the workflow should use AI at all.
- Scope: 1–2 workflows
- Timeline: 2–3 weeks
- Output: written decision report
- Includes: savings model + infrastructure path
What sits in the report
Seven sections. One written report. The go / no-go recommendation is named in one sentence on page one, with the savings and infrastructure logic behind it.
Workflow map
Every manual, software, and AI step in the workflow you scope. Inputs, outputs, hand-offs, where humans still belong.
AI usefulness review
Where AI actually moves the needle, and where a rule, template, or removed step would do the job for less cost and risk.
Data sensitivity assessment
What data flows through each step, what can leave your perimeter, what cannot. Plain language for legal and security review.
Closed API vs open-weight vs private deployment comparison
Side-by-side on your workflow across API, hybrid, private cloud, and self-run paths. Assumptions written down so you can challenge them.
Cost, risk, and speed model
Fully loaded cost, residual risk, latency, and break-even point per option at your stated volume.
Go / no-go recommendation
One sentence on page one: keep on API, move private, simplify, or stop using AI for it. A written argument behind it.
Practical infrastructure path
If the recommendation is to change architecture, the report names the first three concrete steps toward a private or self-run setup — not a slide deck.
This is not implementation
The Architecture Review itself does not include a build. It tells you whether implementation is worth doing, what architecture fits, what savings are realistic and what trade-offs to expect.
Implementation can be scoped separately after the recommendation. The review is not tied to any infrastructure provider or required implementation engagement.
This is not the right fit if
Use this filter before going deeper — it saves both of us time.
- You do not have a specific workflow in mind.
- You only want general AI brainstorming.
- You want someone to build a chatbot immediately.
- You are not ready to share basic workflow and data context.
- You need a full compliance / legal assessment (we touch data sensitivity, not formal DPIA sign-off).
Four steps from first-pass report to Architecture Review
First-pass report
You describe the workflow, current tools, and what is prompting the review. The instant report shows whether the workflow has enough signal for deeper analysis.
Scope confirmation
Plain language: scope, inputs, timeline, and next step. You decide whether the deeper review makes sense.
Architecture Review (2–3 weeks)
We map the workflow, benchmark on your data where applicable, and model cost / risk / speed across the architecture options.
Written report + walkthrough
Recommendation on page one. Walkthrough call. The report is yours; we do not gate-keep conclusions.
How we scope the Architecture Review
- Scope
- One workflow reviewed end-to-end. Two if they are tightly related. Broader scopes need a separate scope confirmation.
- Duration
- Two to three weeks from kick-off.
- Output
- Written decision report + walkthrough call. You own the report.
- Next step
- After the first-pass report, we confirm the scope, inputs, timeline, and next step before the review begins.
- Confidentiality
- Mutual NDA available on request before kick-off.
- Data handling
- See the Privacy Policy. We do not retain your data beyond the engagement.
Common questions before a deeper review
Is private deployment cheaper than closed APIs?
It depends on volume, latency targets, and ops capacity. For many low-volume workflows, closed APIs are cheaper fully loaded. We model your specific case, including the break-even point for private or self-run infrastructure, so you decide on numbers, not slogans.
Do you always recommend moving off closed APIs?
No. A review can conclude that changing architecture is not justified. The written recommendation can be to keep the API, move private, simplify, or stop using AI for the workflow.
What if AI is not the right tool for our workflow at all?
We will say so. “Simplify” covers replacing AI with a rule, template, or removed step where that is the honest answer. We do not stretch a problem to fit a model.
Do you build the system during the review?
Not during the review. The review tells you whether implementation is worth doing, what architecture fits, what savings are realistic, and what trade-offs to expect. Implementation can be scoped separately afterwards.
Can we sign an NDA before sharing details?
Yes. Mutual NDA on request before kick-off; we can sign earlier if counsel requires it before you share sensitive architecture.
Do you work outside the EU?
No. Engagements are limited to EU and EEA organisations, with processing in that footprint.
Why one or two workflows, not more?
Depth beats breadth. One workflow reviewed honestly is more useful than ten skimmed. Two tightly related workflows can fit a single scope; broader scopes need separate scope confirmation.
Have an architecture question behind the workflow?
Start with the process. If deeper cost, risk or infrastructure analysis is justified, an Architecture Review can follow.