Step 02
Architecture and design
Every vendor has shown you the platform that solves everything, and your people already use AI nobody approved. Step 02 settles what runs where, under which rules, before you buy.
Step 02 is design and advice, a partner hands-on with your IT and data owners: models, hosting and access matched to how sensitive each piece of data is, and the rules every agent works under. Sometimes the honest design buys nothing new – it keeps the tools you already pay for and adds the rules they were missing.
When it applies
A use case is chosen, the model and platform are not, and your data must stay protected.
- An AI platform or model to choose
- A shadow-AI incident, or an AI policy to write
What you get
A vendor-independent target architecture matched to how sensitive your data is: frontier, enterprise or open-weights models, access and data boundaries, the rules for your agents and assistants, hosting in the region you need, no vendor lock-in, and a build plan.
Before anything is built, we tell you what the AI will cost to run – per task – so the business case rests on real numbers.
You decide where it runs – in your own environment or on a platform under your own account – under your security rules, and you approve the architecture before anything is built.
“We already gave everyone licenses.”
Then the tool is in place. The next step is one process redesigned end to end – that is where the license starts to pay.
The rules for your agents and assistants
Most companies already run more AI than they have written rules for – assistants, coding agents, integrations, and agents that act on their own. Step 02 settles, in writing, what each one may touch and who answers for it.
An inventory of the AI in use – every assistant, coding agent, AI agent and integration, sanctioned or not: who owns it, what it can read and do, and where its data and logs go.
The rules, written into your architecture – every agent gets an identity, a budget, rules and a record before it gets access: scoped, revocable, logged permissions; a spending cap with alerts; which actions it may take on its own, which need a person, and which models and vendors are allowed for which data; and where its prompts and outputs are kept.
A one-page rule for your people and an AI policy for your board – and, for your counsel, the legal duties that arise.
An independent view of the controls – which ones you already pay for, which ones you need, and what any product must prove in your own environment before you buy it; the commercial terms reviewed on your side.
A 90-day plan with named owners – what to turn on first, what to buy, what to hand to a certified security firm, and the one number the plan is measured against.
We design and advise; we do not test, monitor or run your security – that stays with your security team and certified specialists.
How it runs
A diagnostic of your use cases and data, with a named owner on your side; the scope and terms of every step agreed in writing before it starts.
A partner is accountable for every engagement.
Architecture before rollout. Where your data lives, who can access it and how it is secured are settled before anything goes live – and the design is written so you could have it built elsewhere.
What you decide
Step by step from prototype to everyday use – you decide after every step.
How to start
A 20-minute call about what prompted you to get in touch – no slides, one question. Write to us or directly to either partner; if AI is not the answer, or we are not the right fit, we say so.
Prefer e-mail? Write to letsdiscuss@raimconsulting.com or directly to either partner.
- letsdiscuss@raimconsulting.com
- Roman Antošroman.antos@raimconsulting.com
- Filip Gröschlfilip.groschl@raimconsulting.com