AI Adoption & Governance

We organize, govern and scale AI use across the company: usage policies, licensing, agents connected to business data, and team training.

Partner of Anthropic, Microsoft and SAP. Our recommendation starts from what your company already uses, not from a catalog of our own.

In most companies AI arrived before the strategy: personal accounts, tools built outside IT and licenses with no policy behind them. The board pushes to move forward and IT answers for keeping it under control.

The challenge isn't to stop that use, it's to govern it: letting AI move forward with clear rules, known costs and protected data. Sorting it out before handing out licenses is cheaper than doing it with a fire already burning.

Five areas of work

Licensing: what to buy, for whom, and at what cost

Copilot, ChatGPT, Claude, Gemini. Every tier charges differently: flat per user, API consumption or a mix. The question almost no one answers alone is what it will really cost at month end and what control you gain or lose.

We help you decide with judgment, not with the vendor's calculator:

A realistic cost estimate per usage scenario and number of users, including API consumption, which is where the surprises show up
What level of control each plan brings: auditing, usage limits, blocking of risky features, centralized management
Assessment of Microsoft 365 renewals under the terms in effect since July 2026
No commission from any vendor: we work across all three ecosystems

The usual approach is to align the tool with the suite already in use (Microsoft 365 → Copilot; Google → Gemini), add an advanced option for power users and keep the rest on official accounts.

Governing AI use

When teams adopt AI on their own, the company accumulates tools and automations that are not inventoried: that is shadow AI. The risk rises when someone requests ERP access so their tool works, and control becomes nominal.

Banning it by decree only pushes usage into less visible channels. The point is for IT to stop building everything and start enabling with rules.

Training without governance accelerates shadow AI: after every workshop people build more on the outside. That is why training and the official lane ship together.

We work this front in four layers:

Inventory of tools, accounts and agents in use, official and unofficial
Usage policies: what data can go to which tools, with which accounts and approved by whom
Enablement model: champions per area, a testing sandbox, and a path so that what people build goes through IT instead of around it
Compliance: Law 21.719 is in force from 1 December 2026, creates the Data Protection Agency and contemplates fines of up to 20,000 UTM. The Cybersecurity Framework Law and sector regulation apply on top

Praxia, the platform

Sustained in Praxia: permanent training with an AI tutor on your own material, a living inventory and the adoption dashboard per team.

Agents over your business data

The reports already exist: Power BI, SAP, spreadsheets, in-house systems. What doesn't exist is a direct way to ask them something.

Agents connected to the systems the company already has. A manager asks how much was produced this month at plant X and gets the figure, its source and the detail, without waiting for an analyst to build the spreadsheet.

Connected to existing data models; they don't replace reporting, they make it queryable
Access profiles: not every user can ask everything
They run on the client's infrastructure, with cost control per consumption

The usual starting point is a scoped pilot with one management user and one data domain, in weeks.

Process automation with agents

There are processes RPA never handled: different formats, criteria that require interpretation, exceptions. Where an RPA stops at a field that moved, an agent reads the document and decides.

Typical cases:

Invoice, purchase order and goods receipt matching, with a justified verdict per document
Entering purchase orders from email straight into the ERP, with its approval flow
Reporting that today is assembled by hand consolidating spreadsheets
Validation of documents and field evidence (photos, forms, records)

Every automation ships with an auditable log: what the agent decided and why.

Team training

Programs built on the real work of each area, not generic prompt courses, with separate modules for users and technical teams. Format, duration and cohort size come out of the diagnosis.

Training is what makes licenses actually get used: without it, a significant share of what you pay for sits idle.

Where to start

AI-Ready Assessment

Most AI initiatives don't fail because of the model, but because of the foundation they're built on. Before investing it is worth knowing what you are building on.

The Assessment places your company on the journey map and delivers the prioritized plan. It is a 10-business-day diagnosis across five dimensions:

Dimension

What it assesses

Connectivity

Which systems can feed AI, and how

Data

Quality, access and protection of information

Documentation

What is known, and what is not, about current systems

Security

Leakage risks, access, and regulatory compliance

Adoption

Who uses AI today, with which tools, under which rules

Deliverables: a score per dimension, an inventory of existing use (including the unofficial kind), a plan prioritized by return and a budget for the next phase.

Fixed scope and price. No hourly billing.

How we work

Everything runs in a private Azure cloud or in the client's tenant: data never leaves your infrastructure and never trains third-party models. Every implementation goes through human validation before touching production.

Partner of Anthropic, Microsoft and SAP.

Recent work

Microsoft Copilot adoption program · ~70 users · Manuka

AI training program · 150 people across 5 cohorts, with an additional technical module · Financial services company

Conversational agent over production data for executive leadership · Fishing industry company

Frequently asked questions

We found you through the Anthropic partnership. Do you only work with Claude?

No. We're a partner of Anthropic, Microsoft and SAP. Our recommendation starts from the suite your company already uses, not ours.

Is our data exposed to public models?

No. It runs in a private cloud or on your own infrastructure, with an NDA when the project calls for it. Data never trains models.

How do I calculate what it will really cost me?

It is one of the areas where we help. Generic estimators give ranges so wide they're useless for deciding. We model cost by real usage scenario, API consumption included, so the figure holds up in front of finance.

Do we need to have licenses in place before starting?

No, and it's usually better not to. Part of the diagnosis is determining what to buy and how many licenses are justified by expected real usage.

What happens to what teams have already built on their own?

It doesn't get blocked upfront. It gets inventoried, its risk assessed, and whatever adds value is channeled with rules.

Where do we start?

With the AI-Ready Assessment: 10 days, fixed scope and price. With the diagnosis in hand, every initiative that follows has a priority, a budget and an owner.

Start with a diagnosis, not a purchase

AI-Ready Assessment: 10 business days, five dimensions, fixed scope and price.