Surpoint · The path of AI inside the company
In most companies AI arrived before the plan: teams already use it on their own, leadership wants results, and in between there is disorganized data and systems that do not talk to each other. This map shows in what order it gets solved, where each company usually stands today and what comes next.
The most common trap is asking for step 4 results with step 1 data and systems. AI ambition and system readiness tend to be treated as separate projects, when in reality one depends on the other.
Usage policy · Real cost · Which data enters which tool · Compliance (Ley 21.719, in force 1 Dec 2026) · A live inventory of what people build · Measurement of returns. It is not a step: it is born in the assessment and runs alongside the whole path.
The door
“Where do we stand, and where do we start?”
Why it starts here: diagnosing before buying avoids paying for licenses, courses or projects that do not match your level.
How the ones who adopted well buy: scoped, specific pilots; it works, it scales. Never the other way around.
Use
“Are my people using this, and under what rules?”
Golden rule: the official alternative goes live the same day as the restriction. Training without a governed lane accelerates outside usage, it does not reduce it.
Market data: only a third of purchased corporate AI licenses is used regularly (Recon Analytics, 2026). Adoption does not come with the license.
Connect + Sustain
“Can this answer about MY business? And can my systems take it?”
Data for AI
Source inventory, semantic model, permissions and traceability. An agent is worth what the data it can read is worth.
Modernization (ARISE + SAP)
A legacy system with no API and no documentation connects to nothing. Document Z, prepare S/4HANA, migrate what blocks.
The key idea: modernization does not compete with AI for budget. It is the prerequisite of the AI that was already approved.
Market data: 1 in 2 generative AI projects drags along prior data work (Accenture, FY25 results). This is the step almost nobody budgets.
Automate
“Can it do work, or only talk?”
How it starts: a scoped pilot, one data domain, one management user. Results measured in cost avoided, the currency a committee understands without translation.
Gartner estimates more than 40% of agent projects will be cancelled before 2028. Hence a pilot, not a promise.
Scale
“It worked. How do we take it to the whole organization?”
Closing the loop: cases with proven returns become the budget for the next round. It gets funded with evidence, not with enthusiasm.
How the ones ahead do it: LATAM Airlines reorganized IT to enable (management, cybersecurity, citizen developers, support) and rewards with more capabilities whoever builds something that serves others.
The arrow back: almost every company enters through what is urgent (organizing usage, training) and discovers later that what held results back was at the bottom: the data and the systems. Seeing the full map from day one makes it possible to budget that stage in time.
02
The assessment places the company at one of these five levels. Each level says which step of the journey is theirs. It is the opening question of every meeting: “where would you say you are today?”
Level 1
People use AI with personal accounts. IT does not know what is used or with what data.
→ Steps 1 and 2
Level 2
There is policy and there are licenses, but usage is individual and changes no process.
→ Step 3 (data)
Level 3
AI answers about the business, not about the internet. Still a query.
→ Step 4 (agents)
Level 4
Agents run complete processes with an auditable record.
→ Step 5 (scale)
Level 5
It gets measured, prioritized, and initiatives are funded with evidence.
→ Sustain the loop
03
Governance is not the last step: it is the layer that runs alongside everything.
Policy, licensing and compliance cannot wait until there are agents running: they are born with the diagnosis. Banning without offering an official alternative does not reduce usage, it makes it invisible. What is left for the end is something else: scaling what worked.
Between adoption and agents there is a step almost nobody budgets: the data.
A generic assistant answers about the internet; for it to answer about your business, the information has to be identified, organized and with clear permissions. Half of generative AI projects drag that work along without having planned it.
If leadership asks for level 4 results and the company is at level 1, no tool shortens the path: two intermediate stages are missing, the data and the systems, invisible from the top. Seeing them in time stops readiness from competing with AI for budget.
Where would you say your company is today? The AI-Ready Assessment answers that question in 10 days, with fixed scope and fixed price.
