Most people who have used ChatGPT, Claude, or Gemini in earnest have had the same small jolt: a task that used to take an afternoon now takes a couple of minutes. A rough set of notes becomes a clean first-draft proposal. A sixty-page report collapses into the five points that actually matter. A contract clause, or a stubborn spreadsheet formula, gets explained in plain language by something endlessly patient.
This is the productivity story of the decade. General-purpose AI — the kind we described as Track 1 in an earlier perspective — is quietly reshaping a wide range of information-centred work, and it is doing so for almost everyone, almost at once.
The interesting question is no longer whether AI is useful. It is how much of your world it can actually see. And the answer has been getting better in steps, each one removing something the AI previously could not know.
First barrier: it knows history, but not today
The earliest users hit this within minutes. Ask an AI to list the leaders of every country across the last century and it answers without hesitation. Ask it who holds a particular office today and it may answer with equal confidence — and be wrong.
The reason is simple once you see it. A language model is trained on an enormous but fixed snapshot of text, captured up to a certain date. It has read vast amounts about the past, so the past is where it is strong. But it has no live connection to the present; its knowledge stops at the snapshot. Worse, it does not know what it does not know, so a stale answer arrives sounding just as assured as a correct one.
The fix arrived quickly. Modern assistants now pair the trained model with live web search. When you ask about something current, the assistant goes and reads today's pages before it answers, topping up the frozen snapshot with fresh fact. The recency barrier has been largely removed.
Second barrier: it knows your industry, but not your business
The next barrier is harder, and it does not fall to a web search.
An AI can explain in great detail how your industry works — the way the sector is structured, who the major players are, the business models underneath it. It has read the public record thoroughly. What it has never seen is anything inside your walls: this month's Board pack, the minutes of yesterday's operations meeting, a single row of your ERP.
And consumer AI is all-or-nothing. If something is in the public record, you have full access to it. If it is private to your organisation, you have none. There is no middle setting — no way for the model to know a little about your business. It knows the world, and it knows nothing about you.
The barrier starts to fall: AI that reads your documents
This is where the latest wave of tools begins to change things. Microsoft Copilot, Google Gemini, and Claude can now reach into your own document repositories — the shared drives, the mailboxes, the intranet, and in some cases a handful of business systems.
Ask a question and the assistant can read your actual material before it answers. Summarise the three contracts in this folder. Tell me what we agreed in last week's minutes. Draft a reply that is consistent with our policy document. The AI is now reasoning over your content, not just the world's. For knowledge that lives in documents, this is a genuine step forward.
But documents are the easy half. They are unstructured prose, and the permissions attached to them are blunt: you can open the file, or you cannot. The hard half is your operational data — the live, structured truth in your ERP and line-of-business systems. Who owes what. Which orders are late. What this customer's margin really is. That data is governed by fine-grained rules about who may see which slice of it, and a document assistant cannot safely reach into it, because it has no understanding of those rules.
Teaching AI your business — properly
This is the step Predictiv is built for, and it is the one that justifies the title. Teaching AI your business means three things working together.
A map of what your data means. We maintain catalogues that describe your operations in terms the AI can use — the tables, the relationships, the business rules, the vocabulary your organisation actually speaks. Rather than leaving the AI to guess at a wall of cryptic columns, we hand it a map. This is the heart of teaching AI your business: it is given an understanding of your operations, not just access to them.
A live connection to the systems of record. The AI is connected to the operational databases themselves, so answers reflect the current state of the business rather than a stale export someone took last month.
Your own rules about who may know what. Every question is answered through the same role-based access control that already governs the person asking it. The AI sees exactly what that user is entitled to see, and nothing more. The organisation's existing rights model is honoured, not bypassed. This is the crucial difference from consumer AI: it is no longer all-or-nothing. It is precisely your slice — shared fully, within the boundaries your organisation has already drawn.
Two things sit underneath all of this and matter as much as the capability itself.
Everything is audited. Each access is recorded: who asked, what was retrieved, and under whose permissions. An AI answer drawn from your operational data is as accountable as any other action in the system — no quiet side channel, no unlogged reach into the business.
The system learns your business over time. It notices which questions get asked, which answers prove useful, and where its map of your operations is still thin, and feeds that back to enrich the catalogues and sharpen future answers. The organisation's AI gets steadily better at the organisation's own work — and because every interaction stays inside the audit trail and the permission model, it does so without ever stepping outside who is allowed to know what.
Where this leaves you
General AI gave everyone a brilliant assistant that knows the world and nothing about you. Web search taught it today. Document tools taught it your filing cabinet. The step that actually changes how an organisation runs is the last one: an AI that understands your live operations and answers from them — fully, accountably, and strictly within the rights your organisation already enforces.
That is what it means to teach AI your business. If your team has felt the productivity jolt but run into the wall where the AI stops knowing anything about you, that wall is exactly what we built the Predictiv platform to take down.