Operating model

The access ladder: how we control what AI can do.

Every production AI system follows the same graduated model. Start with zero access. Prove each level before graduating. None of our bots operate beyond Level 3 — and that's the point.

LevelAccessWhat it meansProduction example
Level 0No business dataGeneral drafting, brainstorming, public information only. No system access.Staff using approved AI tools with de-identified examples only
Level 1Read-onlyCan search approved documents and read from designated systems. Cannot create, send, update or delete.MarketplaceBot scanning 450 Pipedrive deals every morning — no mutations allowed without explicit approval
Level 2Draft-onlyCan prepare drafts, summaries, classifications and recommendations. A human must review and approve before any action.FinanceBot preparing monthly reimbursement reminder for the finance coordinator's approval before sending to office@
Level 3Write after approvalCan take approved actions after human sign-off. Logs every action. Rollback capability required.FinanceBot forwarding approved-sender invoices to Xero Files after GST and entity checks
Level 4Limited autonomyCan take predefined actions within strict boundaries. Only after months of proven operation, monitoring and explicit approval.None of the production bots operate at this level. The bar is deliberately high.
Level 5Critical system controlPayments, payroll, legal decisions, autonomous customer communications, regulated advice. Requires specialist review and strict governance.AiBorz does not recommend this level as an early SME project — and rarely recommends it at all.

The gap at Level 4 is intentional. Every AiBorz system starts at Level 0 or 1 and graduates only when the workflow has proven itself with real operational data, human oversight and documented controls. Autonomous action is not the goal — controlled, useful AI is the goal.

In practice

What a typical day looks like.

8:10amMarketplaceBot

Scans 450 Pipedrive deals. Surfaces activity exceptions. Emails the leadership team before stand-ups.

8:30amContent Bot

Delivers daily intelligence briefing to Alborz via Telegram. AI, automotive, investment — bullets only.

Every 10 minAsh + MarketplaceBot

Monitor inboxes. Classify. Route. Draft replies. Escalate sensitive items. Never auto-send to external parties.

Throughout the dayFinanceBot

Processes approved-sender invoices to Xero Files. Answers authorised finance queries with live data. Flags missing receipts.

5:00pmMarketplaceBot

Daily learning report. What the inbox taught the system today. No secrets exposed. No state mutated.

MonthlyAll systems

Governance review. Access audit. Prompt updates. Staff feedback. Output quality sampling. Improvement planning.

What this means for your business

Most AI projects fail because they skip levels. They go from "let's try ChatGPT" to "let it send emails to customers" without proving each step. The access ladder prevents that.

When AiBorz builds a system for your business, the first version is always Level 0 or 1: read data, draft responses, flag exceptions. It never starts at Level 4 — and your team decides when (and if) it ever reaches Level 3. The ladder is not a suggestion. It is the operating model.

See how this works inside real production systems: browse the case studies. For the launch protocol that governs every new workflow, see the controlled launch process.

Start with a readiness check.

Run the free AI Readiness Scorecard to identify your best first workflow — with the right access level from day one.

Start the AI Readiness Scorecard