Partial or Full Augmentation? The Four Levels of AI Employees

Key takeaways

• AI in your team isn’t a choice between a chatbot and a robot replacement. It’s a spectrum with four practical levels.
• Level 1 assists, level 2 owns the routine share of a workflow, level 3 runs most of a role’s admin, and level 4 runs a workflow on its own.
• Autonomy should be earned one task at a time, based on measured results, not granted all at once.
• We recommend most first AI employees start at level 2, where the return shows up quickly and people stay in control.

When operations leaders picture AI in their team, they usually picture one of two extremes: a chatbot that doesn’t do much, or a robot that replaces people. Neither is useful, so the project stalls.

In reality, AI staff augmentation works on a spectrum. Here are the four levels we use, what each one looks like day to day, and how to decide where each workflow should start.

Why a levels model helps

Most AI projects fail for one of two reasons. Some aim too low: a tool that drafts an email nobody sends, so nothing changes. Others aim too high: an autonomous system nobody trusts, so it never goes live. A levels model gives you a middle path. You agree up front what the AI employee is allowed to do on its own, what it has to ask about, and what it must never touch.

It also gives everyone a shared language. When a manager asks “is the AI sending emails to clients?”, the answer isn’t yes or no. It’s “routine reminders run at level 2, anything about pricing stays at level 1”.

Level 1: Assist

The AI employee reads the work, pulls the right records and drafts the response. A person reviews and sends everything. It’s the lowest-risk way to start and it builds trust fast, because your team sees every output.

Example: drafting renewal information requests for a broker to approve and send.

What changes: less time writing and looking things up. The person still clicks send on everything, so the time saving is real but limited.

Controls: every output is reviewed. Track how often drafts are sent unchanged, edited or rejected. That number tells you when a task is ready for level 2.

Level 2: Partial augmentation

The AI employee owns the routine share of a workflow end to end. It sends the standard follow-ups, collects the documents and updates the records. People handle the exceptions, the judgement calls and the relationships.

Example: chasing contractors for quotes and invoices on every open work order, while the manager handles disputes.

What changes: the routine loop stops landing on anyone’s desk. People only see the items that need them, with the history already attached.

Controls: clear rules for what counts as routine, a daily summary of what the AI employee did, and an escalation path for anything unusual.

Level 3: Role augmentation

The AI employee runs most of a role’s admin across several systems. A person supervises, coaches the agent and manages the escalations. This is where capacity changes: a small team can handle the volume of a much bigger one.

Example: one coordinator plus an AI employee running onboarding for every new client.

What changes: the person’s job shifts from doing the admin to managing the outcome. They spend their time on quality, exceptions and improving the process.

Controls: weekly review of volume, turnaround and error rate, and a simple way for the supervisor to correct the AI employee so the same issue doesn’t come back.

Level 4: Full augmentation

The AI employee runs a defined workflow autonomously, with an audit trail and a human override. A person owns the outcome and reviews the numbers each month.

Example: lead follow-up that replies to every new enquiry within the hour, chases non-responders on schedule and hands every reply to sales.

What changes: the workflow runs around the clock without anyone managing it day to day.

Controls: a full audit log, monthly performance reviews, alerts when numbers drift, and the ability to drop any task back to a lower level instantly.

The four levels at a glance

Level	The AI employee	The person	Best for
1. Assist	Drafts and prepares	Reviews and sends everything	Client-facing work, first weeks of a pilot
2. Partial	Owns the routine share	Handles exceptions and judgement	Most first AI employees
3. Role	Runs most of a role’s admin	Supervises and coaches	High-volume teams under pressure
4. Full	Runs the workflow on its own	Owns the outcome and the numbers	Proven, well-defined workflows

One workflow, several levels at once

Levels apply to tasks, not whole workflows. Take a typical client onboarding workflow:

Task	Sensible starting level
Send the welcome pack and document checklist	Level 2
Remind the client about missing documents	Level 2
Check documents are complete and file them	Level 2
Answer a question about fees or terms	Level 1
Approve an exception to the standard process	Stays with a person

The routine steps move quickly. The sensitive ones stay supervised. Over time, individual tasks move up as the numbers prove them, and the ones that should always sit with a person stay there.

How to choose the right level

• Start lower for anything client-facing. Level 1 or 2 until the error rate proves itself.
• Keep advice, pricing and contracts with people. Always.
• Move up one task at a time, not one workflow at a time. Autonomy is earned per action.
• Let the numbers decide. Hours saved, volume handled and error rate tell you when a task is ready to move up.
• Consider the cost of a mistake. A slightly awkward reminder is easy to fix. A wrong figure in a contract isn’t.

Common mistakes

• Starting at level 4. It feels ambitious, but nobody trusts it, so it never leaves the pilot.
• Staying at level 1 forever. If people still send every message, you’ve bought a drafting tool, not an employee.
• No escalation path. The AI employee needs to know exactly who to hand things to, and how.
• Not measuring the baseline. Without the before numbers, you can’t prove the after. Our follow-up cost guide shows how to capture them.

Why we recommend level 2 for most first AI employees

We recommend most first AI employees start at level 2, because that’s where the return shows up quickly without asking anyone to trust a black box. The routine work leaves your team’s desks from week one, and people stay in charge of everything that needs judgement.

Our six-step method is built around moving tasks up the levels as the pilot proves them. For a real-world example, see how an AI employee runs the insurance renewal chase at level 2.

Frequently asked questions

Can a task move back down a level?

Yes, and it should be easy. If the error rate rises or the process changes, drop the task back to level 1 until it’s stable again.

How long does it take to move up a level?

It depends on volume and risk. High-volume, low-risk tasks can prove themselves within a few weeks of a pilot. Sensitive tasks take longer, and some should never move.

Who decides when a task moves up?

The person who owns the outcome, based on the numbers. It should be a business decision, not a technical one.