Exceptions and approvals: what your team still owns after go-live

If you own a 20–80 person shop, the first question about an AI worker is rarely "can it do the work?" It is "who checks it, and what do my people still decide?" This post answers that plainly.

The short answer

AI workers handle the repeatable work in a role: pull the data, apply the rules you set, update your systems. When something falls outside those rules, the worker escalates instead of guessing. Your team reviews the exceptions and signs the approvals. How much oversight time that takes depends on your workflows, and it is written into your SOW — not promised in a sales call.

What the AI worker does on its own

Repeatable steps you already do the same way every time (matching, entering, routing, updating records)
Rules you set during the build — thresholds, vendor lists, routing logic, approval limits
Logging what it did so your team can see the trail

What stays with your team

Exceptions: anything that breaks a rule or looks unusual. The worker flags it and stops; a person decides.
Approvals: payments, refunds, credits, and anything else you decide needs a human sign-off. You set the limits.
Judgment calls: customer relationships, disputes, policy changes, and anything you would not hand to a brand-new hire.
The rules themselves: when your business changes, your team changes the rules. The worker follows them.

How this is set before go-live

20-minute assessment. You and Jonas map the roles and workflows worth handing off, and where the judgment calls sit.
Build. Custom workers for your systems and edge cases, with demos so your team sees what gets escalated and why.
SOW. Your SOW lists the work, the access we use, the escalation path, and the agreed metric. We use only the access listed in your SOW, and only to deliver the services.
Go-live. Target go-live: about 30 days from kickoff. Your timeline is set in your SOW.

Questions to ask before you sign

Which steps will the worker complete without a person, and which will it escalate?
Who on my team receives exceptions, and how quickly do we need to respond?
What approval limits do we set, and how do we change them later?
Where can my team see what the worker did and why?
What happens when our rules or tools change?

What this is not

This is not a hands-off promise. Your team keeps the judgment calls. Any figures you see on our site are illustrative; your assessment estimates the numbers for your roles.

If the agreed Success Metric in your SOW is not met within 90 days of go-live, Leverwork pauses the fees and work listed in that SOW until the metric is met or the SOW end condition applies. The pause is not a warranty of savings or results.

You own your data, accounts, and tools. Client-specific customizations transfer only if assigned in a signed Schedule A to your SOW.

Next step

Book a 20-minute assessment: https://leverwork.com/book/ — we will map which roles fit and where your team's approvals should sit.