Insights · · 3 min read
How do QA checks and approvals work?
The short answer
Every task passes QA checks against your rules (prices, facts, scope and tone) before a person sees it. Work that matters then waits for a person's approval. Corrections are saved as lessons, so later tasks start from them.
Why checks come first
An AI agent can draft a quote quickly. The real question is whether you can trust it enough to send. Checks answer that question every time, so you don't have to reread every word yourself.
In AI operations, quality comes from four layers that work together.
1. QA checks against your rules
A QA check is a written rule the work must pass. Good checks are specific and easy to test:
- Every price matches the current price list.
- The scope matches what the client asked for.
- Names, dates and numbers match the source documents.
- The tone follows your writing guide.
- Nothing is offered that your policies do not allow.
Some checks are simple comparisons. Others need a second agent, a checker, that reads the work against the rules and is separate from the agent that wrote it. Work that does not pass goes back for another draft with the reason attached, before anyone sees it.
2. Your approval where it matters
Not every task needs a person. Agreed internal work, such as a summary for your own team, can go through once it passes its checks.
Work that leaves the business or costs money waits for an OK: a quote to a client, a refund, a page that goes live. Where those approval points sit is decided routine by routine, with the people who own the work.
3. A person's review
Checks cover what rules can describe. A person covers what rules can't: a client who needs a softer tone this month, or a job that looks right on paper but is wrong for the relationship. A short review by someone who knows the business keeps that judgment in the loop.
4. Lessons, so the next task starts better
When a reviewer changes something, the change is saved as a lesson with its reason, for example "This client wants prices with tax included." The next time any agent works for that client, it reads the lesson first.
Over the weeks, these lessons add up to the business's own playbook, written down and shared by every agent.
- DraftQuote for a returning client
- QA checksPrices, scope and tone passed
- Your OKApproved by the account lead
- SentQuote sent to the client
- Lesson savedThis client wants prices with tax included
What to ask before you start
When you talk to anyone about AI in your business, ask to see three things for each routine:
- The checks the work must pass, written in plain words.
- Where the approval points sit, and who approves.
- How corrections are saved and used next time.
Like any tool, AI isn't perfect. A good setup is built so that nothing important goes out without a check, and so you can see which checks each task passed and who approved it.
More questions about checks and approvals.
Do checks slow the work down?
A little, and on purpose. Rule checks run as part of the task. The approval step takes as long as your review, which is why approvals sit only where they matter.
Who writes the QA checks?
They come from your own rules: price lists, policies, writing guides and what your team already checks by hand. We write them with you when each routine is set up.
Can approval points change over time?
Yes. In a well-built setup, approval points are set per routine and can move as your team sees how the work goes.