How it works

Trained people, clear rules and work you can check. How a job starts, who checks what, and what you see every week.

Get a plan

A plan within 2 working days Two-week paid pilot A weekly report from week one

From request to production

No sales call. A plan by email, a short paid pilot on your real work, then the same team at full volume.

What we need from you

Your guidelines, or finished examples we can turn into guidelines
Access for our people in the tools the work happens in
One contact who answers questions during the pilot

5 minutesTell us the task
What the work is, how much there is, and how fast you need it.
Within 2 working daysGet a plan and a price
People, hours, start date and the total price, by email.
2 weeksRun the pilot
We train the team, build the gold set and work on real items, with two weekly reports. Then you decide.
After the pilotKeep going
The same team at full volume, with a weekly report. Add or remove people by telling your team lead.

Who checks what

One trained person does each item. Checks are layered on top, and the weekly report shows what they found.

Step	Who	Which items
Doing the work	One trained person	Every item
Team lead review	Your named team lead	Flagged and unclear items, and each person's error pattern
QA check	A QA reviewer	A sample of finished items
Gold check	Scored against the known-correct answer	Gold items mixed into live work

QA re-checks a sample, not every item, and the weekly report shows the results.

Training before live work: Each person trains on your guidelines and finished examples before touching live work. When your guidelines change, the team retrains, and the weekly report notes the date.
Known-correct answers: We build a gold set from your guidelines and examples: items with answers you approve. It measures accuracy instead of guessing at it, and you can keep it and use it with any team.
QA sampling: QA reviewers re-check a sample of finished items against your guidelines and the gold set. The report shows items done, items sampled and the acceptance rate on the sample.
Escalation, not guessing: Unclear items go to the team lead. If your guidelines answer it, the lead decides and adds the case to them. If not, the lead asks your contact, and the question appears in the weekly report.
Corrections: You can send back any item you think is wrong. Your team lead replies with a fix or a reason.

What you see every week

A report from your team lead: the hours, the output, what QA found, what changed because of it, and the questions only you can answer.

Weekly report, week 2 of 2AI output review pilot, 3 reviewers and 1 team lead

Example

118person-hours

2,360items done

236re-checked by QA10% sample

226accepted on re-check95.8%

What QA found (10 items)

Missed a required step6

Missed the refund policy3

Formatting1

All 10 corrected. The refund errors pointed to one topic, so 140 more refund answers were re-checked and 4 fixed.

Questions for you

Should answers that quote the old 60-day refund window count as wrong or partly correct?

Are Spanish-language chats in scope?

Next week: refund rule added to the guidelines, 20 new reference answers for refund questions.

Where the work happens

The team works where your items already are. Most work needs no new software.

In your own toolsYour helpdesk, CRM, store admin or labeling tool, on accounts you create and control.
In Deepen AI's toolsAnnotate, Calibrate, Curate and Validate, for labeling and sensor data.
In Ground Truth or A2IAs a private workforce in your SageMaker Ground Truth or Amazon A2I portal.
From a shared sheet or bucketA spreadsheet, folder or storage bucket you give the team access to.

How work reaches the team

Questions

Is every item checked twice?

No, not by default. One trained person does each item, QA reviewers re-check a sample, and gold items measure each person's accuracy.

How do you measure accuracy?

With gold items: known-correct answers you approve, mixed into live work without the team knowing which they are. The weekly report shows the acceptance rate on the QA sample.

What happens when our guidelines change?

Tell your team lead. The team retrains on the change, and the weekly report notes the date, so you know which work followed which rules.

What if we think an item is wrong?

Send it back. Your team lead replies with a fix or the reason, and if your guidelines need a change, they're updated for the whole team.

Who is our contact?

A named team lead, from the pilot onwards. They run the team, answer your questions and send the weekly report.

See it on your own work

Get a plan

Two-week paid pilot. Weekly reports from week one.