Quality and Call QA: Four-eye checks
Objective. Open Four-eye checks and read maker-checker as a targeted control that raises confidence on high-stakes scores - not as proof that all scoring is correct or as pure duplicated effort.
Four-eye checks route selected evaluations to a second reviewer before they stand. This screen answers 'which scores got a confirming second look, and were they upheld?' - a control for high-consequence items, not a blanket accuracy guarantee.
See it in the demo
What this screen answers
Maker-checker (four-eye) sends chosen evaluations to a second reviewer who confirms or corrects before the score is final. It answers 'on the scores that carry the most consequence, did a second set of eyes agree?' A CSM uses it to validate that high-stakes or auto-fail scores are controlled; Sales uses it to show the programme has a built-in guard against a single reviewer's error on the calls that matter most.
The task
Scope to the items routed for a second check, open one, and read the maker's score against the checker's confirmation or correction. The task is to confirm the control is applied where it should be - high-stakes, disputed or auto-fail interactions - and to see how often the checker overturns the maker, which is itself a calibration signal.
How to read it
A passed four-eye check raises confidence on that specific item; it does not prove the whole programme is accurate. The classic misread is either dismissing it as doubled effort, or over-claiming that four-eye coverage means all scores are verified. It applies to selected items, so it says nothing about the scores that were not double-checked. Read a high overturn rate not as reviewer incompetence but as a signal that makers and checkers are not calibrated - the same agreement question, surfaced on high-stakes items. The control defines confidence for the items it touches, scoped to exactly those items.
Honest limits
Four-eye checks cover a selected subset, so they bound confidence on those items only, not the aggregate. They also cannot fix a rubric that is wrong - two reviewers can agree on a bad standard. Confirm with product how maker-checker routing, thresholds and roles are configured and entitled; the demo shows the control on seeded data.
Interpretation check
A passed four-eye check raises confidence on that specific high-stakes item only - it does not prove the scores that were never double-checked, and a high overturn rate is a calibration signal, not reviewer incompetence.
In the field
KnackGlobal (RCM): routing auto-fail and compliance-flag scores to a second reviewer means a costly mark is confirmed before it stands - confidence where it matters, without claiming every score was double-checked.
Practical task
Open the Four-eye checks screen, pick one checked item, and write it up: the maker's score, whether the checker upheld or changed it, and why - then state which categories of interaction you would route to four-eye for a compliance-sensitive customer.
Self-check
Does four-eye coverage mean all scores are verified?
No. It covers a selected subset - typically high-stakes, disputed or auto-fail items - so it raises confidence on those, not on the scores that were never double-checked.
Checkers overturn makers often. What is the diagnosis?
Usually a calibration gap between makers and checkers, not incompetence. It is the same agreement question surfaced on high-stakes items; feed it back to calibration and rubric clarity.
Which interactions most deserve a four-eye check?
High-consequence ones - auto-fails, compliance findings and disputed scores - where a single reviewer error would be costly.
Can a four-eye check fix a wrong rubric?
No. Two reviewers can agree on a bad standard. The control guards against individual reviewer error, not against the rubric measuring the wrong thing.
Verified vs demo. Screen is live in the demo at /#/review/maker-checker; exact controls and figures are SME-validation-pending.