Analytics
How much of your volume settles without a person, what stops the rest, and how your reviewers are doing
Analytics answers one question in several ways: how much work is your integration creating for humans, and why. It is aggregate only. No applicant is identifiable anywhere on this page, which is why an owner can open it to a read-only seat without disclosing anybody.
Pick a reporting period at the top. Everything below follows it.
Automated
The headline figure is the share of decided verifications that settled without a person, split three ways: our engine, your own automatic decision policy, and your team. Automation over time charts the same thing daily, with the filled area being what settled automatically against everything decided that day.
A falling automation rate is rarely a change in the engine. It is usually a change you made, an influx of a document type we read less confidently, or a rule sending more cases to a person than you intended. The next two panels are where you find out which.
What sends work to a reviewer
Findings on the cases your team actually had to read, ranked, across a stated number of cases.
Read this panel with one thing in mind: a finding appears here because your flow sends it to a person. That may be the engine being unsure, or it may be a rule you set. If the top row is something you configured to always go to review, the queue is behaving exactly as instructed and the fix is on the flow, not in the engine.
Why applicants are rejected
The same idea over rejections rather than reviews: which findings sat on the verifications that ended in a no. If one reason dominates, it is worth asking whether your flow is rejecting the people you meant to reject, or the people whose phone camera is bad.
Export rejections downloads every rejection in the period with the reason recorded against it, including the reviewer's internal note. It asks for a code from your authenticator app first.
Not in the rate
Volume that is counted but deliberately kept out of the automation percentage, because none of it measures our engine deciding: sessions still open, sessions abandoned by the applicant, decisions made out of band, and anything carried over from a previous platform. Keeping them out is what stops a quiet week of abandonment from reading as an automation collapse.
Your reviewers
Everybody on your team who decided a case in the period, busiest first, with how many they took, how they split between approve and reject, and the reasons they most often gave.
Two different times sit beside each other, and mixing them up is the usual mistake.
| Column | What it measures |
|---|---|
| To decision | From the moment the case was handed over, so it carries the queue with it |
| Engaged | From the moment the reviewer first opened the evidence, so it is time actually spent |
A case can wait hours before anyone looks at it. If to decision is long and engaged is short, you have a staffing or an assignment problem, not a slow reviewer. If a reviewer shows no evidence opened at all, they decided without opening the captures, which is worth a conversation.