Confirmed
- The customer said the answer solved it.
- The action completed — the refund posted, the order shows delivered.
- Someone on your team marked it resolved.
Reports that will not flatter you.
Aqvitus never shows a resolution it confirmed and one it only assumed as a single number. Every figure is a projection of the event log, and clicking it opens the conversations it counted.
Overview · last 30 days Example data
Autonomous resolution rate
- Confirmed 58%
- Assumed 16%
- Went to a person 26%
There is no combined 74%. Aqvitus will not print one.
- Draft approval rate81%
- Median review time4m 12s
- Compensated actions0
A customer going quiet is not a success
Most reporting counts a closed conversation as a win. Aqvitus separates the ones where something actually confirmed it from the ones where the conversation simply stopped — in the inbox and in every report.
Assumed
- The customer stopped replying.
- The chat window was closed.
- Nobody reopened it, so it aged out.
Autonomous resolution rate Example data
- Confirmed 58%
- Assumed 16%
- Went to a person 26%
58% + 16% = 74% Aqvitus will not print this number. Add them up yourself if you want to.
Example data, over four weeks. Clicking any figure in the console opens the conversations it counted.
- One chart, built from the event log. Nothing separate to instrument or tag.
- Confirmed: the customer said so, the action completed, or a person marked it resolved.
- Assumed: the conversation simply stopped. Its own line, its own label, on the same chart.
- The two are never added up. There is no combined figure here, and there will not be one.
The same split runs through the numbers built on top: cost per resolution, CSAT by lane, every trend. You can add them up yourself if you want to. Aqvitus will not do it for you.
Every metric has a written definition
Not a tooltip. The formula is part of the specification, so the number means the same thing in month one and month twelve, and the same thing in your workspace as in anyone else’s.
Autonomous resolution rate
Conversations that closed in the Auto lane, divided by everything that closed — with spam left out. Always shown as two numbers: confirmed and assumed.
Draft approval rate
Drafts a person approved, with or without edits, divided by drafts that were resolved. Beside it, the median time from a draft appearing to someone dealing with it.
Cost per resolution
The model, telephone and transcription spend Aqvitus can attribute, divided by resolutions — broken out by which deployment served them.
CSAT by lane
What customers said, reported separately for conversations the AI finished and conversations a person finished, as well as overall.
Escalation mix
Why conversations went to a person, by reason code, trended — so you can see which reason is growing.
Response and resolution times
Medians and the 95th percentile, plus SLA attainment reported per policy rather than averaged into one figure.
How well the AI is actually doing
A tab of its own, and most of what is on it is uncomfortable on purpose.
Was it as sure as it should have been?
When the AI says it is confident, does that hold up? Aqvitus plots the confidence band it predicted against what actually happened, so over-confidence shows as a gap rather than hiding inside an average.
Example data. Predicted band against the share that actually resolved.
What the checks caught, and what got out
- Verifier interventions by check — which pre-send check stopped a reply, counted separately, so you can see whether one rule is doing all the work.
- Automated review scores, as a distribution rather than an average.
- Compensated actions — the count of things that had to be undone. The target is zero, and it is shown even when it is not.
What your customers are asking about this week
Intent volumes with the change on the period before, the sentiment mix, and a watch for spikes.
A spike you did not go looking for
When one intent or a cluster of phrasings jumps statistically, Aqvitus raises it as an anomaly card with the affected customers attached.
And something to do about it
An anomaly card suggests the workflow that would address it — reaching out to the affected batch, for instance — rather than leaving you with a chart.
A forecast with its uncertainty shown
Projected volume for the next period with a confidence interval, and a staffing hint from the volume expected to reach a person.
No separate tracking to set up
Every report is computed from the event log. There is nothing to instrument, nothing to tag, and no chance of the dashboard and the history disagreeing.
See where the events come from- Filter by date range, segment and channel, everywhere.
- Click any figure to open the conversations it counted.
- Four surfaces — overview, AI quality, what customers are asking, revenue.
- Export on a schedule, sent from your own domain.
How this connects
Workflows
A metric crossing a threshold can start a workflow — including pausing the Auto lane for one intent.
WorkflowsKnowledge
A growing intent with no article behind it is exactly the gap the knowledge system is already drafting.
KnowledgeSales
Revenue attribution has four separate measures, each one traceable back to the conversations behind it.
SalesApprovals
Verifier interventions and compensated actions are reported by check, so you can see where the rules are earning their keep.
Approvals
AI First. Human Always.
Numbers you can defend in a meeting, because you can open the conversations behind them.