What metrics means here
Metrics are aggregates over a deployment's calls. The raw material is the per-call record: see call logs and post-call analysis. This page covers what each number counts, where it appears, and how it misleads.
Metrics are read in the dashboard. There's no metrics endpoint. To compute your own, use the call log CSV (exports) or the call.completed webhook (webhook-post-call).
Two kinds of "success"
Keep these apart.
| Measure | What it counts |
|---|---|
| Connected | The call ended with status completed, meaning someone picked up and the call ended normally. A telephony fact. |
call_successful | The post-call analysis judged that the call met its goal. Set per call by the analysis, not by one of your questions. |
If your goal is narrower ("booked a demo", "agreed to pay"), add a Yes/No question for it and count that answer yourself from the CSV or webhook. See analysis fields.
Deployment analytics
The cards at the top of a deployment's page:
| Card | Definition |
|---|---|
| Call counts | Calls in the deployment, with how many were picked up. |
| Connect rate | Picked-up calls ÷ total calls. |
| Call duration | Average duration of the deployment's calls. |
| User sentiment | Average sentiment across the deployment's calls. |
Deployments that dial in parallel also show:
| Metric | Definition |
|---|---|
| Success Rate | Calls with call_successful = true ÷ the deployment's planned contacts while it runs, or ÷ calls made once it's stopped. |
| Completion Rate | Contacts processed ÷ the same denominator. |
| Total Cost | The sum of the deployment's call costs. |
Because a running deployment divides by planned contacts, Success Rate starts low and climbs as calls are made. Read it after the deployment finishes.
Live Deployments
| Card | Definition |
|---|---|
| Call counts | Calls so far, split into completed and failed. |
| Active calls | Calls currently initiated, ringing or in progress. |
| Success rate | Completed calls ÷ calls dialled. This is a connection measure: it doesn't use call_successful. |
| Avg duration | Average duration of completed calls between 1 second and 30 minutes. |
Other views
| Page | What it shows |
|---|---|
| Dashboard (home) | Call counts, call duration and user sentiment for the week, month or year, with the change against the previous period. |
| Analytics | The deployment execution log. Select deployments to Export ROI report (an HTML report) or build a Merged report of their post-call analysis. |
Where numbers mislead
Running deployments move. Rates are computed over the calls made so far. Wait for the deployment to finish, or for at least 100 calls, before you act on a number.
Small samples are noise. A rate over 12 calls tells you little. The Connect rate card shows how many calls it's based on.
Dashboard data can lag. Deployment data can be up to 5 minutes old. If a number hasn't moved after a change, wait and reload before assuming the change did nothing.
Same word, different measure. "Success rate" on Live Deployments counts connected calls. "Success Rate" on a parallel deployment's analytics counts call_successful. Compare like with like.
Drift is usually the list. A success rate that slowly drops over weeks most often means the audience is going stale, not that the Finn changed. Check connect rate and call outcomes before you rewrite prompts.
Reading a bad campaign
Work in this order. Each step rules out a cause before you spend effort on the next.
- Connect rate. Low connect rate with healthy results on connected calls is a list or number problem. No prompt change fixes it. See spam labeling.
- Call duration. Very short calls mean people hang up on the opening. Check the welcome message.
- Blank analysis answers. A question that is blank on most connected calls means the agent never gets to that topic.
- The worst calls. Open a handful of negative-sentiment calls and read the transcripts. Patterns usually show within three calls.
- A baseline. Run the same audience on a known-good Finn for 50 calls and compare the same numbers.
Don't read every call. Read the summary numbers, then the ten best and ten worst calls.
Related
- Post-call analysis: the questions you can count against.
- Monitoring: watching a deployment while it runs.
- Evaluations: testing a Finn before it produces metrics.
- Exports: CSV and PDF output.