Analytics and reports
AI voice agent analytics for every agent and campaign
Reports shows how many calls you made, how many connected, how many reached the outcome you count as success and how many minutes they used, each against the period before. Filter by agent, campaign or channel, then download the calls as CSV.

What is AI voice agent analytics?
AI voice agent analytics is the count and comparison of what your AI agents' calls did: how many there were, how many connected, how they ended and how long they took.
In Bananaflow it is one Reports page, counted from every finished call, with the same filters for the CSV.
New to voice agents? Read how a voice agent call works, one turn at a time.
Building a floor dashboard? Read which metrics a call center dashboard should show.
How do Bananaflow reports work?
Every finished call is counted the moment it ends, so you pick a date range and filters and the page fills in. There is nothing to build. Setup takes five steps.
Mark which outcomes count as success
For each agent, choose the outcome codes that mean success, such as qualified. It is one setting, and it is what the success rate counts.
Run calls
Campaigns, inbound, web or chat. Each finished call is counted.
Open Reports and pick a range
Choose Today, 7 days, 30 days or a custom range of up to a year.
Narrow it down
Filter by agent, campaign or channel: phone, web or chat.
Download or open the calls
Download the same calls as a CSV, or open any one of them.
Want your reports set up around your own outcome codes?
What do the reports show?
Four numbers, each against the period before: calls, success rate, connect rate and minutes used. Under them sit a trend chart, the outcomes list and the busiest hours. Four things make that useful.
The four numbers a call floor runs on
Calls, success rate, connect rate and minutes used, each with its change.
Connect rate 36%, +9 pts vs prior 7 days.
- Calls, success rate, connect rate and minutes used, with the average call length under the minutes.
- Each shows its change against the same-length period just before: a percentage for counts, points for rates.
- Success counts the outcome codes you choose for each agent, so "success" means your goal. It needs outcome codes set up, and the tile stays hidden until then.
Trends, outcomes and the hours that answer
See how the days went, how calls ended and when people pick up.
Most calls at 11:00 · Asia/Kolkata. 41% connected at 18:00.
- Daily bars, or hourly bars when you look at one day, split into successful and other calls.
- Every outcome code with its count and share of calls, most common first.
- A 24-hour strip of calls and connect rate in your own time zone, plus a one-year heatmap on Overview with streaks and your busiest weekday.
Filter any way, export the same calls
Narrow the page, then download exactly the calls it counts.
loan_amount, city, callback_time as CSV columns.
- Reports filters by date, agent, campaign and channel. The Agent Runs list adds disposition, duration, direction and the caller's or called number, and keeps the filters in the link so a teammate sees the same view.
- The CSV has one row per call, with one column for each field your agent extracts.
- It also carries the recording and transcript links, and it has no cost column and no QA columns.
Campaign numbers and the minutes each call used
Read a campaign on its own, and see what each call used.
v7: 50% target, 49.6% actual, qualified 212.
- Filter Reports to one campaign, or download its own CSV for any time window.
- Each agent version in an A/B split shows its target and actual share, completed calls and outcomes.
- Billing's minute history lists every call with the minutes it used, linked to the call, on hosted billing.
Dates are calendar days in your organization's time zone. A warning shows how many calls failed to start in the range. Reports does not include Call QA results: for scores, tags and summaries, see Call QA.
To run a campaign and compare agent versions, see calling campaigns.
How does a call center use the reports?
An operations lead reads connect rate by hour, moves the calling slots, and checks next week's numbers against this week's. This example is made up: the names and numbers are illustrative, not a customer.

Monday, 9:30 am
Meera opens Reports
A call center in Kochi runs home-loan lead qualification for a lender: about 3,000 outbound calls a day in Malayalam and English, in slots from 10:00 to 13:00 and 15:00 to 17:00. Operations lead Meera Nair used to pull dialer exports into a spreadsheet every Friday. Now she filters to the campaign and Phone, and picks 7 days. Calls 20,400, connect rate 27%, success rate 11% (outcome qualified).
Monday, 9:40 am
The busiest hours say move
The strip shows most calls at 11:00, but only 19% connected then, and 38% at 16:00. She moves one morning slot to 17:00 to 19:00 in the campaign's calling slots.
Wednesday
One version wins
The campaign's Agents table shows the A/B split: version 7, with a shorter opening, has qualified on 212 calls against 164 for version 6, at the same share of traffic. She tells the developer to move all traffic to version 7.
Friday
The client gets the week's CSV
She downloads the week's CSV. It has loan_amount and city as columns next to each outcome and recording link, so no one copies notes by hand.
Campaigns can tag each call with an outcome you define. See AI outbound calling.
How do you get the data out?
Download a CSV from any filtered view, receive a webhook after each call, or send call events to your own BigQuery table. Three ways to take the numbers further.
A webhook after each call
When a call ends, a webhook step you add sends the outcome and extracted fields you have set up, and the links to the recording and transcript to your own endpoint. A failed delivery is retried up to 5 times, with longer waits between tries.
Call events in your BigQuery
Switch it on in Settings and Bananaflow sends one row per diagnostic event into your own table: speech, turns, tool calls, response times and errors. It holds no transcripts or outcomes, and delivery is best effort, so use the webhook or the CSV for business results.
The call behind each number
Every number comes from real calls. Open a call to hear the recording, read the transcript and see its outcome, if you have set outcome codes up. Call QA scores sit on the call's page too, but they are not counted in Reports.
Score each call with Call QA.
Calling campaigns
Run a CSV list, split traffic between agent versions and read each one.
Call QA
Score each call and tag what went wrong, on the call's page.
AI outbound calling
Campaigns that tag each call with your outcome codes.
Pricing
What a minute costs on each plan.
How a voice agent call works
One turn at a time, from the caller's last word to the reply.
FAQ
Analytics questions, answered
What is AI voice agent analytics?
AI voice agent analytics is the count and comparison of what your AI agents' calls did: how many there were, how many connected, how they ended and how long they took. In Bananaflow it is one Reports page, counted from every finished call, with the same filters for the CSV.
How is success rate worked out?
It is the share of calls that ended with an outcome code you marked as success for that agent, such as qualified. If no codes are marked, the tile stays hidden instead of showing a guess, and outcomes read as unknown.
What counts as a connected call?
A phone call that reached a conversation. Calls the carrier reports as not connected, and calls rejected before they start, are left out. Connect rate covers the phone channel only.
Which time zone do the reports use?
Your organization's own time zone, for the date range, the daily bars and the busiest hours.
What is in the CSV export?
One row per call: IDs, campaign, agent version, time, phone number, outcome, duration, a column for each field your agent extracts, tags, and links to the recording and transcript.
Can we see what each call cost?
You can see the minutes each call used in Billing's minute history, linked to the call, on hosted billing. Your plan's per-minute rate turns that into money, and the pricing page lists the rates.
Can we send the data to our own tools?
Yes. A webhook step sends the outcome and fields you have set up after each call, and diagnostic call events can go to your own BigQuery table. Transcripts and outcomes go out by webhook or CSV, not through BigQuery.
Bring us one process. We'll pilot it on your real calls, with your team as the fallback.
Plan a pilot