Capture structured data from calls

Turn conversations into rows. Which fields to collect, what Confirm and Required do, and the one thing about keys that will bite you later if you get it wrong.

7 min readUpdated 26 August 2026

A call is a conversation, and a conversation is hard to count. Data capture turns each call into a row with named fields, which is what makes calls reportable rather than merely listenable.

The panel describes itself plainly: fields collected from each call, where you toggle what to capture, the agent confirms the marked ones live, and the rest are filled after the call.

How a field is put together

Each row in the list has the same parts.

PartWhat it does
On/off toggleWhether this field is collected at all
LabelThe human name, such as Customer Name
Typetext, phone, email or date
BadgeExtracted, or Auto for something the system already knows
KeyThe column name in Collected Data and in exports
ConfirmWhether the agent checks the value out loud during the call
RequiredWhether the call is expected to produce it

Rows can also be reordered and deleted. Order matters more than it looks, because it is the order the agent works through the ones it has to confirm.

Extracted and Auto are different things

A field marked Auto is something the platform already has, such as the phone number the call came from. Nothing has to be asked and nothing can be misheard.

A field marked Extracted comes out of what was said. That is where the value is and also where the errors are, which is what Confirm exists for.

When to turn on Confirm

Confirm makes the agent check the value out loud before the call ends. It costs a few seconds and it is the difference between a usable record and a plausible one.

Turn it on for anything you will act on without a human checking first.

  • Email addresses. Almost always worth confirming; they are misheard constantly and a wrong one fails silently.
  • Names, if you will use them in a follow-up. Spelling matters more than people expect.
  • Dates and times for anything being booked.
  • Amounts, reference numbers, anything with digits.

Tip

Do not turn Confirm on for everything. A call that reads eight fields back to the caller is a call people hang up on. Confirm what you will act on, and let the rest be filled after the call.

What Required actually means

Required marks a field as expected. It is a signal about what the call is for, not a barrier that traps a caller who does not want to give you their email.

Use it on the handful of fields that make the record worth having. Marking everything required tells you nothing when you later look at which calls came out incomplete.

Keys are the part to get right first

The panel warns about this directly, and it is the single most consequential thing on the screen: each field's key becomes its column name in Collected Data and in CSV exports, so it should be short and readable, in the style of buyer_full_name. Changing a key later starts a new column, and rows collected under the old key keep it.

So a key rename does not tidy your data. It splits it. Two columns, half the rows in each, and every export from then on carrying both.

  • Lower case, words joined by underscores.
  • Say what it is, not what the agent asked. customer_email, not their_email_address.
  • Keep them stable across agents. The same thing should have the same key everywhere, or reporting across agents becomes manual work.

Careful

Decide your keys before you collect anything real. Renaming after a hundred calls leaves you merging columns by hand.

Where the data comes out

Collected Data in the sidebar. It describes itself as structured data captured from your calls, where you review flagged fields and export.

Its empty state points back the way you came: configure capture fields on an agent, and rows appear there after each call. If that page is empty, the fields were never turned on for the agent that took the calls.

Rows can be filtered by agent, status and source, the visible columns can be chosen, and the result exported.

Common questions

Collected Data is empty. Why?
Capture fields are configured per agent. Open the agent that took the calls, open Data capture, and check the fields are on. Rows only appear for calls taken after that.
What is the difference between Extracted and Auto?
Auto is something the platform already knows, such as the caller's number. Extracted is taken from what was said, which is why those are the ones worth confirming.
Can I rename a field key later?
You can, and it starts a new column. Rows already collected keep the old key, so you end up with both. Settle keys before collecting anything you care about.
Should I mark everything Required?
No. Required is a signal about what matters, and marking everything makes the signal useless when you look at which calls came out short.
Does confirming a field make calls longer?
Yes, by a few seconds each. That is the trade. Confirm what you will act on and leave the rest.
Can I export what was captured?
Yes, from Collected Data. Each field's key is the column name, which is why readable keys matter.

Was this helpful?

Related