This week at Roark

See everyone who has talked to your agents in one place, with a profile for each caller and chat visitor that fills itself in from your conversations and that your team can correct.

What you can do:

• Browse and slice your customers: customizable columns (including any customer field), filters, sort and saved views you can share with your team.
• See where every detail came from: Overview, Activity and Notes tabs, with each captured value linked to the customer's own words and the moment in the conversation.
• Fix a wrong detail in one click: open a field's history and pick "Use it" to restore an earlier value. Edited fields are locked, so newer conversations can't overwrite them.
• Choose what you capture: in Settings > Customers, add your own fields, switch built-in ones off and choose which fields Roark fills from conversations.

Also in the API, SDK and CLI as the customer resource (/v1/customer).

Off by default while we roll it out. Ask us to turn it on for your account.

Get a Run Debrief for your production calls: a scheduled report that grades your live calls against the targets you set and lands in your team's inbox or Slack, daily, weekly or monthly.

What you can do:

• Set the bar: targets like a task completion pass rate of at least 85% or average frustration of at most 2, for one agent or all of them, narrowed by modality, dataset or call properties.
• See what changed and why: a verdict and executive summary, each target against the previous period, failure themes checked against transcripts with the calls behind them, and recommendations.
• Share it: email and Slack delivery with a PDF attached, any past period one click away, or write the latest one on demand.

Also in config as code (kind: liveReport) and the API (/v1/live-report).

👉 Read the guide

Test your agent against phone menus that behave like the real thing, from hold music to keypad timeouts to a hand over to a live person.

What's new:

• Hold music: a new Audio step plays hold music, a hold beep or ringback for a set time. It counts as a turn, so your metrics don't read the hold as a pause.
• Strict IVR into improv: a strict scripted menu can open a simulation, route on the agent's actual keypresses, speech or silence, then hand over to the person the caller reaches.
• Hang up on silence: an Expected DTMF step can hang up when nothing is pressed, the way real menus do.
• End the call from the script: a scripted path can end the call itself, for example after "That extension is not available. Goodbye."
• Improv openings: an improv flow can speak first with its own line instead of waiting for the agent.