Post-call triage desk for AI phone calls, wired to HubSpot

Build me a post call triage desk where my sales team works the results of the AI phone calls we run with Bland AI, so the outcomes actually reach HubSpot instead of sitting in a call log. It opens on a queue of recent completed calls and the team works that queue down to zero.

The queue view loads recent completed calls with the Bland AI List Calls operation, defaulting to the last 7 days with a date range control, and buckets every call into booked, wants a callback, not interested, voicemail and failed. Show the buckets as tabs or columns with counts, plus filters for date range, phone number and whether the call has been handled. Each row shows the number called in E.164, the call time, the duration, the outcome bucket, the matched contact name and the reviewed and handled state.

Opening a row shows a detail panel. Pull the full transcript and call metadata with Get Call Details, and show the answers to my standard question set from Analyze Call. The question set is configurable in app settings, something like whether they booked, what objection they raised, when they want to be called back and who the decision maker is.

Analyze Call is billed by Bland per question every single time it runs, even for a call that was already analyzed, so run it at most once per call. The first time a rep opens a call that has no stored answers, run the analysis once, save the answers against that call id, and serve the stored copy on every later open. Never re-run it from a background refresh or when the queue reloads. If someone genuinely needs a fresh pass after editing the question set, put that behind an explicit Re-run analysis button with a confirmation that says it will be billed again.

Give me a player for the call recording using Get Call Recording. Only calls that were dispatched with recording enabled have audio, so when there is no recording show a plain no recording available state next to the transcript rather than throwing an error.

Match every call to a HubSpot contact on the phone number using Search Contacts, normalizing the number to E.164 first and searching both the phone and mobile phone fields. Cache the match against the call so the lookup does not repeat on every render. When more than one contact matches, show the candidates and let the rep pick the right one. When nothing matches, say No contact matched clearly on the row and offer a create action that uses Create Contact with the number and anything useful the transcript gave us, so a call is never silently lost.

From any row a rep can write the call up with Create Note associated to the matched contact, pre-filled with the outcome, the extracted answers and a short summary they can edit before saving. They can update fields such as lifecycle stage on the contact with Update Contact. They can hand a hot one to a human by creating a follow up with Create Task, associated to the contact with an owner and a due date. And they can redial straight away with Send Call, reusing the task or pathway from the original call and confirming the number before it dials.

Persist per call state inside the app so two reps never work the same row twice. Store reviewed, handled, who claimed it and when, alongside the cached analysis answers, the cached contact match and a record of which CRM actions have already been fired for that call. Show the claim on the row, hide handled calls by default with a toggle to show them, and warn when someone opens a call another rep claimed in the last few minutes.

Everything else is read fresh from Bland AI and HubSpot. The only things the app owns are the cached analysis answers and the per call working state. This is a shared surface for a small sales team, so keep the queue dense and fast, make it easy to move between rows, and always show how many calls are still unworked.