Legal · Stowe Family Law

Why enquiries don't convert, answered from hundreds of recorded calls in one click instead of a month of listening.

Understanding why new-business enquiries did not convert meant a person listening to hundreds of recorded calls spread across three separate phone systems.

Talk to us about something similarAll case studies
ClientStowe Family Law
SectorLegal
WhenLive
Used byClient services and operations staff, an offices manager, an operations console user, Hub admins
Plugs intoAzure Blob Storage, Azure SQL, Vonage Contact Center, RingEX, Deepgram, OpenAI, Prolink Hub
AITranscription with speaker diarisation; batch analysis returning themes, blockers, coaching and compliance risks; grounded Q&A that cites calls and admits gaps; plain-English filters validated against live values
What we built

Ingest every call, transcribe it, analyse a thousand at a time, and let people ask follow-up questions with evidence.

An ingest and transcription pipeline across three phone systems, every call transcribed with speaker diarisation and timestamps.
A Calls browser filtering the library by date, source, region, cohort, extension, queue, outcome and business flags, turned into an analysis batch in one click.
Batch analysis returning a summary, top themes, conversion blockers, coaching opportunities, compliance risks and a per-call breakdown with quotes linked to the call.
Batch Q&A answering free-text follow-ups with per-call evidence and explicit admissions of what the data does not show.
Presets that save a filter plus a brief as a one-click reusable run, with a history of recent runs and live progress.
An operations console showing provider status and credit, storage and database health, job backlog and one-click run-now.
ai

Smart Filter

Type 'new business calls to client services last month' and the filter form fills itself, every value checked against the real database so it cannot invent an option.

cost

Idempotent batches

An identical call selection with the same brief reuses the existing batch instead of paying for a duplicate run.

operations

Self-monitoring jobs

Every job logs start, finish and errors to one table, records even fatal crashes, and a daily reaper catches the rest.

How it plugs in

Three phone systems, one library.

Recordings are pulled from each system on their own schedules, enriched from the firm's employee warehouse, transcribed once, and embedded for semantic ranking in the weekly digest.

The applicationAzure Blob StorageAzure SQLVonage Contact CenterRingEXDeepgram
job Source ingest on cron: recording scans about every ten minutes, contact-centre polling on a rolling window, short and outbound calls skipped to save cost
job Download and transcribe every two minutes; batch analysis every five to eight minutes with email to subscribers
job Monday 08:00 digest: top lawyers by volume, the five most concerning calls, five strongest new-business signals
What changed

The monthly client services retro that meant listening to hundreds of calls became a one-click preset run, with non-conversion reasons returned as labelled buckets with percentages and example calls.

one detail only this project hasThe weekly digest ranks the week's transcripts against probe queries to surface potentially concerning calls, an angry client, a complaint, a threat to leave, alongside the best new-business signals.
Figures from the project record: batches of up to 1,000 transcripts, split into 50-call chunks; about 15 background jobs on the health board
Something similar?

What are your recorded calls telling you that nobody has time to hear?

Book a walkthroughRequest a callback