Call Intelligence: From Recordings to Outcomes

Act 01Context

We started by helping business owners replay conversations. We ended by helping them understand outcomes.

When I joined LinkedPhone, calls were essentially entries inside a list: caller, timestamp, direction, team member, status. The system knew a call happened. It understood nothing about it.

To verify a request, confirm a detail, or recall a promise, a business owner had to find the call, open the recording, listen to the whole thing, remember the details, then act. Everything happened in the owner's head. The software only stored the audio.

Chapter theme

The real cost wasn't storage. It was accessibility.

"I know the information is there. I just don't have time to listen to it."

Executive Summary

Nobody actually wanted a recording.

Challenge

Important information lived inside conversations. Once a call ended, that information was often lost, or locked behind minutes of audio nobody had time to replay.

Breakthrough

Customers asked for recordings, then transcripts, then summaries. But nobody actually wanted any of those. They wanted to know what happened, and whether they needed to do anything about it.

Outcome

Call Intelligence evolved from recording storage, to conversation understanding, to business outcome detection, the foundation Lisa would later be built on.

The evolution

RecordingRaw reality, stored

TranscriptText representation

SummaryMeaning, extracted

OutcomeWhat actually happened

Duration: ~2 yrsContinuous evolution
Platforms: MobileWeb
Systems: MultipleCalls · Contacts · Tickets
AI layers: 4Transcription to sentiment
Legacy: FoundationFor Lisa

Where We Were

Calls were historical records. Nothing more.

The mental model was simple: call, then history. The system knew a call happened. It understood nothing about it.

Screen · Version 1, the call list

Simple, and completely flat

Caller, timestamp, direction, status. No context, no understanding, just records.

Act 02Problem

The Problem

Recordings solved storage. Not understanding.

A customer called, asked about pricing, requested a callback, mentioned a date, promised to send documents. All of it trapped inside audio nobody had time to replay.

Conversation

Recording

Time required to listen

Information hidden, action missed

The problem wasn't information availability. It was information accessibility.

Investigation

Nobody was consuming calls start to finish.

Users rarely opened calls because they wanted a recording. They opened them because they wanted an answer: what did they ask, did we promise anything, do I need to call back.

When users did open recordings, most skipped through the audio, searched for a specific moment, and wanted only one or two pieces of information. Nobody was consuming calls linearly. Everyone was hunting for a signal.

Artifact · Listening behavior

Skipping, searching, hunting for signal

Playback data made the real question obvious: what do I need to know, not what was said.

Insights

Every layer we built, users skipped past.

Insight 01

Users don't want recordings. They want information.

Observation: Most listens ended within seconds of finding the answer.
Meaning: The recording was the path, never the destination.
Implication: Surface the answer directly.

Insight 02

Users don't want transcripts. They want conclusions.

Observation: Transcripts were opened, then abandoned almost immediately.
Meaning: Reading everything is no faster than listening to everything.
Implication: Summarize, don't just transcribe.

Insight 03

Users don't want analysis. They want decisions.

Observation: Even summaries left one more step: what do I do now.
Meaning: Understanding isn't the finish line, action is.
Implication: Extract the action, don't stop at the meaning.

Insight 04

Calls are actually work.

Observation: Every important call created tasks, follow-ups or new customer data.
Meaning: The conversation isn't the outcome. The work it creates is.
Implication: Design for the work a call generates, not the call itself.

Act 03 · Turning Point

We thought users wanted transcripts.

They wanted conclusions.

Later this evolved further: we thought users wanted conversations. They wanted outcomes. Everything afterward was designed around that idea.

Exploration Lab

Five versions, each one closer to the answer.

Calls as recordsStore it, nothing moreV1

Expandable callsContext lives where the call livesV2

Summaries appearUnderstand a call in secondsV3

Action extractionUnderstanding leads directly to actionV4

Dedicated call detail pagesIts own experience, not squeezed into a list itemV5, final

Breakthrough

Calls stopped being conversations. They became outcomes.

The question changed from "what was said" to "what happened." That single shift changed everything downstream.

The call intelligence stack

RecordingRaw reality

TranscriptText representation

SummaryMeaning extraction

SentimentEmotional context

OutcomeAppointment booked, lead captured

Calls stopped being conversations. They became work. And once calls became work, everything else became possible.

Act 04System

System Architecture

Seven layers between a sound wave and a decision.

Call recording

Raw reality

Transcription

Text representation

Summary

Meaning extraction

Sentiment

Emotional context

Action detection

Follow-ups, tasks

Outcome detection

Business intelligence

Design System Impact

A vocabulary for intelligence, reused across the product.

Summary blocks

Component

Sentiment chips

Component

Action cards

Component

Event markers

Pattern

AI attribution

Pattern

Confidence indicators

Pattern

Many of these later became Lisa's own patterns.

Product Showcase

The final leap: outcome as the hero.

Once Lisa started answering calls nobody on the team had attended, the software needed to explain everything: reason for calling, information captured, callback and appointment requests.

Screen · Call detail page

Summary, sentiment, action, in one place

Its own dedicated experience, not squeezed into a list item.

Screen · Outcome card

Appointment scheduled, transfer failed, lead captured

The business owner understands the entire situation in seconds, not minutes.

Interaction & Motion

Recording, transcript, summary, outcome: one motion.

Progressive disclosure carried the intelligence layer the same way it carried Auto Attendant: audio timeline mapping, call flow breadcrumbs, and outcome cards designed for immediate scanning.

Act 05Impact

Validation

Users skipped straight to the outcome, on their own.

Transcript opens declined over timeUsers increasingly skipped straight past the raw text.

Outcome cards became the default viewMost users never needed to go deeper than the card itself.

The hypothesis heldNobody actually wanted a transcript. They wanted an answer.

Ripple Effects

Without this chapter, Lisa doesn't exist.

Call Intelligence to Lisa Missed Calls, to Lisa All Calls, to the Unified Work Model and every AI system since. This is the chapter where LinkedPhone's AI story truly began.

The evidence vault.

Early call list screens

Figma

Accordion explorations

Figma

Summary iterations

Figma

Sentiment experiments

Doc

Prototype videos

Video

Internal AI experiments

Doc

Lessons

What we believed. What we learned.

Lesson 01

We believed users wanted recordings.

We learned they wanted information.

Lesson 02

We believed users wanted transcripts.

We learned they wanted conclusions.

Lesson 03

We believed calls were communication.

We learned calls create work.

Lesson 04

We believed AI should explain conversations.

We learned it should explain outcomes.

Legacy

The moment calls stopped being media.

This chapter marks the point LinkedPhone stopped treating calls as recordings, and started treating them as business events. That reframe made everything that came after, especially Lisa, possible.

Closing

Calls stopped being conversations. They became work. And once calls became work, everything else became possible.