Will DiMaio
Index
05In Progress· 2026· Sales Tech

Building inside the rep's actual workflow · v2 in flight

SDR Copilot

Real-time call coaching, objection detection, and auto-drafted follow-ups — built inside the rep's existing dialer, CRM, and inbox, with sub-200ms objection latency that surfaces a coaching prompt before the rep finishes hearing the prospect's pushback.

<200ms
Objection detection
0
Sequenced interventions
0
Auto-sent messages
MV3
Chrome extension target
01Key Insight
WHY ON-DEVICE OBJECTION DETECTION

Objection detection at 200ms latency in a Chrome extension means no server round-trip — the coaching prompt appears before the rep finishes hearing the objection. A round-trip to a backend would add 200-400ms of network latency and break the perception of 'live' coaching. Running pattern detection client-side on the streaming transcript turns a server-side feature into a UX moment.

02The Problem

Why this exists.

The bottom 90 minutes of an SDR's day are tax. Pre-call research, post-call notes, Salesforce hygiene, post-call emails — none of it is selling, all of it is mandatory, and most of it gets done badly under time pressure. The rep ends up either skipping the work (and the pipeline rots) or doing it on Sunday night (and the rep burns out). Neither outcome is what anyone signed up for.

Existing AI sales tools mostly try to replace the dialer or the CRM. That's the wrong wedge. The dialer and CRM are sticky — reps and managers don't switch them on a whim — and the workflow tax lives between those tools, not inside them. The right wedge is a thin layer that runs alongside the existing stack and removes friction at every step without asking the rep to change a single tool.

03The Approach

How it works.

SDR Copilot is a Chrome MV3 extension that injects into Orum (the dialer), Salesforce (the CRM), and Gmail (the outbound channel). It lives where the rep already works. There is no separate app to learn. The rep's workflow stays exactly the same — Copilot just makes each step take less time.

Four interventions, sequenced across the rep's day. Two minutes before a call connects, Copilot synthesizes a 3-bullet pre-call brief from the prospect's LinkedIn and 90 days of company news. During the call, a Deepgram WebSocket streams transcription with sub-300ms latency, surfaces talk-track suggestions inline, and flags commitments in real time. Thirty seconds after the call ends, a Gmail draft is staged in the rep's voice with resources pre-filled and a calendar next-step proposed. At 6pm, an end-of-day audit fixes Salesforce hygiene without the rep touching a record.

The discipline is restraint. Copilot never auto-sends an email, never closes a deal stage, never books a meeting without a human approving. The rep stays in control of every commitment that goes out under their name — Copilot just makes the commitment cheap to make.

04WORKFLOW INTEGRATION

Lives in the tools the rep already uses.

01 / 03Orum

Dialer overlay

Content script injects into Orum's call screen. The pre-call brief appears in the corner; the live transcript streams below. The rep doesn't open a new app — Copilot is where their eyes already are.

02 / 03Salesforce

Activity sync

Calls are logged automatically with full context: outcome, mentioned resources, commitments. The EOD audit fills missing fields; the rep approves with one click. CRM hygiene becomes a passive side-effect of the workflow, not a Sunday-night cleanup.

03 / 03Gmail

One-click follow-ups

Drafts stage in the rep's voice, in their drafts folder, ready to send. The rep reviews and hits send — Copilot never auto-sends. Speed of follow-up goes up; volume of clerical work goes to zero.

05Features

What it does.

01 / 04T−2 min

Pre-call brief generator

Two minutes before the dialer connects, Copilot scrapes the prospect's LinkedIn, pulls 90 days of company news, and synthesizes a 3-bullet brief plus two conversation hooks. The rep walks into every call with the same context the best reps used to spend ten minutes building manually. The bar for 'prepared' goes up by an order of magnitude.

02 / 04During call

Live transcription and talk-track surfacing

A Deepgram WebSocket streams the call audio with sub-300ms latency. Copilot detects objection patterns, surfaces matching talk-tracks inline next to the dialer, and flags commitments and next-step language in real time. The rep gets a quiet copilot in the corner of the screen — never in the way, always one glance away.

03 / 04T+30 sec

Post-call drafts in the rep's voice

Thirty seconds after the call ends, a Gmail draft is staged in the rep's voice with resources mentioned during the call pre-filled, a next-step calendar slot proposed, and the Salesforce activity log queued. Copilot never auto-sends — the rep reviews and clicks send — but the cost of sending a thoughtful follow-up drops from ten minutes to ten seconds.

04 / 04EOD

End-of-day Salesforce audit

At 6pm, a cron runs over the rep's Salesforce activity for the day, flags stale records, fills missing fields where the call transcript clearly answered them, scores hygiene, and sends the rep a digest mail with one-click fixes. The CRM finally works the way it was supposed to — kept current, without a Sunday-night cleanup session.

06Achievements

What's shipped so far.

OBJECTION DETECTION
<200ms

Pattern detection runs on the streaming transcript client-side — no server round-trip. The coaching prompt appears before the rep has finished hearing the objection.

SEQUENCED ACROSS THE DAY
0 interventions

Pre-call brief → live overlay → post-call draft → EOD audit. Each one is a small, scoped feature; together they are roughly 90 minutes of daily tax removed.

REP STAYS IN CONTROL
Zero auto-sends

Copilot never sends an email, closes a stage, or books a meeting without explicit approval. Every commitment goes out under the rep's name with their review.