< 900 → 4,087
calls/month on SuperSales CRM
Case Study · SuperAGI · Dialler Module
From a call button to a full outbound calling engine
SuperAGI · Associate Product Manager · October 2024 — March 2025
~5x
in 3 months
< 4 months
zero to parity
" I inherited a single 'call' button. Three months later, we had
a full outbound dialling system with ~5x the call volume
and first-to-market AI features. "
The Situation
When I took ownership of the Dialler module in November, it wasn't a product — it was a button. Inside the SuperSales CRM, you could open a contact/lead and press call. Settings let you purchase a phone number to dial out. That was the whole thing. No inbound calls, no call campaigns, no parallel dialling, no dialler section at all.
Meanwhile the category had mature players — JustCall, Nooks.ai, RingCentral — offering full outbound suites. My brief was clear and on a clock — take this from a button to parity with the leaders, and do it fast enough that the sales team could start landing customers off the back of it.
In the month before I took over, the platform placed under 900 calls. That was the baseline I started from.
The Call
Two decisions shaped this track.
First — the build had to be a system, not a feature pile. Before writing a single PRD, I mapped what a mature outbound dialler (the "v3") would eventually need, then worked backwards. I scored every feature twice: how necessary it was to the user, and how much engineering effort it took. Necessity set the floor — the P0 features without which nothing ships meaningfully. Effort, inverted, sorted the rest into P1/P2, with genuine nice-to-haves parked at P3. Then I sequenced the build against the actual bandwidth of my two engineers, occasionally pulling a P1 forward and pushing a P0 back when that's what dev bandwidth allowed.
Second — we needed one feature the leaders didn't have. Parity alone doesn't move customers off an incumbent. My CEO set the harder goal: while we closed the gap, we needed a genuinely first-to-market feature — something that solved a live, unsolved problem — so the sales team had a reason to pull customers across before parity was even achieved.
I didn't start with an idea. I started with users. I spoke to customer-success reps I knew — one running JustCall, the market leader, in his daily workflow, and between them they'd cycled through six or seven platforms including RingCentral — and went hunting for the real friction. I found it in bulk outreach.
Here's the problem in plain terms. When you dial a long list of leads in parallel, every call that hits voicemail breaks the flow — the rep has to stop and manually leave a message. In markets like the US, where most people have voicemail enabled, this slows domestic outbound to a crawl. The only fix on the market was auto voicemail drops (Nooks.ai did this): the system drops a pre-recorded voicemail on every call that goes to voicemail, so dialling never stops. But that message was generic — the same recording to everyone — which quietly killed the lead. So you could scale your calling, but you'd bleed leads doing it. Nobody had solved scale without the bleed.
That was the gap I chose to build into.
What I Built
AI Voicemails — personalised voicemail drops, at scale, from a text prompt.
The insight was that SuperSales already had some pieces; nobody had connected them. Our internal AI SDR could generate a personalised outreach message from a lead's latest LinkedIn activity. I plugged that into a customisable layer where the rep defines what the voicemail should say, so the generated message reads like a real, relevant reason to call back — not a template. To get it spoken aloud, I used ElevenLabs: a rep records a sample of their own voice, the system learns it, and every drop goes out in the rep's own voice. Twilio handled the delivery. My job was the insight, the wiring, and a UI clean enough that a rep could run it without thinking.
The timing was the unlock. This feature was only possible the moment it shipped. Voice models had just crossed the line from obviously-synthetic to genuinely human — ElevenLabs had cracked it. Anything built 6 months earlier would have sounded like a machine and lost the lead anyway. We were early to a door that had only just opened.
Wireframes · Voicemail Settings library, and the Add Voicemail modal with a file uploaded and with the record control in its paused state
How AI Voicemails works
Lead's LinkedIn activity is fetched
↓→
AI SDR generates personalised script
↓→
Rep's voice clone reads the message out loud (ElevenLabs)
↓→
Twilio delivers the Voicemail
↓→
Lead hears their name with a custom message
Shipped · the Voicemails library in production, three drops with audio players
Around the hero feature, I also built the system the dialler actually needed to reach parity: inbound calling, lead lists to call through, parallel dialling, lead temperatures, and a clean UI on a stable backend.
I then picked up a second track in parallel — Sequences — once the Dialler roadmap was fully PRD'd out for the next three months and my engineers had a clear queue. Sequences was a standalone multi-channel outreach engine: a rep adds a lead and the queued steps run automatically. When I took it over it could only send AI-SDR emails. I added a wide range of task types including actions on linkedin, wired AI Voicemails into it, added a plethora of sequence templates, and let reps drop leads into a sequence from anywhere — the Chrome extension, the phone app, and several other entry points — bringing it to close parity with Apollo, Salesloft, Outreach, and Reply.io.
I ran both tracks end to end: competitor analysis, roadmaps, PRDs, wireframes, and UAT, working with one frontend and one backend engineer dedicated to each track, plus shared designers and QA — eight people in the orbit of the work. My SPMs oversaw and approved; the build was mine.
The Result
Zero to parity in four months. From a call button on October's basic baseline to a full outbound calling system at parity with the category leaders by the end of February.
Calls placed / month
< 900
inherited baseline
Oct
1,811
Nov
3,226
Dec
4,087
Jan
~5x
growth in 3 months — existing customers choosing to dial more
I'll be straight about what drove that, because it matters. This growth came majorly from two existing paid customers — Agora and SpatialChat — expanding their usage as the system matured. A third client, a US healthcare firm, had also been signed for 100 seats on SuperSales to primarily use Dialler, but hadn't started using the product yet — so they contributed nothing to these call numbers. That means the call volume growth is a clean signal: existing users chose to dial more because the platform got better.
On North Star metrics: the honest measure here would be calls per active user per week — it strips out the new-seat effect and isolates actual product engagement. I didn't have visibility into that number at the time, but the growth pattern here tells the same story: the users we had dialled more. That's what good looks like.
"I work back from the version three releases out, so nothing I ship needs to be torn down to get there. "