The starting point
SaaSify's founding team was doing the right things — they just didn't scale. The two founders ran sales themselves, splitting roughly 30 hours per week between LinkedIn prospecting, list-building in Google Sheets, and writing cold emails by hand.
The output: about 200 outbound emails per week, a 0.8% reply rate, and a $480 customer acquisition cost. That was viable when they were closing 4 deals a quarter. It wasn't viable for the growth plan their investors had signed off on.
The real problem wasn't volume. It was that their best two people were spending 60% of their time doing work that should have been infrastructure.
What we did
We built them an AI prospecting system that runs continuously in the background — identifying ideal-customer-profile (ICP) matches, enriching them with intent and firmographic data, generating personalised outreach, and routing replies to a real human. The founders went from doing prospecting to auditing it.
1. ICP definition that the model could actually use
We spent the first week not on tooling, but on the ICP itself. We interviewed five of SaaSify's best-fit customers to surface the patterns: company stage, tech stack tells, recent triggers (funding, hires, product launches), and the specific job-to-be-done language they used. The output was a structured ICP document — the model's brief.
2. The data layer: Apollo + Clay
Apollo gave us the universe of 2.4M companies matching the firmographic shape. Clay enriched each company weekly with intent signals — recent press, hires posted, tech stack changes, growth velocity. The result: a continuously-refreshed shortlist of ~2,000 ICP-matched accounts the team could actually act on.
3. The personalisation layer: GPT-4 + structured prompts
This is where most AI outbound systems fail. Generic LLM personalisation reads worse than no personalisation at all. We built a structured prompt that fed the model: the prospect's role, the company's recent triggers, the SaaSify product context, and a tightly-controlled tone guide. The model returned three opening lines per prospect; the system picked the one with the best match score to the ICP brief.
4. The delivery layer: Instantly + warm-up
Three rotating sending domains, full SPF/DKIM/DMARC setup, gradual ramp from 20 emails/day per inbox to 80, dedicated IPs warmed over 14 days. Deliverability is the silent killer of cold outbound — if your emails land in spam, the rest doesn't matter.
"Our pipeline went from dripping to flooding. The AI does in an hour what used to take our team a full week."
Rajan K., Founder — SaaSifyWhy it worked
The 6.4% reply rate isn't because GPT-4 wrote magical emails. It's because three things lined up: (1) we only contacted prospects who were currently in-market, (2) the opening line referenced something only a human researcher would notice, and (3) the emails actually arrived in the inbox.
Most agencies sell one of those three. We treated them as a system. Take any one out and the number falls back to industry average.
What's next
SaaSify hired their first full-time AE in month 4 — using the qualified-lead flow we built as the AE's pipeline floor. The founders are now in their next product line, and we run the lead-gen system on a thin retainer. CAC continues to decline as the model gets better at predicting which prospects convert.