How it works
From a raw CSV to 10 signals per company in 30 minutes.
Six steps. Two are your decisions (ICP and output approval), four are the engine reading signals fast and deterministically. Simple split: humans decide, the machine measures. Never the other way around.
We define your ICP together
Week 1: we build your ICP together across 6 layers (filters, technographics, trigger events, champion, buying committee, anti-ICP) and map which of the 10 intent signals most strongly predict a purchase in your market. Yours, tailored to you, not generic.
You upload the starting list
CSV with company names. From Sales Navigator, a partner, a conference, your own database. SOutreach swallows dirty formatting and mangled names. This is your starting point for signals, nothing else to prepare.
You run the pipeline
You pick which of the 10 signals you want for this campaign. SOutreach runs them in order: identity → activity → fit → pain → risk → technographic → behavior → language → outreach drafts. Around 30 min for 400 companies.
You review the output
Open the result sheet. Every row is a company with 10 signals: who they are, what they do, what painpoints they carry, how they fit, whether it's worth writing at all. Plus a message draft based on those signals. You approve every message before sending. Always, no exceptions.
You start the sequence
You export to your sender (MeetAlfred, Apollo, Instantly, manual LinkedIn). 8-touch sequence designed in step 1, each message tied to the signals of that specific company. No copy-paste-to-everyone.
Weekly signal review with the team
During Beta we sit with you on Fridays for 30 min. Together we look at which signals correlate most with replies in your market, which are false-positive, how to tune the ICP. You iterate on data, not on intuition or gut feel.
Why intent data, not autopilot
AI autopilot in prospecting failed 95% of buyers in 2025.
UserGems reports 50-70% AI SDR churn within 90 days. 11x.ai got loud for the wrong reasons. We watched that play out and built the reverse: the engine reads 10 signals per company (input), you decide how to use them (output). Slower than the broken alternative. Works, though.