The outbound system
your team can't build.

AI-native companies are too new and too narrowly funded to show up cleanly in Apollo or ZoomInfo. AI Reserve needed all of them anyway.

• i.Sourced from three places at once — SEC Form D filings (a free Crunchbase alternative), ~50 scraped VC/PE portfolios, and Apollo — to catch the companies any single database misses.
• ii.Resolved, deduped and size-gated every company down to one clean universe by domain.
• iii.Scored each one with an LLM against a custom rubric, then split into tiers — qualification no keyword filter can do.
• iv.Result: 781 qualified target companies across the three sources — and a reusable engine they own, not a one-off list.

Pattern: multi-source fusion — SEC filings + VC portfolios + Apollo, deduped and LLM-scored into one universe.

PE firms and investment banks needed targeting data — AUM, portfolio companies, sectors — that sits behind premium subscriptions they didn't want to buy.

• i.Estimated each firm's AUM by matching Crunchbase against SEC Form ADV regulatory filings — no premium data subscription.
• ii.AI agents extracted portfolio companies and sectors from PitchBook profiles.
• iii.An enrichment waterfall found the decision-makers; a 3-step sequence ran in Email Bison; positive replies routed straight to Slack.
• iv.Result: a 2.5% reply rate with half of them positive — into PE and investment banking, a segment where most cold outbound never clears 1% — running 12 campaigns/month, with meetings booked at firms like Lex Capital Group and Aurum Capital.

Apollo and ZoomInfo barely list local contractors. A US home-services company needed to reach them anyway.

• i.Scraped and deduped every business in their target metros from Google Maps — the local contractors that don't show up cleanly in B2B databases because they're small, local, and not "in the dataset."
• ii.Qualified to their exact ICP with a config rule set — size, reviews, category, claimed status, chains filtered out.
• iii.Found each owner / decision-maker from the business's own website and Google — a real person, not a generic info@ inbox.
• iv.Generated a {business_type} token for every lead, so the sequence personalized to whether the recipient was a roofer, HVAC contractor, or plumber — personalization-ready data, not raw leads.
• v.Result: a 9% reply rate, half of them positive — ~4.5% of everyone contacted into a real sales conversation, against the 1–2% norm for local cold outbound. They own the engine: re-point the config at a new metro or vertical and it runs again at ~98% owner-match accuracy.

Pattern: config-driven engine — deployable to any geography or vertical by changing the config, not the code.

An agency had campaigns underperforming across multiple clients — and no systematic way to tell whether the problem was deliverability, copy, offer, or targeting.

• i.Connected directly to Instantly, Smartlead or HeyReach via API and pulled the full account — the real data, not a screenshot.
• ii.Pinpointed the actual bottleneck — sending/domain health, copy, offer, or list targeting — graded against a fixed rubric and the client's own GTM.
• iii.Categorized the reply pool and ran inbox-placement checks.
• iv.Result: a prioritized fix list they used to turn around live campaigns across the whole roster — diagnosis in hours, not a week of guessing.

Pattern: API-driven audit — connects directly to Instantly / Smartlead / HeyReach via API rather than scraping the UI. More reliable, faster, and runs against the actual sending data.