An outbound engine for a vertical SaaS

RepairDesk sells to phone, computer, jewelry and other repair shops, most of them small and independent. They rarely appear in lead databases, so research was manual, slow, and rarely reached the shops most likely to buy.

• Find. An internal lead-generation app, built with our developers to a spec my team shaped, searches local listings ZIP by ZIP for each vertical and filters out chains and franchises. My team runs it and decides its upgrades.
• Enrich. PhantomBuster and Outscraper add phones and socials, owner names are added by hand with AI help, and a third-party enrichment and scoring tool runs when a lead is assigned.
• Verify. Every email is checked before sending. Rebuilding deliverability early cut bounce rates by about 80%.
• Sequence. Vertical-specific sequences of five emails over ten business days, moved from Zoho Campaigns to Instantly as volume grew.
• Hand off. Once a lead engages enough, it leaves the sequence, is assigned to an account executive with an alert, and starts a sales cadence. Enrichment to first contact takes under five minutes.
• Recycle. About 10,700 lost leads re-segmented into three sequences by why they were lost.

For lost leads, the system reads each lead's CRM notes and call summaries and drafts personalized lines with an LLM. A person approves every lead before it is sent. Getting the output right took several rounds:

• It wrote from the rep's point of viewPain points came out as sales notes. The prompt now asks for the shop owner's perspective.
• It invented specificsThin notes produced confident, wrong details. Rules now tie every claim to something in the record.
• Personal details leaked inSocial media notes appeared in emails. Personal information is now excluded explicitly.
• Everything sounded the sameAn example sentence in the prompt became the default answer. Removing it restored variety.
• Some fields were never reliableTwo fields kept drifting in tone, so I replaced the AI with fixed copy chosen by vertical.

Outbound became a repeatable pipeline rather than a research task, reaching 10,000+ prospects directly and surfacing 450+ existing stores ready for an upsell.