Lead-gen: the approach

Thesis

We hunt customers of existing support-agent builders — teams already running a support agent in production. Where they built it themselves, they own its quality, which is the regression pain testpath solves. We mine each platform’s customers (case studies, reviews, job posts) into a high-signal list. The platforms are in the candidate list.

Builder vs. managed

What decides whether a platform’s customers are worth targeting: who owns the agent’s quality. With a builder, the customer owns the agent — and its testing, which nobody’s doing. That’s our wedge. With a managed / turnkey service, the vendor owns quality, so the customer feels no pain and the vendor gatekeeps. Builders’ customers are targets; managed customers are noise. The teardown proves it: every builder ships testing, none handle stochasticity — so their customers can’t tell a real regression from LLM noise.

Targeting the builders

We don’t sell to the builders — they make the tool, not the production agent. We target their customers, starting with the largest, most reachable bases: Botpress and Voiceflow (broad SMB/mid-market, public case studies). Rasa is enterprise/slow; Agentforce and Cognigy are too big or absorbed. Pull each builder’s customers from case studies, G2, and job posts, then cold-email on the curiosity angle: they built it, they own quality, nobody’s testing it. Longer term, a builder partnership (integration or referral) reaches a whole base at once — but only after direct outreach has proven the pitch lands.

Building the list

Per priority builder, run the same pipeline (this is what the prospecting CLI automates, writing rows to src/artifacts):

1. Find accounts — companies that use that builder. Sources: the builder’s case-study / customer-logo pages, G2 / Capterra reviews (reviewers name their employer), job posts naming the platform (Greenhouse / Lever / Ashby / LinkedIn — “built our agent on Voiceflow”), chat-widget / tech-stack detection on company sites, and partner / “X chose Y” announcements.
2. Find the person — whoever owns the agent. Small company → founder / CTO; larger → a “conversational / applied AI” or “support automation” lead. Triangulate from LinkedIn, the company site, and the original job poster.
3. Find the email — verify via Hunter (guess the pattern, then check). Never send to an unverified guess; bounces wreck the warmed sending domain.
4. Record it — one datastore row per contact: company · builder · person · role · email · hook · source · status. The hook is the specific signal (e.g. “uses Voiceflow, hiring an AI engineer”).
5. Score + dedup — keep only ICP fits (customer-facing agent in prod, can pay ~$500/mo, reachable); drop duplicates across sources.

Quality over volume: a vetted 10–20 contacts/week with a real hook beats a 1,000-row blast. The hook is what makes the cold email land — it ties straight back to the teardown (“your Voiceflow eval scores one run; it can’t tell a regression from noise”).

Target accounts (first pass)

Found manually via the pipeline above (vendor case studies, chat-widget detection, job posts, engineering blogs). Verify each before outreach — case studies are vendor-published, widget detection can be stale, job posts rotate. Emails are the next step.

Channel leads (agencies that deploy these builders repeatedly — one relationship reaches many SMB clients): Helpline Hero (Bilal, founder), Parkfield Commerce, Streamline Connector.

Best first cuts (SMB/mid, technical, clear in-house ownership): Central, Firecrawl, Gusto, PostHog, Socure, eSnipe, Roam, Eddy Travels, Tia Health, Able. Monzo, JetBrains, Hopper, StubHub, Turo, and Trilogy lean enterprise — likely already have internal eval, so deprioritize.