01 The problem
A private capital firm wanted what every firm that lives on relationships wants from outbound: more qualified conversations, without sounding like a mail merge.
The bar is higher in capital than almost anywhere else. A construction company that sends a clumsy cold email loses a prospect. A capital firm that sends one loses standing, because the people it writes to talk to each other, and because the message itself is evidence of how the firm thinks. Volume without research is not just ineffective in this market. It is actively expensive.
So the firm faced the standard capacity ceiling, hours of research per prospect against a finite week, with an extra constraint bolted on: the floor for message quality was wherever a partner would be comfortable signing their own name.
02 What had to be true first
Three constraints were fixed before the build, and they will be familiar if you have read the Breez case, because they are the same discipline pointed at a more conservative market.
Research before outreach, every time. No message leaves the system without a per-lead analysis behind it: who this prospect is, what their situation looks like, why a conversation makes sense. In a reputation market, the strategy layer is not a feature. It is the licence to operate.
The firm's data lives in the firm's instance. The engagement runs on its own isolated database, with authentication handled separately. Pipeline state, lead records and campaign history are the firm's asset, in the firm's instance, under our standard isolation architecture. No commingling with any other client.
Measured before switched on. The firm agreed baselines in writing before deployment: what prospecting costs in hours today, what current output looks like. That is the same method behind our guarantee, and it is why this case can be honest now and numerical later.
03 What we built
The engine is our outbound pattern, adapted to the firm's market: one pipeline, raw lead in, researched outreach out.
- Lead sourcing. Continuous automated sourcing against the firm's target profile, replacing one-at-a-time manual list building.
- Enrichment. Contact and company data gathered and merged into one verified record per lead, so everything downstream reads the same truth.
- Per-lead strategy. A generated analysis for every prospect, the layer that separates personalization from a template with a first-name token.
- Personalized outreach. Messages written from the strategy layer, with every line traceable to something true about the prospect.
- Reply detection. Responses caught and routed to a person the moment they land. The machine never negotiates; in this market especially, a reply is where automation ends and the firm begins.
Around the pipeline sits a client dashboard, built from our client platform template: campaigns, leads and workflow runs visible to the firm in real time, on its own instance.
04 The technologies behind it
- Next.js for the client dashboard, built from the same hardened template every client platform of ours starts from.
- Supabase holds pipeline state: one record per lead, one isolated instance for the firm.
- Modal runs the execution scripts on serverless schedules: sourcing, enrichment, detection.
- Claude powers the strategy and personalization layer, the judgment-shaped work the pipeline is built around.
Full reasoning for each choice lives at /stack.
05 Where the numbers are
The infrastructure is built and the system is in deployment, pending go-live. We publish this case now, without outcome numbers, for a simple reason: we would rather show you what a deployment actually looks like than wait and show you only the wins.
What is true today: the pipeline above exists and is verified against the build. What is not yet true: measured results. The baselines are agreed, the before state is recorded, and first measurements land this quarter. When the numbers exist, they appear here with their measurement basis, the same way every measured case on this site carries one. Until then, the only honest value for the results section is the date we expect them.
06 What the team learns and owns
The firm's team is trained on the system as part of deployment, not after it. They own the directives that set tone, targeting and follow-up rules, in plain text, editable without a vendor ticket. They learn how the strategy layer reasons, so they can audit it, and how to read the pipeline dashboard, so the system is never a black box that sends things in their name.
That last point is the whole engagement in one sentence: in a reputation market, a firm should be able to defend every message its systems send. After handoff, this firm can.
07 Related work
The pattern is productized as the outbound engine, and the measured version of this story is the Breez case: same pipeline shape, different market, numbers already in. For the sector view on AI inside investment firms, see AI for private equity. If you want your own baseline agreed before anything is built, start with the free audit.