Automated Deal Sourcing for Venture Capital

The Challenge

Deal sourcing in venture capital is legwork. Working through accelerator batches, screening startups against the fund thesis, creating contacts, following up. One analyst can do this - but not at the breadth and not at the cadence needed to be early in a deal.

Our client, a Hamburg-based Food & AgTech VC, came with a tightly scoped pilot: automate the evaluation of accelerator programmes. Around 30 hours of work, two weeks. Over the following months, that pilot grew into load-bearing infrastructure.

Our Approach

Sourcing pipeline. A weekly run covers 40 accelerators, extracts current batch participants and scores each startup via LLM against the fund’s investment criteria. Results land in the CRM, in the data warehouse, and as a curated top-ten digest in the team chat. What used to happen as a side task between meetings now happens by itself on Monday morning.

Chief-of-staff system. In parallel we extended the managing partner’s existing assistant system: classification of incoming mail, daily briefings, automated meeting preparation, synchronisation between CRM and data warehouse, and an approval workflow for generated draft replies.

Service split instead of monolith. We extracted the external data sourcing layer into its own service - incrementally, in individually reviewable slices rather than one large rewrite.

Operations that surface themselves. A fully infrastructure-as-code managed development environment mirrors production. A dashboard shows service health, model spend and error rates - important for a system whose running costs depend directly on usage.

The Results

A two-week pilot became the fund’s sourcing infrastructure. The analyst no longer spends time compiling batch lists, but on the ten startups that survived the pre-screen.

The build-out ran deliberately in slices. Each extension was deliverable and verifiable on its own - no big bang, but infrastructure that grew alongside the trust placed in it.