The problem M&A advisors win mandates person by person. Mergero, a cross-border M&A advisor, wants to scale mandate origination, but the first touchpoint is moving: owners thinking about succession and acquirers looking for off-market targets increasingly ask ChatGPT, Perplexity or Gemini first. Whoever the answer names gets the first call. We asked 40 real owner and buyer questions (Claude with web search, one run): Mergero was named in 0 of 40 answers. Its advisors have the expertise, but no time to write.
What we built Mergero Engine, a working internal tool that turns advisors' spoken expertise and Mergero's own data into content that AI assistants can cite, then into manual, advisor-led outreach. - Talk to blog: press one button and talk. An AI voice interviews the advisor and asks follow-up questions based on what was just said. - Review: every sentence is traced to the advisor's words or a data row. Anything the AI added is flagged and blocks publishing until a human accepts or removes it. - One article becomes six languages (fi, en, sv, de, it, fr) plus a LinkedIn post, carousel, newsletter, letter to owners and a video script. - Meeting insights: upload internal meeting notes (Word, PDF, text), which are anonymised on the way in. Ask questions by voice, e.g. "What matters most to family offices when choosing targets?", and turn recurring questions into article topics. - Seller readiness tracker: a shortlist built from Finnish public registry data (16,868 IT companies → 772 established employers), with timing signals, a letter-to-call workflow and a buyer view that shows only anonymous counts. - Search demand per market (DataForSEO), with an optional SE Ranking connector for trending terms.
How it works, and what's AI about it - Claude drafts the article from the interview transcript only, then a second pass checks every claim against the transcript and data. A quote counts only if its words really occur in the recording. - ElevenLabs gives the interviewer a voice and transcribes the answers (Finnish included). - Confidentiality rules run as code before an LLM pass: no client or company names, euro amounts only as ranges, no group smaller than five, no identifying combinations. - Published pages are built to be cited: answer first, dated, a named author, structured data (JSON-LD), hreflang and llms.txt. - Humans stay in charge: nothing is posted or sent automatically, and willingness to sell is never predicted. An advisor asks, in person. Stack: Next.js, Postgres (Drizzle), Claude, ElevenLabs.
What's next - A public succession-wave report per industry and region: the number nobody else has. - A measurement loop: re-run the 40-question AI-answer baseline to see whether Mergero starts to appear. - An 8–12 week pilot in Finland and one DACH market with Mergero's advisors.
