Real case · first-week results
55%same-day response rate

A dormant base of 1,800 contacts, awake in one week.

A European specialised technical recruitment consultancy had years of contacts piled up on LinkedIn, an outdated CRM and zero hours for manual follow-up. This is what happened when we built them an AI assistant.

11/20
replies on launch day (55%)
20
candidates validated in one week
51
contacts recovered from old threads
700+
CRM records audited and fixed

The problem was never the data. It was the hours.

No human has time to review 1,800 records, write 1,800 personalised messages and keep 26 conversations going at once without dropping the ball. Every passing week, that base was worth less.

What we built (in one week)

The starting point could not have been simpler: the official LinkedIn data export — that ZIP any user can download from their own account. No scraping, no shady tools. Claude read it whole from the terminal (1,800 profiles, 18,569 messages, years of history) and everything else grew from there:

A transparent WhatsApp assistantReaches out with pre-approved personalised messages, always introduces itself as what it is (an assistant), holds the conversation, verifies requirements and asks for the CV. Checks replies every 15 minutes during work hours.
A CRM that updates itselfEvery reply lands on the contact record instantly. A glance at the record before calling someone saves half the conversation.
Mining the archiveIt re-analysed 18,569 old LinkedIn messages and recovered 51 phone numbers and 5 emails lost in years-old conversations.
An audited, certified databaseOver 700 records verified, corrected or enriched. The base went from questionable to reliable.
Structured profiles + matchingEvery incoming CV becomes a structured profile that a matching engine scores against each new vacancy, with evidence.

The panel the whole team sees

None of this lives in one person's head or in a lost spreadsheet: everything lands in a shared, always-current web panel with five tabs:

  • Cold pool — all 1,439 source contacts, each with its status and priority.
  • Candidates — 135 living profiles: skills, tech stack extracted from the CV, and the dated history of every conversation.
  • Technical match — for each vacancy, a candidate ranking scored out of 100 by the AI, with a transparent rubric and the evidence behind every point.
  • Request a match — self-service: anyone on the team pastes a vacancy and the engine returns the ranking, no gatekeeper needed.
  • Discarded — nothing gets deleted: every discard is recorded with its reason.

The rules that make it trustworthy

  • It never pretends to be a person: it introduces itself as an assistant, always.
  • It never contacts anyone outside an approved list, and never replies twice to the same message.
  • The moment anything sensitive comes up (rates, complaints, personal matters) it goes quiet and alerts a human instantly.
  • The AI does the volume. A person makes the decisions.

Tooling cost: one AI subscription. No new software to learn: everything runs through WhatsApp and a web page. And it worked so well that another team in the company wired the panel into their own database via API — the kind of adoption you cannot fake.

What does this have to do with you?

This loop (reach out, converse, verify, record, escalate) is the same one a real-estate agency with cold leads needs, or a clinic with patients who never come back, or any business with a list it plans to work on someday.

Tell us which list of yours is asleep WhatsApp us