Cold pitches get ignored. Relationships don't.

Relationships bring work and referrals. Cold pitches mostly get ignored.

But finding the right people, researching them properly and remembering every conversation doesn't scale for one person.

We wanted the research and the memory of a full-time assistant, with every email still personal and every fact true.

Research first, then a personal introduction

The Relationship Manager introduces the founder to a few well-chosen people every night and keeps track of every conversation. It was built for our own companies, BinaryBits and TheDataHQ, and it is used every day.

The product screen, shown with test data. Real screens use the same layout.
📋
A short list without AI
Suitable people are picked from data first, which keeps costs low. The agent then checks cheaply whether each person is still in their role and has something current to talk about.
🔍
Deep research only where it counts
A research card with sourced, dated facts about the person and their company, only for the people it will write to.
✅
A reviewer who checks the facts
It opens the sources and confirms facts about both the person and the company.
♻️
Research reused
Company research is saved and never repeated.
✉️
No-pitch introductions
A genuine question or a small useful offer (a sample, a trial or a demo idea), never a sales pitch. Only one gentle follow-up.
🧠
Memory and stages
What each person said and cares about, with stages from introduced to talking, warm, opportunity and client.
🔔
Check-ins only with a real reason
And a message to the founder when someone becomes warm.

From a short list to a real conversation

AI agentAutomated stepPersonData or serviceShort listpicked from data, no AIQuick checkstill there? anything current?Deep research cardsourced, dated factsIntroductiona question or a useful offerRevieweropens sources, checks factsEmail Managerpaced sendingConversation & memorystages and notesCheck-insonly with a real reasonsuitable peoplestill relevantresearch carddraftchecked emailrepliesa real reason
Cheap checks first, deep research only for the people it writes to, and a reviewer that confirms every fact.
1
Short list
Suitable people are picked from data, without AI, so no money is spent on people who are a poor fit.
2
Quick check
An agent checks cheaply whether each person is still in their role and has something current to talk about. If not, they are skipped.
3
Deep research
For the people it will write to, the agent builds a research card of sourced, dated facts about the person and the company. Company research is saved and reused.
4
Introduction
The email starts a real conversation: a genuine question or a small useful offer, never a pitch.
5
Review
A second agent opens the sources and confirms the facts about both the person and the company.
6
Paced sending
Checked emails go to the Email Manager, which sends them one at a time at a human pace.
7
The conversation, remembered
What each person says is remembered, their stage is tracked, and check-ins happen only when there is a real reason.

Research depth, balanced against cost

Each step uses only as much AI as it needs. The agents run on Anthropic's Claude, through Claude Code skills, and work comes up only when the conversation opens the door, with the founder's OK.

WhoWhat it does
Short list (automated)Picks suitable people from data first, without AI, which keeps costs low.
Quick checker (AI)Cheaply checks whether each person is still in their role and has something current to talk about.
Researcher (AI)Builds a research card of sourced, dated facts about the person and their company, only for the people it will write to.
Writer (AI)Writes a no-pitch introduction: a genuine question or a small useful offer.
Reviewer (AI)Opens the sources and confirms the facts about both the person and the company.
The founder (person)Hears when someone becomes warm, and gives the OK before work is discussed.
Cost

A short list without AI, a cheap check first, deep research only where it counts, and company research saved and reused.

Quality

Every fact on the research card has a source and a date, and a second agent checks them before anything is sent.

People stay in control

Work comes up only when the conversation opens the door, and only with the founder's OK.

Memory

What each person said and cares about is remembered, with stages from introduced to talking, warm, opportunity and client.

Technologies used

Our own Python code, web research and AI agents built on Anthropic's Claude. Checked emails are sent through our Email Manager.

Python Claude agents Web research

Challenges and how we solved them

❌ Challenge
Old contact data: some people had moved on.
✓ What we did
A cheap "still there?" check before any research.
Result: No emails to people who have left.
❌ Challenge
Research too thin at first.
✓ What we did
A research card with at least four sourced facts about the person and four about the company.
Result: Personal, verifiable emails.
❌ Challenge
Cost.
✓ What we did
Deep research only for the people it writes to, and company research reused.
Result: Depth without waste.

Personal, verifiable introductions

Daily
In daily use, every night
Skipped
People who had changed roles, before any research
Checked
Facts about both the person and the company

In its first runs it skipped people who had changed roles and wrote fully sourced introductions for the rest. The reviewer confirmed facts about both person and company. Business results will be added here once they are measured.

Last updated: November 2026

What this means for you

The same research-first approach works wherever relationships matter more than volume.

  • Founder-led networking. A few well-chosen introductions a day, each personal and researched.
  • Account research for sales. Research cards with sourced facts about people and companies, ready before the first conversation.
  • AI-assisted partnerships. Finding and starting conversations with the right partners, without pitching.

For client systems we build on PostgreSQL, MySQL, MongoDB or Redis, and cloud data warehouses where needed, chosen for the project's scale and team. See our AI agent development services.

Frequently asked questions

Isn't this just cold email?

No. It doesn't pitch. It starts a real conversation with a specific question or a useful offer, and work comes up only if the person wants it.

How personal is it?

Every email is built from a research card of sourced facts, checked by a second AI.

How do you keep costs down?

A short list without AI, a cheap check first, deep research only for the people it writes to, and saved research reused.