New business clients, and a human approves every message
A maker of corporate merch sells to other companies. We built them a system of three agents: the first finds the person who decides at each company and writes to them where they reply; the second runs conversations with clients who write in themselves; an AI assistant helps the whole team. Every first message goes out only after a manager's OK.
- Client
- A maker of corporate merch and brandingsells to other companies
- Service
- emilus. Leadgenfinding new clients among companies
- Status
- In production at the clientevery message waits for a manager's approval
- Before → After
- Companies that become clients: 1.3% → 3–4%20–25% reply to the first message
No message goes out on its own. A manager reads and approves every message before it is sent.
per client data the client gave us this numberestimate we calculated it; the assumptions are next to it
Where the client started
The company makes corporate merch and sells it to other businesses. The team looked for new business clients by hand: 2,000 companies in 5 months, and 26 purchases. Inbound messages on Instagram and Telegram arrive as text, photos, voice notes and files. A manager answered each one personally, dug up examples of past work, passed the details on for the invoice and tracked payment across several systems. Nobody replied in the evening or at the weekend, and a client's files got lost somewhere in the chat history.
companies in 5 months, worked by hand
per client datapurchases from those leads
per client datarevenue from those purchases
per client dataspent on this work
per client data26 purchases from 2,000 companies: 1.3% became clients, and behind each one sat hours of manual research and back-and-forth.
The client's main conditionNo message goes out without a human's approval.
What the system does
- Researches every company in the list on its own: website, job posts, LinkedIn. It finds the decision-maker and gives the company a score, so a manager doesn't lose half an hour to manual searching.
- Writes to a specific person in a specific channel: a personal email, a shared inbox, LinkedIn or Telegram, with the company's real projects in the text. It keeps sending reminders until the person replies.
- Shows every message to a manager before it goes out. The manager reads, edits and approves it, so the company's reputation stays in its own hands.
- Handles Instagram and Telegram conversations like an account manager, with new and existing clients and by the company's rules. It finds out what each client needs and passes on only what needs a person.
- Understands the client's photos, files and voice notes and keeps every material in one company card, by date, so the order history doesn't vanish into the chat.
- Quotes prices from the company's own calculator and picks examples of past work to fit the request. One click turns a manager's correction into a rule.
- Keeps in touch and sells again to existing clients: a feedback request after delivery, holiday greetings, a reorder. It stops as soon as the client replies.
- Sends the approved mockup to the production team's Telegram group and keeps reminding the people responsible until they answer, so no order gets stuck between departments.
- Issues the invoice in FinMap, the company's finance app, spots the payment and confirms it to the client in the same conversation, so the manager doesn't have to check bank statements.
- Works as an AI assistant for the whole business. It answers any question about clients, conversations, orders and figures, pulls together analytics, helps train new staff and carries out actions in the system after one confirmation.
How it is built
A web app for the manager on desktop and phone, a log of every action, and daily and monthly spending limits. When finding clients, the AI works in three roles: researcher, critic and copywriter. Two more agents work alongside it. The inbound agent talks to new and existing clients, and the AI assistant works with all the data in the system for the whole team. A human stands between the AI and the send button.
- Agents
- finding clientsinbound agentAI business assistant
- Channels
- emailLinkedInTelegramInstagram
- Data sources
- company websitesjob sitesLinkedInpublic registersemail verification
- Integrations
- FinMap: invoices and paymentsproduction team's Telegram group
- Notifications
- Telegram bot for the manager
How it works
- managerCompany listUploads companies or contacts
- AI researcherProfile and scoreDecision-maker, sources and a score
- AI writerMessage for the channelEmail, LinkedIn or Telegram
- managerA human confirmsReads, edits and gives the OK
- clientReplyThe reminders stop
- inbound agentConversationTalks with the client on Instagram and Telegram
- agentPrice and examplesFrom the company's calculator
- FinMapInvoiceCreated with no manual entry
- systemPayment confirmedThe client sees it in the same chat
- agentRemindersFeedback, holidays, reorders
Results and payback
of people we wrote to for the first time reply: on average across email, LinkedIn and Telegram
per client dataof the companies we wrote to become clients
per client datacompanies researched: who decides there and how good a fit they are
per client dataProject payback: ≈5 months per client dataThe same contacts now bring 2.5–3 times as many clients, at an average order of 32,885 UAH. One system of three agents: finding clients, replies to inquiries and a business assistant.
What's live
- live
A company profile, a message to the right person and the manager’s approval
- live
Campaigns by email, LinkedIn and Telegram
- live
The inbound agent and an AI assistant for the whole team
A similar task in your business?Solomiia, Pavlo's AI assistant, will ask about the details and prepare your conversation with him.
emilus. Leadgen
Finding clients among companies: a list of companies, a profile of each, and a message to the right person where they reply. No message goes out without a human's approval. Diagnostic $300 → pilot $2,500–3,500 → system $6,000–15,000; we name the exact system price after the diagnostic.