Finding B2B clients by cold email: AI writes each one from facts about the recipient's company
When sales reps write cold emails themselves, most of their time goes on research: who makes the decisions at a company, how the business is doing and what to write to them about. In this setup Apollo finds the right people, Clay collects facts about each company, AI writes an email to each person, Instantly sends it and replies go to amoCRM. The sales reps keep the quality checks and the conversations.
A good cold email takes time to prepare
For this estimate, take a B2B company that sells services to mid-sized businesses. A deal takes several months, and the decision is made by the CEO, COO or IT director.
An email that says nothing about the recipient's company reads like a mass mailing. So a sales rep first looks for a reason to write: checks the company's revenue, reads its news, looks at new job openings and appointments. Then they write an email to that particular person. The rep works as a researcher and a copywriter at once.
Even if the research and one email take just 10 minutes, a rep will write at most 240 emails in a 40-hour week and have no time left for meetings. Writing more means hiring and training more people.
Five steps from a list of companies to a lead in the CRM
The idea is simple: take the repetitive work off the sales reps and leave them what they were hired for, checking quality and holding meetings.
Senior decision-makers who fit the customer profile, for example the CEO, COO or IT director of companies with 50–500 staff. The list is split by industry.
Revenue from SPARK, company news, the technologies on the company's website, new job openings: the facts each email is built on.
GPT-4o writes 3 versions of the email from 4–7 facts, and Claude picks and checks the best one (both through ProxyAPI.ru). Subject lines are A/B-tested.
Eight warmed-up mailboxes take turns. If there's no reply, four more emails follow. Bounced addresses are removed and emails are checked for spam.
Replies go to amoCRM tagged by tone. Interested people go to a sales rep; those who decline are unsubscribed.
Apollo finds the right people
Apollo is a database of companies and contacts. In it I pick the senior people who fit your ideal customer profile and split the list by industry: manufacturing, wholesale, retail, B2B services.
Clay collects 4–7 facts about each company
Clay adds data to each contact: revenue from SPARK, recent company news, the technologies on its website via BuiltWith, open job postings (a sign of growth) and mentions in the trade press. The goal is 4–7 specific facts that let the AI write to this particular person.
Each email has a specific reason to get in touch
GPT-4o quickly writes 3 versions for each contact (through ProxyAPI.ru, so it can be paid for with a Russian card). Claude picks the best version, removes stock phrases and checks every claim against the collected data. The finished email mentions a specific event at the company: a new branch, a new head of digital, a product launch.
Instantly: warmed-up mailboxes and follow-ups
Emails go out from 8 mailboxes on different domains. Before the campaign starts, the mailboxes are warmed up in Instantly for two weeks, building a history of ordinary correspondence. Then they take turns, and each sends 35–45 emails a day to stay out of spam folders. Subject lines are A/B-tested. Anyone who doesn't reply gets four follow-ups, after 3, 7, 14 and 28 days. Bounced addresses are removed automatically.
Replies are sorted by tone and go to amoCRM
A small language model reads each reply and tags its tone. If the person is interested, a lead with a "ready for a call" tag appears in amoCRM and goes straight into a sales rep's queue. A neutral reply gets the tag "send more materials". If the person declines, they are removed from the sequence.
A dedicated service for each step
A database of companies and contacts, filtered by the ideal customer profile
Adds data from SPARK, news feeds, BuiltWith and job postings: 4–7 facts per contact
Cold-email service: mailbox warm-up, rotation, follow-up sequences, bounce handling
Quickly writes 3 email versions and a subject line from the collected facts
Reviews the best version: stock phrases, facts, tone
Access to OpenAI and Anthropic models, paid with a Russian card
Company revenue and financial statements by tax ID
The technologies behind a website: CMS, analytics, hosting
Qualified leads, tone tags and the sales reps' queue
- In my missed-call auto-reply, a language model writes the reply around a specific customer's order. The system is not allowed to promise anything the data doesn't support.
- I find out what a website is built on and which services it connects to with my own tool for researching websites from public sources. In this setup BuiltWith does the same step inside Clay.
What to expect after launch
The figures are worked out for the setup described above: eight mailboxes, 35–45 emails a day from each, five working days a week and up to five emails per recipient. I'll redo the numbers for your company during the audit.
eight mailboxes at 35–45 emails each, follow-ups included
each gets up to five emails: the first one and four follow-ups
what 300 personal emails would take by hand at 10 minutes each
out of 300 new recipients, every percentage point of replies is about three people
The main change is the sales reps' job. The services take over the research and the writing. The reps spot-check emails and hold meetings with the people who reply.
How many replies turn into leads depends on the industry, the customer profile and the reason for writing. The first month of sending shows which subjects and reasons work best. After that, the cost per lead can be worked out from your own numbers.
You can also test a new industry without hiring: it takes new filters in Apollo and a fresh data set in Clay.
How long the launch takes
4 weeks of setup and 2 of mailbox warm-up; sending starts once the warm-up is done
I show the team how to run the email sequences, change the filters and handle replies, so they can carry on without me
- Together we describe your ideal customer, and I set up the Apollo filters to match
- I set up Clay to collect 4–7 facts about each company for the AI to write from
- I write the GPT-4o prompts and set up review in Claude and A/B tests for subject lines
- I prepare the mail: 8 mailboxes, a 14-day warm-up, follow-up sequences
- I connect Instantly to amoCRM: tone tags and a queue for the sales reps
For the cost of setup and monthly subscriptions, see the Pricing page.
When automated client search makes sense
The setup suits companies that sell to top managers of mid-sized and large businesses, with deals that take more than a month and bring over ₽500k. Typical examples:
- → B2B software sold at ₽500k+ a year: a well-defined customer profile and clear reasons to buy
- → Consulting and agencies: digital transformation, automation, custom development
- → Business services: legal, audit, valuation and HR consulting for mid-sized companies
- → Industrial technology (IoT, MES, WMS), where the person to reach is the production or IT director
- → Financial products for business: card acquiring, factoring, business banking
- → Any B2B service where sales growth is held back by how many emails the reps can write
Related cases and ready-made solutions
Conversions go to the ad platforms straight from the server, and the team's report builds itself
The Keitaro tracker collects clicks and conversions on your own server. An nginx layer with Lua removes…
Weekly client reports that put themselves together
Every Monday a workflow in n8n, an automation tool, pulls the numbers for all eight clients from Facebook…
From speech to a finished ad: transcription, translation, a new voice-over and lip-sync
Whisper transcribes the speech, GPT-4o translates the script and rewrites it for the market, ElevenLabs…
Start with the 5,000 ₽ audit: a written report and a payback estimate
I'll tell you what to automate first, when it will pay for itself and whether your task needs AI at all. Sometimes it doesn't.
Or just ask a question. During business hours I reply within 2 hours