1. Home
  2. Work
  3. 24 interested replies for a consultancy using an AI outreach

AI outreach system in Lebanon: 24 interested replies from 7,279 personal emails

I'm David Geha, an AI consultant in Beirut, and I built an AI outreach system for my own consultancy. Between 2026-05-25 and 2026-11-10 it ran 247 times, screened 26,170 businesses, sent 7,279 personalised first emails and brought 24 interested replies from real people. This page shows what the system does, what it replaced and where its limits are.

one-person AI consultancy (my own) · Beirut · Delivered 2026-05

SupabaseVercelClaude CodeGitHubAI Email AgentsAI CRMsOutreach SystemsSEO & GEOAI ReceptionistsAI Web Design
Businesses screened automatically
26,170
Personalised first emails sent
7,279
Interested replies from real people
24

This is my own consultancy, so there is no outside client here. I use the system to find businesses that fit my work and to write to them. It has run every day since May 2026. The numbers below were measured from 2026-05-25 to 2026-11-10 over 247 runs.

Outreach by hand took hours for every handful of leads. I had to find businesses that fit, check who actually decides, and write a first email that read as if a person had researched the company. Because of that, outreach happened in bursts and then stopped. A channel that only runs when I have spare hours does not bring in a steady flow of conversations.

  • Source businesses against an ideal-customer profile, using free map and web data from Google Maps and Overture.
  • Drop every business that does not fit the profile before any email is written.
  • Find the real decision-maker at each company and prove who it is.
  • Write a personal first email from what the system read about that company.
  • Send each email from my own domain on a schedule.
  • Log every reply so I can see who answered and what they said.

The system runs on Supabase for data, Claude for reading and writing, Google Maps and Overture data for sourcing, and Brevo for sending. Each run screens businesses, drops the ones that fail a check, and writes and sends emails for those that pass. A lead with no proven decision-maker or a failed check is dropped and never sent a weaker email. In total 18,891 leads that failed a check were dropped and never emailed. Across the period, 10,290 emails were delivered, counting follow-ups, and there was 1 spam complaint in 10,290 delivered. The 24 interested replies came from real people. I read them myself and take it from there. The system finds and writes to prospects; it does not hold the conversation or close the work.

Supabase, Claude, Google Maps and Overture data, Brevo. Built in the client's own accounts and handed over running.

MetricAfter
Businesses screened automatically26,170
Personalised first emails sent7,279
Emails delivered (with follow-ups)10,290
Interested replies from real people24
Spam complaints1 in 10,290 delivered

Timeline: Running every day since May 2026.

This is one of the systems described on AI solutions in Lebanon; the audit-first process is on the how an engagement works page.

Are these results from a real client?

No. They come from my own consultancy, where I run the system for my own outreach. I measured them from 2026-05-25 to 2026-11-10 over 247 runs. I can build the same kind of system for another business, but this page does not claim client results.

What happens to leads that do not fit or fail a check?

They are dropped and never emailed. If the system cannot prove who the decision-maker is, or a check fails, it does not send a weaker email instead. In this period 18,891 leads were dropped this way.

Will automated outreach get my domain marked as spam?

In my case there was 1 spam complaint in 10,290 delivered emails. The system sends from my own domain on a schedule, and each first email is written from what it read about the company. I can only speak for my own results, and I would look at your list and domain before promising anything similar.

How much would a system like this cost?

Fixed-price projects start from $1,000, and the exact price is set after the audit. The audit looks at who you want to reach, what data exists and how replies would be handled. I would rather price a real scope than guess.