Learn to build a solo AI consulting agency using automation blueprints, real prompts, and low‑cost tools — then decide if you want the ready‑made package.
Running an AI consulting agency often feels like juggling client work, prospecting, and delivery while trying to keep costs low. After reading this guide, you’ll have a repeatable system for finding leads, crafting proposals, delivering AI automation projects, and invoicing — all powered by a few AI‑driven blueprints.
How do you find qualified leads without spending hours on LinkedIn?
Most solo consultants start by scrolling LinkedIn, sending connection requests, and hoping for a reply. That approach burns hours with little return. Instead, I built a lightweight scraper that pulls public profiles matching a keyword and exports them to a CSV.
The tool I use is Apify with its LinkedIn scraper actor. You sign up for a free Apify account, grab the actor, and feed it a simple JSON input. The script below shows the core logic — nothing fancy, just a request loop.
import requests
from apify_client import ApifyClient
def scrape_leads():
client = ApifyClient("YOUR_TOKEN")
run = client.actor("apify/linkedin-scraper").call(run_input={
"searchQuery": "AI automation consultant",
"maxResults": 50
})
for item in client.dataset(run["defaultDatasetId"]).iterate_items():
print(item["profileUrl"])
if __name__ == "__main__":
scrape_leads()
You run this once a week, get a list of 30‑50 URLs, then feed those into a cold‑email pipeline. The total cost? Apify’s free tier gives you 10 000 compute seconds per month, enough for dozens of runs. If you need more, the $49/mo plan adds plenty of headroom.
One gripe: the LinkedIn actor sometimes returns duplicate profiles because it treats variations of the same URL as distinct entries. I had to add a deduplication step in Python — annoying, but fixable.
What most guides get wrong about AI proposal writing
Many tutorials tell you to copy‑paste a generic template and fill in brackets. The result sounds robotic and fails to show you understand the client’s stack. I realized the missing piece is a quick diagnostic prompt that forces the AI to talk about the client’s actual tools.
🤖
Recommended Reading
AI Side Hustles
12 Ways to Earn with AI
Practical setups for building real income streams with AI tools. No coding needed. 12 tested models with real numbers.
Here’s the prompt I use, wrapped in a Make.com module that pulls the lead’s website text via a simple HTTP fetch:
You are an AI consultant. The prospect’s website says: "{{website_text}}".
Based on that, list three specific AI automation opportunities that would save them time or money. For each, give a one‑sentence benefit and a rough effort estimate (low, medium, high).
The output is concrete, not fluff. I then paste those three points into a Google Doc template, add a price range, and send it as a PDF. This step takes under five minutes per lead.
My love: the Make.com HTTP module returns the raw HTML, and I can strip tags with a built‑in text parser — no extra code needed. It feels like having a junior analyst on standby.
A concrete lead‑gen scraper built with Python and Apify (example)
Let’s walk through a full example so you see the numbers. I target solo founders who post about “AI workflow” on Twitter. Using Apify’s Twitter scraper actor, I pull the last 200 tweets containing that phrase, extract the author handles, and then query each handle for their bio and location.
First, I set up the Twitter actor with these inputs:
- searchQuery: “AI workflow”
- maxItems: 200
- includeRetweets: false
The actor returns a JSON array of tweet objects. I run a short Python script that:
- Loads the JSON from Apify’s dataset.
- Extracts unique author usernames.
- For each username, calls the Twitter actor again with “profileLookup” to get bio and location.
- Writes the results to a CSV with columns: username, bio, location, tweetUrl.
The whole process costs about 0.02 USD per run on Apify’s compute pricing. At 4 runs per month, that’s under $0.10 — basically free.
One concrete failure I hit: the Twitter actor occasionally returns a 429 error when you exceed the rate limit. I added a simple exponential backoff in the script (wait 2 s, then 4 s, then 8 s) and the problem vanished.
How to debug when the cold‑email pipeline stalls
Even the best scraper is useless if your emails land in spam or get ignored. I use MailerSend for sending and Make.com to orchestrate the flow. The pipeline looks like this: new lead in Google Sheets → Make.com triggers → MailerSend sends a personalized template → reply detection via Gmail webhook → if no reply after three days, send a follow‑up.
When I first launched it, open rates hovered around 12 %. I dug into the MailerSend logs and saw that many messages were flagged because the subject line contained the word “free”. Spam filters love to penalize that term.
I changed the subject line to “Quick idea for {{company}}” and open rates jumped to 28 %. The lesson: always check the raw SMTP logs before blaming the list.
Another common snag is the personalization token failing when a field is empty. I added a fallback in Make.com: if the company name is missing, use “there”. That tiny tweak stopped a handful of emails from sending with broken greetings.
One mild aside: the MailerSend dashboard can be slow to refresh when you have over 5 000 records — and good luck finding docs for this — but the API itself stays snappy.
Pricing your services: what’s fair and what’s ridiculous
Pricing is where many solo consultants either undercharge and burn out or overprice and scare away clients. I started at $75 per hour for basic workflow automation, then realized that most clients care about the outcome, not the hours.
Now I sell packaged outcomes: a lead‑gen scraper setup for $450, a cold‑email pipeline for $800, and a full “AI ops” bundle (scraper + email + invoicing) for $1 500. These prices reflect the value of time saved, not the time I spend.
I think $1 500 for a done‑for‑you automation stack is fair for a small business that saves at least 10 hours a week. If you charge less than $500 for the same stack, you’re likely leaving money on the table.
On the flip side, I’ve seen some agencies list a “AI consulting retainer” at $5 000/mo with vague deliverables. That feels ridiculous unless you’re delivering custom model training or ongoing prompt engineering — most solo clients don’t need that.
One concrete love: the invoicing step in Make.com that pulls data from a Google Sheet, creates a PDF via Pdfcrowd, and emails it through MailerSend. It costs $9/mo for Pdfcrowd’s basic plan, and it eliminates manual invoice work.
One concrete gripe: Pdfcrowd’s free tier watermarks every PDF, which looks unprofessional. I had to upgrade to remove the watermark — worth it, but annoying that the free plan is unusable for client‑facing docs.
Adjacent reading: deeper coverage of AI agent platforms.
If you’d rather skip the build and deploy a working version in an afternoon, we’ve packaged this workflow as a blueprint at deepusecase.com/vault/packages/agency.
