Case Study

AI Automation Systems

Four real systems I built for companies to remove a bottleneck.

+100
Leads scored & drafted / day
20
Hours saved / week
10
Minutes for research that used to take days
AI automation case study

Anyone can ask AI to write a post or generate an image. That's not what's separating the companies pulling ahead.

The ones getting real leverage out of AI didn't start with a tool. They started with mapping a pain point: one repeatable workflow that eats up your time, drawn out step by step.

  • Someone manually copying data between two tabs or extracting numbers from screenshots.
  • A request sitting in an inbox for seven days because you've got too many emails.
  • A progress or performance report that's incomplete because you're still manually chasing for an answer.

Until we finally build a system shaped around that specific problem.

A workflow that runs on its own and pulls a human in only at the point where judgment actually changes the outcome.

A generic prompt was never the solution. A custom AI system is.

Outreach & lead scoring

Every new lead scored, researched, and written to before anyone opens the sheet

Built in Make. It watches your lead list or spreadsheet for any new row, scores the lead, researches the context, and drafts a personalized email in Gmail. Then it logs everything back automatically. No manual work needed. It handles 100 leads a day and takes 20 hours a week of targeting, scoring, and research off your team's plate.

New lead lands in a list - You only have the company name
AI scores & researches the lead

Fit, context, and signal, pulled automatically, no manual digging.

minutes later
Contact found and personalized email is drafted

Claude writes a custom message to match the lead's context, not a generic template.

same run
Draft lands in Gmail, ready to send

No copy-pasting from an LLM, no writing from scratch. Human review before sending.

Lead logged, database or CRM updated
Outreach and lead-scoring automation, built in Make.com

Actual automation, built in Make.com.

Inbox to task

An assistant that turns incoming email into documented, assigned work

Also built in Make. It scans your inbox daily for messages from key stakeholders or clients, and uses AI to spot an actual task in the thread, skipping it if there isn't one. Then it creates that task in Notion, assigned to the right person or team. Works across sales requests, marketing briefs, anything. The triage that used to eat a morning isn't a person's job anymore.

New email arrives from your client or provider
Assistant reads it every day

Watches the inbox on a schedule. Nothing waits for someone to check.

instantly
Decides if it's actually a task

No real task in the thread? It's skipped. Nothing clutters the board.

Task documented & routed in Notion, sales or marketing, either works
Inbox-to-task automation, built in Make.com

Actual automation, built in Make.com.

Database/CRM Enrichment

A fallback chain that fills in the email addresses your list is missing

Built in n8n, it pulls the prospect list, flags every row missing a work email, then cascades through three email-finding providers in sequence: if the first comes back empty, it tries the next, and writes whatever it finds straight back to the sheet. One provider always has gaps.

Prospect list pulled, missing emails flagged
Tries the first email-finder

Provider one searches for a match.

no match? next
Builds the full outbound pipeline

Two more providers try in sequence, so every contact comes out enriched and ready to run.

Fully enriched prospect list, ready for your outbound campaigns
Lead enrichment waterfall automation, built in n8n

Actual automation, built in n8n.

Multi-agent research

50 companies researched, scored, and reported on in 10 minutes

Using Claude Code, I built a system for my client that runs 22 research agents in parallel across a batch of companies (ICP) to analyze their posts, PR activity and digital presence. Checking LinkedIn and Instagram, scoring each one for fit and for relevance. Then merges the findings into a single spreadsheet report, uploaded to Drive to trigger an outbound automation. Done by hand, that's entire days or weeks of work; here it finished inside a coffee break.

5 research agents working through target sites, checking LinkedIn / Instagram and scoring fit and genuineness.

✳ Waiting for 22 background agents to finish…

Agent "Research + draft: batch C (6 companies)" finished

Agent "Research + draft: batch D (6 companies)" finished

Agent "Research + draft: batch E (4 companies)" finished

→ Merging batches, building final report, uploading to Drive

50 companies · 10 min · what used to take days
Claude Code running 22 parallel research agents

Actual run, in Claude Code.

Send me the workflow you never want to do by hand again

That's the one worth building a system around.

Book a Call