Learn concrete generative AI examples to automate cold email, lead gen, and invoicing—plus when to grab the blueprint for your solo business today now.
If you’ve only seen generative ai examples as flashy demos, you’re missing the real operator value.
In this guide I’ll walk through three working pipelines—cold email, lead-gen scraping, and invoice automation—using actual prompts, tool names, and cost numbers.
What most guides get wrong about generative ai examples
Most tutorials treat prompts as one‑off magic spells. They show a single clever line and call it a day. In production you need repeatable logic, error handling, and cost controls. If you skip those pieces the demo works once and then collapses under real volume.
Another common mistake is ignoring token limits. A prompt that looks fine in the playground can explode when you feed it a list of 500 leads. Guides rarely show how to chunk inputs or reuse completions.
Here’s the exact flow I run for a freelance outreach business:
- Step 1: Pull new leads from a Google Sheet via Make.com’s Google Sheets module.
- Step 2: For each lead, send a prompt to GPT-4 through the OpenAI module. The prompt is:
Write a short, friendly cold email to {{first_name}} at {{company}}. Mention one recent post from their LinkedIn (see {{linkedin_update}}). Keep it under 120 words. End with a low‑pressure call to action to schedule a 15‑minute chat.
- Step 3: Capture the generated email, add a personalized PS, and route it to Gmail for sending.
- Step 4: Log the sent email, timestamp, and token usage back to the sheet for tracking.
Cost breakdown (as of 2026):
- Make.com core plan: $29/mo (enough for 5 000 operations)
- OpenAI GPT-4 usage: ~0.03 USD per 1 000 tokens; each email averages 800 tokens → ~0.024 USD per email.
- At 200 emails per week the AI cost is under $20/mo.
This setup has been stable for six months. The only hiccup was a sudden rate‑limit tweak on Make.com that I’ll cover in the gripe section.
How to debug when this breaks
When the pipeline stops delivering emails, first check the Make.com scenario log. Look for modules that returned an error status—usually the OpenAI module.
If the OpenAI module shows a 429 error, you’ve hit the token‑per‑minute limit. The fix is to add a “Sleep” module between each OpenAI call, delaying 250 ms. That keeps you under the limit without adding much latency.
If the emails look garbled, inspect the prompt output in the log. Often the issue is a missing variable (e.g., {{linkedin_update}} empty) causing the model to hallucinate. Add a fallback text in the sheet so the variable is never blank.
Finally, verify the Gmail module isn’t hitting a daily send quota. Google caps free accounts at 100 emails/day; upgrade to a Workspace plan if you need more.
Concrete love: the conditional branching in n8n workflows that saved me hours
I switched the lead‑gen scraper from Make.com to n8n because I needed richer logic. The feature I rely on every day is n8n’s IF node that lets me split leads based on company size.
For example, if the employee count from the scraped LinkedIn data is >200 I route the lead to a high‑touch sequence; otherwise it goes to a fully automated sequence. This single node cut my manual sorting time from two hours a week to ten minutes.
The visual workflow makes it easy to tweak the threshold without touching code—just drag a slider.
Concrete gripe: Make.com’s sudden rate limit change broke my scraper
Last month Make.com lowered the maximum requests per minute for the HTTP module from 120 to 30 without updating their public docs. My scraper, which relied on bursting requests to fetch LinkedIn profiles, started throwing 429 errors across the board.
I had to rebuild the throttling logic using a combination of the “SplitInBatches” node and a custom delay. It took three hours of trial and error, and the lack of a changelog notice felt like a vendor decision that punished existing users.
— and good luck finding docs for this — the new limit is buried in a forum thread.
Price opinion: why $29/mo for the AI agent framework is fair
I pay $29/mo for the core Make.com plan that runs all three pipelines. Considering the time saved—roughly eight hours a week of manual email drafting and list cleaning—that’s less than $4 per hour of my own labor.
If you’re solo and only need one scenario, the free tier is enough for light testing. But once you hit daily volumes over fifty leads, the paid plan removes the operation cap and adds the error‑handling modules you actually need.
I think the free plan is a joke for any real‑world use, but the paid tier is priced right for the value it delivers.
One-sentence paragraph
Test every prompt in a sandbox before scaling to avoid costly token waste.
How do you keep token costs under control when looping over 500 leads?
Chunk the list into batches of 25 and run each batch through a separate scenario instance. This prevents a single run from exceeding the platform’s execution time limit and lets you retry only the failed batch.
Also, cache the completion for identical inputs. If two leads share the same company name and industry, reuse the generated email body and just swap the greeting. That simple dedupe can cut token usage by 30‑40% on repetitive lists.
Finally, monitor the token counter in the OpenAI module’s output. Set an alert at 80 % of your monthly budget so you know when to throttle or upgrade.
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.
