How to create AI workflows for marketing

You spend hours each week copying data between spreadsheets, drafting follow‑up emails, and pulling ad performance numbers — time that could be spent on strategy. After reading this, you’ll be able to stitch together a repeatable AI workflow that pulls leads, writes copy, and updates a dashboard without manual steps.

What most guides get wrong about AI marketing workflows

Many tutorials treat AI as a magic button that replaces thinking. They show a single prompt and call it done. In reality a workflow needs reliable data ingestion, error handling, and a way to move output into your CRM or ad platform. Skipping those steps leaves you with a demo that dies after the first run.

Another common mistake is to rely on a single tool for everything. Trying to make GPT‑4 scrape LinkedIn, write copy, and push to Google Ads in one call creates brittle dependencies. Splitting concerns — scraping, generation, distribution — makes each piece easier to test and replace.

Why does my workflow stall when the lead list grows?

I hit this wall last quarter when I tried to process 2 000 leads in one Apollo.io batch. The API returned a 429 error after the 500th record, and my Make scenario stopped cold. The fix was simple: add a pause module and chunk the list into groups of 400. That tiny tweak let the whole run finish without manual babysitting.

Rate limits are the silent killer of “set‑and‑forget” automations. Most docs mention them in a footnote, but you need to design for them from the start. A pause of two seconds between chunks keeps you under Apollo’s limit of 100 requests per minute.

— and good luck finding docs for this — the official Apollo rate‑limit page is buried under three layers of FAQs.

Building the lead‑gathering step with Apollo.io

First mention of Apollo.io: I use its API to pull new leads that match a title and location filter. The request looks like this:

GET https://api.apollo.io/v1/mixed_companies/search
Headers: {"Cache-Control": "no-cache", "Content-Type": "application/json", "X-Api-Key": "YOUR_KEY"}
Body: {"q_keywords": "founder", "person_titles": ["CEO", "CTO"], "person_locations": ["United States"], "page": 1, "per_page": 400}

The response returns a JSON array of company and person objects. I keep only the person’s email, first name, and LinkedIn URL. Anything missing an email gets filtered out immediately.

My concrete gripe: I hate that Apollo’s API returns duplicate records unless you dedupe yourself. The same lead can appear in two different pages if they recently changed companies, so I added a deduplication step based on email hash before passing data forward.

Cost note: Apollo’s Growth plan at $49/mo gives 5 000 credits, enough for roughly 2 500 leads at 2 credits each. For a solo operator that’s more than enough; the free tier is a joke because it caps at 100 credits.

Turning raw leads into copy with GPT-4

Once I have a clean lead list, I send each record to OpenAI’s GPT‑4 with a prompt that creates a personalized first line. Here’s the exact prompt I use:

You are a friendly sales assistant. Write a one‑sentence icebreaker that references the recipient’s recent LinkedIn post about {linkedin_post_topic}. Keep it under 20 words and sound natural.

I replace {linkedin_post_topic} with a short snippet scraped from the lead’s LinkedIn activity (using a lightweight scraper like Phantombuster). The model returns a line such as “Loved your post about scaling remote teams — how’s the new hiring going?”

Concrete love: I love how GPT‑4 can spin a personalized intro line in under a second, turning a cold list into something that feels human.

Price opinion: OpenAI’s pay‑as‑you‑go rate of $0.03 per 1k tokens means a typical icebreaker costs less than $0.001. Even at 10 000 leads the API bill stays under $0.30, which feels negligible compared to the time saved.

Storing results and sending follow‑ups via the Make platform

Now we move to the orchestration layer. I built a Make scenario with four modules:

1. HTTP → Apollo.io (get leads)
2. Iterator → process each lead
3. OpenAI → GPT‑4 (generate icebreaker)
4. HTTP → Google Sheets (append row with email, icebreaker, timestamp)
5. HTTP → Gmail (send draft email using the icebreaker as subject line)

Each module has error handling: if the OpenAI call fails, the scenario logs the lead ID and continues; if Google Sheets is unreachable, it retries three times with exponential backoff.

One‑sentence paragraph: It works.

Concrete named example: the Google Sheets module uses the endpoint https://sheets.googleapis.com/v4/spreadsheets/{SPREADSHEET_ID}/values/A1:E?valueInputOption=RAW with a POST body of {“values”: [[“alice@example.com”, “Loved your post about scaling remote teams — how’s the new hiring going?”, “2026-09-26”]]}.

Price mention: Make’s core plan at $9/mo gives 10 000 operations, enough for my weekly lead flow of ~2 000 operations. I think $9/mo is fair for the automation it gives; the free tier would exhaust after a few hundred runs and force constant upgrades.

Cost breakdown and opinion on pricing

Here’s what I actually pay each month for the stack:

• Apollo.io Growth: $49
• OpenAI API (estimated): $2
• Make.com Core: $9
• Google Workspace (for Sheets & Gmail): $12 (already have it)

Total: ~$72/mo. If you already pay for Google Workspace, the incremental cost is $60.

Direct opinion that could be wrong: I think paying $79/mo for Jasper AI’s boss mode is overpriced for the simple copy tasks I need; GPT‑4 plus a prompt does the same job for a fraction of the price.

Mild aside: (which, yes, is annoying) the Make UI sometimes hides the “add another route” button behind a scrollbar when your scenario gets wide — just a tiny CSS quirk that makes you hunt for it.

For more on this exact angle, 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.