I need help with Speed Optimization: Apify → Web Scraping → GPT-5 mini/GPT-5 Email Personalization Workflow

Help me Build my Workflow
1

Speed Optimization Help Needed: Apify → Web Scraping → GPT-5 Personalization Workflow

Hey guys, here’s a great, working workflow that creates personalized cold emails I’d love to put into Instantly.ai for outreach. Unfortunately, it’s way too slow — it currently takes about 3 minutes per lead — so uploading 1000 leads just isn’t feasible.

What It Does

It generates hyper-personalized cold email icebreakers (and does a great job creating those):

  • Pulls leads from Apollo with Apify

  • Scrapes multiple pages from each website

  • Uses GPT-5 mini and GPT-5 for every lead (summarization + icebreaker generation)

  • Outputs to Google Sheets

Results are great, but it’s painfully slow.

The Problem

  • Current: ~3 minutes per lead

  • At scale: 1000 leads = 50 hours :scream:

  • Bottlenecks: sequential processing, 2x GPT calls per lead (GPT-5 mini + GPT-5). The workflow spends too much time on the two GPT nodes. Switching to smaller GPT models didn’t help much. Ideally, it should handle many leads at the same time.

Current Architecture

  • Apollo/Apify → filter leads with websites

  • Loop through each lead individually

  • Scrape homepage + 3 internal pages (for personalization in case the homepage doesn’t contain much info)

  • GPT-5 mini summarizes each page

  • GPT-5 generates icebreaker from the summaries

  • Write to Google Sheets

How do you handle batch processing with AI nodes, and how would I need to set up parallel processing for this? The sequential loop is killing performance.

Looking for any ideas to get this from 3 minutes to under 30 seconds per lead.

I used Nick Saraev’s template as the base for this automation but tweaked it for my use case. Maybe someone else has also used his template and made it work? I’d really appreciate your help! :slightly_smiling_face:
Happy to share the JSON if anyone wants to take a look.

Icebreaker-Workflow
Icebreaker-Workflow2038×576 143 KB

2

Switch from chatgpt to google gemini. Had a client who had the same problem recently, that cut down the processing time by a lot. I’ve recently noticed that OpenAI has become slower, not sure if it’s because of the thinking models or are they slower now in general.

3

Ah, thank you so much, artur! I’ll try it out!

4

You might also want to try to remove the loop and instead create a sub workflow replacing it (Run once for each item).

5

switching LLM model might improve things a bit, but to get massive improvements you just need to make sure you leverage concurrency properly. Once you got 20 links from Apify, you should have another workflow which runs concurrently - e.g. 5 threads - so 5 websites are processed at the same time. So even if a single lead still takes 3 min to do a research, 1000 leads will take 5x less time.