How We Cut a German SEO Agency's Automation Costs by 50%
A German SEO agency I work with ran a workflow that sent automated SEO audit reports to their newsletter subscribers. It pulled the list of subscriber websites, ran an audit through an API, passed the results to ChatGPT for analysis, and packaged the top recommendations into a newsletter.
It worked. That part matters, because nothing here was broken and there was no bug to fix.
The problem
The workflow cost $394 a month. Of that, $314 went to Clay. The rest was OpenAI and the SEO audit API.
Break the bill apart and the shape of the problem shows up immediately. The parts doing the actual work, the audits and the analysis, came to about $80. The orchestration cost four times as much as the thing being orchestrated.
Why Clay was the wrong fit here
Clay bills per operation. That model makes sense when each operation is doing something valuable and unpredictable, like enrichment across multiple data providers where you're paying for hit rate. This workflow wasn't that. It was a scheduled batch job that walked the same path every run: read a list, call two APIs, write to a sheet.
For that shape of work, per-operation pricing charges you repeatedly for a decision you only made once.
I want to be fair to the tool, because the answer isn't that Clay is overpriced. Clay earns its cost when non-technical people need to build and change the workflow themselves, or when the enrichment layer is the actual product. Neither applied here. The agency had someone technical, and the workflow hadn't changed in months.
The migration
Same logic, same vendors: DataForSEO for the audits, OpenAI for the analysis, Google Sheets for the output. The only thing that really changed was where the workflow runs. Self-hosted n8n on a $7 a month VPS, instead of Clay's per-operation billing.
Three days of work, and rebuilding the logic was the small part. Most of the time went into edge cases and error handling, specifically making sure a failed job surfaced instead of disappearing quietly. A batch job that silently drops a subscriber is worse than one that crashes, because nobody notices until a client asks why the reports stopped.
The move was also a chance to delete things. The original had accumulated formatting workarounds to fit how Clay handled data. Those came out in the rebuild, along with a cleaner approach to parallel processing and a tighter prompt structure.
Results
The new bill is $87 a month: $7 for the VPS, and roughly $80 of the same API calls as before. The $314 orchestration line is gone. That's about $300 a month, or $3,600 a year, for three days of work.
Output quality is the same, and arguably a little better, for a reason I didn't expect. Under per-operation billing, testing a prompt change costs money, so you test less and you second-guess more. Once iteration was free, the prompts got better simply because we were willing to try things.
Built by Ilyes Haddad, n8n automation and AI workflows. If you're paying more for orchestration than for the work being orchestrated: contact@ilyeshaddad.com
