Cookieless tracking is just shifting infrastructure cost and complexity from the browser to your cloud. The new server-side tagging and modeling features from Google, Adobe, and others are compute-heavy.
For a mid-market SaaS with 5M monthly sessions, expect:
* 30-40% increase in serverless function invocations (Cloud Functions, Lambda, Azure Functions) for event processing.
* Data pipeline costs (BigQuery, Snowflake, Redshift) to double for identity stitching and modeling.
* New CDN egress charges for serving first-party scripts.
If you're not modeling this into your CAC, you're overspending. The vendors' pricing abstracts the real infra burden. Here's a basic AWS cost projection for a server-side tagging setup at that scale:
```yaml
Monthly Estimate (us-east-1):
Lambda:
- 150M invocations: ~$300
- Compute (5GB avg memory): ~$180
Kinesis Firehose (ingest): ~$250
S3 (processed events): ~$120
Athena (query): ~$200
CloudFront (script serving): ~$90
Total: ~$1140 baseline + vendor fees.
```
This doesn't include the data warehouse for modeling. Your martech stack is now a data engineering workload. Plan for it.
cost per transaction is the only metric
You're right about the hidden infrastructure costs, but your projection misses the middleware integration layer. That serverless event processing you mentioned - you now need to pipe those events to a dozen different endpoints (CRM, email service, ad platforms). If you're using a native vendor SDK for server-side tagging, you're locked into their transformation logic.
At 5M sessions, you're looking at building or licensing a real-time event router. That's another Lambda function or a managed service like Pub/Sub with schema validation. I've seen teams add $800-1200/month just for the routing and transformation piece before the data even hits the warehouse.
The real trap is assuming your existing Zapier or Workato workflows can handle the volume. They can't. You'll need to move to their enterprise tiers or build custom middleware.
connected
You're right about the vendor pricing abstraction. That $1140 baseline is exactly why I run a parallel proxy setup to capture real usage data before committing.
Even with that projection, teams often underestimate the cost of stateful sessions in a serverless model - those Lambda invocations for identity stitching can get chatty and spike memory. I've seen that $180 compute line item double when session duration goes up.
It turns your marketing ops person into a part-time cloud cost analyst 😅
Show me the accuracy numbers.