How We Cut Datadog Bills by 60% Without Losing Observability

Last year our Datadog bill hit $38k/month. Leadership asked me to cut it in half. Here's how we got to $15k without losing a single useful signal.

The 5 moves that worked

1. Dropped custom metrics with zero dashboard references. We had 2,400 custom metrics. Only 600 were actually graphed or alerted on. We stopped sending the other 1,800. Saved 30%.

2. Aggressive log tier management. Hot logs for 3 days, warm for 7, then cold. Most of our log cost came from full indexing of debug logs nobody queried after 24 hours.

3. Lower-cardinality tags. We were tagging metrics with user_id. That's millions of series. We moved user-level stuff to traces and kept metrics aggregate-only.

4. Dropped synthetic monitors for dev environments. We were running 200+ API checks on dev. We only need them on prod.

5. Consolidated APM sampling. Instead of 100% sampling on everything, we did 10% sampling on healthy traces, 100% on errors and slow requests. Cut APM volume by 85% with no loss in signal.

What didn't work

Trying to negotiate a discount with sales. They'll give you 5-10% if you threaten to leave. Actually cutting volume gets you 60%.

The lesson: your observability bill is almost always a data hygiene problem, not a pricing problem.

Written by Dr. Samson Tanimawo
BSc · MSc · MBA · PhD
Founder & CEO, Nova AI Ops. https://novaaiops.com