Observability Cost Audit
Datadog cost audit — Sample Co (fictional)
Billing month 2026-09 · generated 2026-11-17 15:06 · read-only analysis
Estimated monthly bill
$31,476
at the rates in the pricing sheet
Conservative savings
$1,088/mo
3% · high-confidence only
Likely savings
$5,662/mo
18% · high + medium
Annualized (likely)
$67,948
upside $100,222/yr
Where the money goes
| Log indexing | $7,858 | |
| Infrastructure hosts | $6,000 | |
| Log ingestion | $6,000 | |
| APM hosts | $3,720 | |
| Custom metrics (overage) | $2,800 | |
| Synthetics | $2,020 | |
| Containers (overage) | $1,752 | |
| Indexed spans (overage) | $1,326 | |
Recommendations, ranked
1
8 custom metrics nobody has queried in 30 days
$1,088/mo
high confidence
effort: lowrisk: low
metrics
Configure Metrics without Limits on these with no queryable tags (keeps ingestion, drops indexing), or remove them from instrumentation. Not used by any dashboard, monitor, SLO, notebook or API query in 30 days.
Evidence (8)
- legacy.worker.queue_depth: 6,200 series
- kafka.consumer.lag_by_partition: 5,400 series
- app.cache.hit_by_key: 3,900 series
- billing.job.duration_v1: 2,100 series
- test.loadgen.latency: 1,800 series
- etl.rows_by_table: 1,500 series
- app.feature_flag.eval: 900 series
- old.api.v1.requests: 400 series
2
Index 'main' keeps logs searchable for 30 days
$1,739/mo
medium confidence
effort: lowrisk: low
logs
Drop hot retention to 15 days; keep long-term copies in Flex Logs or a cloud-storage archive (rehydrate on demand). Check compliance requirements first.
Evidence (1)
- 2,173.2M events/month at $2.5/M vs $1.7/M
3
9 synthetic tests run more often than every 15 min or from >2 locations
$961/mo
medium confidence
effort: lowrisk: medium
synthetics
Run non-critical tests every 15 minutes from 2 locations. Keep tight intervals only on revenue-critical flows (login, checkout).
Evidence (9)
- Checkout flow (browser): every 300s from 4 locations → $346/mo
- Login flow (browser): every 300s from 3 locations → $242/mo
- Homepage up (api): every 60s from 5 locations → $105/mo
- API health /v2 (api): every 60s from 4 locations → $84/mo
- Docs site (api): every 60s from 3 locations → $62/mo
- Old v1 API (api): every 60s from 2 locations → $40/mo
- Staging smoke (browser): every 600s from 2 locations → $35/mo
- Search API (api): every 120s from 3 locations → $30/mo
- Webhook receiver (api): every 300s from 5 locations → $19/mo
4
High-cardinality tags on app.request.duration
$811/mo
medium confidence
effort: lowrisk: medium
metrics
Stop indexing user_id on this metric (Metrics without Limits allowlist of endpoint, env, service). Confirm with the owning team that no dashboard groups by these tags.
Evidence (2)
- Datadog estimator: 18,000 → 1,440 series
- tag user_id: 5,000 distinct values
5
High-cardinality tags on checkout.cart.value
$426/mo
medium confidence
effort: lowrisk: medium
metrics
Stop indexing session_id on this metric (Metrics without Limits allowlist of env, region). Confirm with the owning team that no dashboard groups by these tags.
Evidence (2)
- Datadog estimator: 9,000 → 300 series
- tag session_id: 2,400 distinct values
6
High-cardinality tags on search.query.latency
$321/mo
medium confidence
effort: lowrisk: medium
metrics
Stop indexing query on this metric (Metrics without Limits allowlist of env, service). Confirm with the owning team that no dashboard groups by these tags.
Evidence (2)
- Datadog estimator: 7,000 → 450 series
- tag query: 800 distinct values
7
Index 'audit' keeps logs searchable for 30 days
$240/mo
medium confidence
effort: lowrisk: low
logs
Drop hot retention to 15 days; keep long-term copies in Flex Logs or a cloud-storage archive (rehydrate on demand). Check compliance requirements first.
Evidence (1)
- 299.6M events/month at $2.5/M vs $1.7/M
8
High-cardinality tags on payments.auth.latency
$76/mo
medium confidence
effort: lowrisk: medium
metrics
Stop indexing request_id on this metric (Metrics without Limits allowlist of env, provider). Confirm with the owning team that no dashboard groups by these tags.
Evidence (2)
- Datadog estimator: 1,600 → 40 series
- tag request_id: 900 distinct values
9
Index 'main' has no exclusion filters
$924/mo
low confidence
effort: mediumrisk: low
logs
Use Log Patterns to find the top noise (health checks, debug, repeated INFO) and add exclusion filters or sample them at 10%. Excluded logs still feed log-based metrics and archives.
Evidence (3)
- filter query: *
- 2,173.2M events/month, ~$5,433/month
- assumes 25% of volume is excludable — validate against patterns
10
2 custom retention filters index ≥50% of matching spans
$663/mo
low confidence
effort: lowrisk: low
apm
Lower these to 10% unless a team actively searches those spans. Error and latency traces are still kept by Datadog's intelligent retention filter.
Evidence (2)
- 'checkout-all' query='service:checkout' rate=1.0
- 'debug-traces' query='env:staging' rate=0.5
11
Containers exceed the included allotment
$526/mo
low confidence
effort: mediumrisk: low
hosts
Exclude short-lived and system containers (pause, sidecars, kube-system jobs) with the Agent's container_exclude settings.
Evidence (1)
- container overage ≈ $1,752/month; assumes 30% is excludable
12
60 hosts billed through cloud integrations without an Agent
$450/mo
low confidence
effort: lowrisk: low
hosts
Every VM a cloud integration discovers is billed as an infrastructure host. Add host tag filters to the AWS/GCP/Azure integration so only instances you actually monitor are collected (e.g. exclude dev, batch and autoscaling scratch instances).
Evidence (12)
- source aws: 60 hosts
- batch-340
- batch-341
- batch-342
- batch-343
- batch-344
- batch-345
- batch-346
- batch-347
- batch-348
- batch-349
- assumes 50% can be filtered — confirm with the platform team
13
Index 'audit' has no exclusion filters
$127/mo
low confidence
effort: mediumrisk: low
logs
Use Log Patterns to find the top noise (health checks, debug, repeated INFO) and add exclusion filters or sample them at 10%. Excluded logs still feed log-based metrics and archives.
Evidence (3)
- filter query: source:audit
- 299.6M events/month, ~$749/month
- assumes 25% of volume is excludable — validate against patterns
Worth fixing, no savings today
- Index 'main' has no daily quota — Set a daily quota with a warning threshold so a logging bug can't produce a surprise bill.
Method & assumptions
- All data was read through Datadog's API with a read-only key. Nothing in the account was changed.
- Costs use the rates in the pricing sheet (Datadog list prices — replace with contract rates before quoting).
- Savings on custom metrics, containers and indexed spans are capped at the billable overage — cutting below the included allotment saves nothing.
- High confidence = proven by usage data (e.g. never queried). Medium = Datadog's own estimator or simple arithmetic, needs owner sign-off. Low = rule-of-thumb share, validate before committing.
- 2026-09: 400 hosts, 120 APM hosts, 96,000 custom metrics (allotment 40,000), 3,200 containers.
- Metrics: 16 custom metrics read; top 8 queried ones (by peak series) inspected for tag cardinality.