Custom metrics and log indexing together drove 30 to 50 percent of spend, far ahead of the host counts everyone negotiates, so the meter nobody watches is the one setting the bill
More than twenty meters bill behind one logo. The renewal is decided by which of them you tame before the quote, not by which discount you win after it.
Prepared by Redress Compliance · August 18, 2026 · Observability advisory. 20 to 25 Datadog contracts benchmarked, 2024 to 2025.
Executive summary
Custom metrics and log indexing drove 30 to 50 percent of spend in most estates. Host counts, which is what the negotiation usually argues about, came a long way behind.
Usage cleanups before renewal cut baseline consumption 20 to 35 percent without losing any observability coverage the teams noticed. That is the discount you give yourself.
Annual commit rates ran 20 to 40 percent under on demand for the same usage, but only helped buyers who committed against a cleaned baseline rather than a dirty one.
Agents on decommissioned infrastructure keep billing. Dormant hosts renew as quietly as active ones, because nothing in the process asks whether the machine still exists.
What is actually being billed?
More than twenty separate meters, each turned on independently by a different team. Infrastructure hosts, APM hosts, custom metrics, indexed log volume, synthetics runs and RUM sessions all price separately on the Datadog pricing page.
The invoice is the sum of those decisions. Nobody chose the total, which is exactly why the total is negotiable in a way the rate card is not.
| Meter type | What it counts | What drives it up |
|---|---|---|
| Host meters | Infrastructure and APM per host per month | High water marks, which punish bursty estates |
| Volume meters | Custom metrics, indexed logs, ingested spans | Tag cardinality and retention defaults |
| Per run meters | Synthetics, RUM, CI visibility | Execution volume nobody caps |
| Dormant agents | Hosts that no longer exist | Nothing, which is the problem |
Cardinality is the mechanism
Custom metric counts scale with tag cardinality, not with how much you are watching. One unbounded tag can multiply a metric into thousands, and the billing documentation explains exactly how, which is the reading the seller hopes you skip.
Why does hygiene beat negotiation here?
Because a discount applies to whatever baseline you bring, and the baseline is usually 20 to 35 percent larger than it needs to be. Cleaning first changes the number the percentage multiplies.
Across the contracts benchmarked, cleanups removed that much consumption without losing coverage any team noticed. Nothing was turned off that anyone was using, which is the test that matters.
A commitment signed against a dirty baseline locks the waste in for the term. The discount looks the same on paper and costs more every month, because you have promised to keep consuming what you were about to remove.
The Datadog renewal brief
The meters that actually drive the invoice, the cleanup sequence before the quote, and the commit structure that survives a usage change.
Get the brief →What 20 to 25 Datadog contracts showed
Across roughly 20 to 25 Datadog contracts Morten Andersen benchmarked between 2024 and 2025, estate hygiene moved more money than negotiation in almost every renewal.
Custom metrics and log indexing together drove 30 to 50 percent of spend in most estates. Host counts, which is where the commercial conversation usually starts, were well behind both.
Usage cleanups before renewal cut baseline consumption 20 to 35 percent. The removals were unbounded tags, full indexing of logs nobody queried, and agents still reporting from decommissioned infrastructure.
Annual commit rates ran 20 to 40 percent under on demand for the same usage. That gap is real and worth having, and it only helps a buyer who has cleaned the baseline first.
Which is the whole argument. The rate is the last decision, not the first, and the three meters above decide what the rate is applied to. Usage attribution shows all of it before the quote arrives.
- Your quote benchmarked against 500,000+ real closed deals, adjusted for size, region, and industry
- Meter by meter consumption checked against the commit you are being asked to sign
- Commit and overage language flagged with the exact quote, the page, and the replacement text
Where does the spend leak first?
Through custom metric cardinality, then log indexing defaults, then agents on dead infrastructure. All three are visible in usage attribution before a renewal opens.
- Find the unbounded tags, because a single one can multiply a custom metric count without adding any insight.
- Index what you query and archive the rest, since full indexing is an expensive default that nobody chose deliberately.
- Reconcile reporting agents against infrastructure that still exists, because dormant hosts renew as quietly as live ones.
The same meter discipline applies across the observability estate. Our briefs on Splunk Cloud and the Datadog enterprise renewal cover the neighbouring cases.
Coverage is not what gets cut
The cleanups that removed 20 to 35 percent of baseline consumption did not remove observability any team noticed. That is the honest version of this argument, and it is the version that survives an internal review.
How should the commit be structured?
Against the cleaned baseline, with room for the meters that move fastest. The Datadog master agreement sets the frame, and the commit sits inside it.
An anchor helps. Grafana, New Relic and native cloud tooling are credible alternatives on parts of the estate, and a priced comparison is what makes the commercial conversation two sided.
Sequence the year
Clean, measure for a period long enough to be defensible, then commit. Committing first and cleaning afterwards is the order that gives the saving back to the vendor.
What the contracts measured, 2024 to 2025
Two cuts of the engagement file frame where the invoice is really decided.
Custom metrics and log indexing together, in most estates benchmarked, well ahead of host counts.
Before the renewal, with no loss of observability coverage that any team noticed.
The second number is only available before the commitment is signed, which is what makes the sequence matter more than the negotiation.
Watch the briefing · 5:27Negotiate the Billing Mechanics, Not the Rate CardWhich meters decide a Datadog invoice, and why the rate is the last thing to argue about.
Your first five moves
- Pull usage attribution before the renewal opens, because every leak in this brief is visible there without asking the vendor anything.
- Find and bound the high cardinality tags, since custom metrics and log indexing together drove 30 to 50 percent of spend.
- Index what you query and archive the rest, which is the single largest reversible default in most estates.
- Reconcile agents against infrastructure that still exists, because dormant hosts renew as quietly as live ones.
- Only then commit, and commit against the cleaned baseline. The negotiation practice runs the cleanup before the quote, which is the only order that keeps the saving.
Frequently asked questions
What actually drives a Datadog bill?
Custom metrics and log indexing, which together drove 30 to 50 percent of spend in most estates benchmarked. Host counts, which the negotiation usually argues about, came well behind.
How much can a cleanup remove?
Between 20 and 35 percent of baseline consumption before renewal, without losing any observability coverage the teams noticed.
Why do custom metrics grow so fast?
Because they scale with tag cardinality rather than with how much you are watching. One unbounded tag can multiply a metric into thousands of billable series.
Is full log indexing necessary?
Rarely. Index what you query and archive the rest. Full indexing is an expensive default rather than a decision anyone made deliberately.
How much does an annual commit save?
Rates ran 20 to 40 percent under on demand for the same usage, but the saving only holds for buyers who commit against a cleaned baseline.
What happens if we commit before cleaning?
The waste is locked in for the term. You have promised to keep consuming what you were about to remove, so the discount costs more every month than it appears to.
Do dormant hosts really keep billing?
Yes. Agents on decommissioned infrastructure keep reporting and keep billing, because nothing in the renewal process asks whether the machine still exists.
Where is all of this visible?
In usage attribution, before a renewal opens. Every leak described here can be found without asking the vendor for anything.
Does Datadog have credible competitors for anchoring?
Yes. Grafana, New Relic and native cloud tooling are real alternatives on parts of the estate, and a priced comparison is what makes the conversation two sided.
What is the right sequence?
Clean, measure for long enough to be defensible, then commit. Committing first and cleaning afterwards hands the saving back to the vendor.
