Investigate first, contract later

A recent beta run against a mid-market healthcare SaaS company's Datadog org produced a report with two clearly separated sections: a short list of things to investigate, and a longer list of things to consider committing to. That split isn't cosmetic. Treating a temporary anomaly like a structural cost and a structural cost like a temporary anomaly are two different, both-expensive mistakes — and the report's own explicit timeline is the clearest way to see why the order matters.

Two different questions, two different actions

"Investigate this" findings are anomalies: cost jumps, forecasted jumps, things that broke from a recent baseline. The right response is to look — pull the relevant metrics, check what changed in a deploy, confirm whether it's real usage growth or something else (see our post on serverless cost spikes and error rates for exactly this kind of check). Nothing here commits you to anything; the report is explicit about that, marking each investigation item with "what you give up: nothing."

"Commit to this" findings are different in kind: sustained, multi-month, 100%-on-demand spend where the number has held steady long enough to be a pattern rather than a blip. The right response there is a genuinely different action — a conversation about a contract, not a config change.

Conflating the two produces two failure modes. Commit to a 12-month rate based on a spike that turns out to be a retry storm from a bug, and you've locked in an inflated baseline for a year over something that would have resolved itself with a bug fix. Or: dismiss a real, sustained cost increase as "just noise, we're not committing to anything" and you miss both the chance to fix an underlying problem and the chance to negotiate a fair rate on the spend that's actually going to stick around.

The report's own timeline

This run's guidance was explicit about sequencing: this week, investigate the flagged cost jumps and forecasted jumps. This quarter, re-measure on-demand spend after that cleanup, then evaluate a committed-use contract — schedule the call with your Datadog account rep once you know the real, post-cleanup numbers. Ongoing, keep watching for new anomalies rather than treating this as a one-time exercise. The order is the point: contract decisions get made on numbers that have already survived a round of "is this even real" scrutiny, not on whatever the dashboard happened to show the day the report ran.

The nuance our other committed-use post didn't cover: sometimes the answer is no

In an earlier post, about a different anonymized company, every 100%-on-demand line item was worth taking into a contract conversation — the dollar amounts were large enough that committing made sense across the board. This report reached the opposite conclusion for four on-demand families at a smaller healthcare SaaS company, despite every one of them also having $0.00 committed coverage:

Product family	On-demand spend	Verdict
serverless_apps_apm	$90.00/month	Low ROI — keep flexibility
serverless_apps	$73.44/month	Low ROI — keep flexibility
synthetics_api_tests	$64.29/month	Low ROI — keep flexibility
ingested_spans	$41.27/month	Low ROI — keep flexibility

The reasoning is the same for all four: at this volume, the administrative overhead and the loss of flexibility that comes with a 12-month lock-in outweigh the saving on offer. $41 to $90 a month is real money over a year, but it's not enough to justify losing the ability to walk away from that specific line item if usage drops or the service gets decommissioned. Same underlying check — $0 committed against real usage — completely different recommendation, because the dollar volume changes the math. There's no blanket rule here that says "100% on-demand always means negotiate a contract." It depends on the size of the number.

The one way to make small lines worth committing anyway

The report names the exception explicitly: bundling several small on-demand families into one account-wide master contract, so the administrative overhead gets paid once instead of four separate times. Four lines that individually don't clear the bar for a standalone commitment might collectively be worth negotiating as a single package — but that's a conversation to have explicitly with your Datadog account team, not something to assume or self-serve. Worth raising directly: "we have several smaller on-demand families — is there a bundled or account-wide rate that makes committing to all of them worthwhile, even though none of them individually would be?"

What this tool actually does — and doesn't

Worth stating plainly, because the report itself does: this measures and reports, it doesn't act. Every finding reflects exactly what Datadog has received and stored at the time the tool ran — if your own account's tags, retention filters, or usage data are incomplete or misconfigured, that limitation flows straight through into what the report can tell you. And nothing in the report changes anything in your Datadog account; every recommendation only produces a result once your team actually implements it. That's a feature, not a caveat — see our decision framework for evaluating any committed-use offer for how to take a recommendation like this and actually decide, rather than just act on the report's word.