Metrics worth tracking in Metabase

The share of deals that close won — by cohort, segment, and stage entered.

How long deals take to close, and why medians beat averages for a skewed distribution.

Open pipeline divided by the quota or target it has to cover in a period.

Stage-to-stage and overall funnel conversion — where deals advance and where they stall.

The typical value of a won deal (ACV), and how mix and outliers distort the mean.

The share of contacted prospects who reply to outbound — by sequence, rep, and segment.

Qualified meetings generated by outbound — booked vs. held, without double-counting.

How close the forecast came to actuals — why you need snapshots, and how to track bias.

The share of sent emails that fail to deliver — hard vs. soft, and protecting deliverability.

The normalized monthly value of active subscriptions — the base for churn, retention, and LTV.

Revenue vs. customer churn, and voluntary vs. involuntary — where recurring revenue leaks.

Expansion minus contraction and churn across your existing customers.

Average recurring revenue per user or account, and how it differs from ARPA.

The expected gross-margin value of a customer over their lifetime.

The share of payments that fail — the main driver of involuntary churn.

The four keys: deployment frequency, lead time for changes, change failure rate, and MTTR.

How often you ship to production — a DORA throughput metric.

Time from commit to running in production — a DORA throughput metric.

The share of deploys that cause a failure — a DORA stability metric.

How fast you recover from a failure in production — a DORA stability metric.

Time from work started to done — how fast issues move once picked up.

Time from request created to delivered — end-to-end issue flow.

Successful CI runs over all completed runs — segmented by branch and pipeline.

Median and p95 CI wall-clock time, with queue time tracked separately.

Tests that both pass and fail on the same code — the tax on every build.

How long customers wait for the first human reply — the median, not the average.

Time from ticket created to resolved, and whether to subtract pending time.

Customer satisfaction — positive ratings ÷ rated tickets, and why response rate matters.

Tickets created vs. solved per period, plus the open backlog behind them.

The typical revenue per order, and how mix and discounts move it.

The share of customers who buy again — the core signal of retention and loyalty.

How long a candidate takes from application or first touch to accepted offer.

How long an open requisition takes to reach an accepted offer or start.

Accepted offers divided by extended offers, with pending and rescinded offers separated.

Stage-to-stage recruiting funnel conversion from ATS stage history.

Which recruiting sources produce qualified candidates, offers, and accepted hires.

The share of candidates who advance after a completed interview stage.

Available operating cash divided by average monthly net burn, with restricted cash and FX handled explicitly.

Open invoice balances grouped by days past due at a specific as-of date.

The number of days of credit sales represented by average accounts receivable.

Gross operating cash outflow and net cash consumed after operating inflows.

Revenue remaining after direct cost of revenue, based on an approved account mapping.

Actual results compared with one approved budget version at a shared reporting grain.

Payments allocated to an invoice cohort divided by the value invoiced in that cohort.

Foreign-currency balances and commitments translated at a documented valuation rate.

Completed work divided by due or committed work in a period.

How much work is completed per period, by project, team, or work type.

Open work past due divided by open work with due dates.

How open work is distributed across owners, teams, projects, and statuses.

A transparent risk score for overdue, blocked, stale, and incomplete work.

The share of active content reviewed or meaningfully updated inside its policy window.

Required topics and workflows with current, owned documentation.

The share of eligible boards, spaces, sites, or channels with meaningful recent activity.

Time from a question or new thread to its first useful human reply.

Conversion value over ad spend, by campaign and channel — with the attribution caveats spelled out.

Spend over new customers — blended vs. paid-only, and the payback math that follows.

Spend over leads generated — raw vs. qualified, so cheap channels don't hide bad leads.

The price of attention per channel, with CPM alongside — and why cross-channel comparisons mislead.

Clicks over impressions or delivered emails — the denominator changes by context.

Conversions over sessions per landing page, with sample-size thresholds that keep rankings honest.

Search visibility from Search Console rows — branded vs. non-branded, position, and CTR together.

Signups reaching the product's value moment — the bridge between marketing spend and retention.

Successful pipeline runs over all runs per destination — with a row-weighted version that catches partial syncs.

How old the newest delivered data is, per destination — the metric that notices pipelines going silent.

Qualifying incidents per period, segmented by severity and service.

How fast a responder picks up an incident — the paging-health half of response.

Failed events over total requests, sessions, or users — per service and release.

The share of alerts that never became incidents — the alert-fatigue early warning.

Successful requests or minutes over total, per service — uptime, done per service.

Reliability against explicit targets — error budgets and burn rates.

Used capacity over requested or provisioned — the bridge between reliability and cost.

Prompt and completion tokens per period, by model and app — and why to track them separately.

Your blended LLM unit cost — the honest number for comparing models and catching drift.

LLM spend over active users of AI features — the unit-economics headline, with its skew.

The tail response time of LLM calls — why percentiles, not averages, and how streaming changes it.

Failed LLM calls over all calls — provider errors, timeouts, rate limits, and guardrail blocks.

Cached responses and tokens over totals — the most direct LLM cost and latency lever.

Passing eval examples over all examples — the test-suite green rate for LLM quality.

Daily LLM call volume by app and model — the denominator behind every other LLM metric.

Open critical and high findings, deduplicated across scanners — by asset group and age.

Median days from a finding's first-seen to resolved, by severity — remediation velocity.

Findings closed within their severity SLA over all findings due — including the overdue.

Median time from first malicious activity to detection — a lower bound, honestly caveated.

Passing controls over monitored controls per framework, from daily snapshots.

Active users with a strong second factor enrolled — segmented by factor type and privilege.

Failed authentication attempts over all attempts — user friction and attack signal, separated.

Assets scanned or protected within a window over the true inventory — the honesty denominator.

Spend per period under one stated convention — amortized, unblended, or net of credits.

Allocated infrastructure cost per customer — the unit-economics headline, joined to revenue.

Month-over-month spend change, attributed to services and compared to business growth.

Commitment-covered share of eligible spend — always read together with utilization.

Realized savings against on-demand-equivalent cost — the honest commitment-program number.

Spend with no owner tag over total — the ceiling on every allocation dashboard.

Budget consumed vs. period elapsed — the early-warning pace metric for cloud spend.

Spend per provider or application service, with the resource-to-service mapping made explicit.

Volume-weighted mean search rank — and why GSC and rank trackers report different numbers.

Tracked keywords ranking in positions 1–10 — the visibility count that survives averaging.

Vendor link-authority scores (DR, Authority Score) — directional, and never cross-vendor comparable.

Unique domains linking to a target — count domains, not raw backlinks.

Net new referring domains per week — gains minus losses, with recrawl-spike caveats.

Your volume-weighted slice of tracked-keyword visibility against a fixed competitor set.

Clicks over impressions by position bucket from Search Console — and how SERP features bend it.

Percent promoters minus percent detractors on the 0-10 scale — always with response volume.

Responses over invitations, opens, or views — and why the denominator must be labeled.

Completed over started responses, with drop-off by question to find the leak.

New votes on open feature requests per period — what customers want now, not ever.

Median time from feature request to shipped and announced — the feedback honesty metric.

Adopting accounts over eligible active accounts, with tried vs. adopted tiers.

Frustration clicks per 1,000 sessions by page — the canonical digital distress signal.

Net new community members per period, with joins and departures shown separately.

Share of members who posted, replied, or reacted in a window you define and state.

Registrations that turned into check-ins — only measurable when attendance is recorded.

Composite account health, re-derived from its components instead of trusting a vendor black box.

Retained recurring revenue excluding expansion — the number that can never exceed 100%.

Forecast vs. actual renewals, graded against a forecast frozen at period start.

Median time from signup or contract start to a value milestone you have explicitly defined.

Booked meetings that never happened, with cancellations and reschedules kept out of the denominator.

Refunded revenue over revenue by order cohort — count vs. value rates, and the lagging-metric caveat.

Abandoned checkouts over started, from event data — cart vs. checkout stages, by device and step.

Total order value net of cancellations — GMV vs. revenue vs. take rate, decomposed with AOV.

COGS over average inventory, annualized — snapshot-based history and days of inventory.

Fully-loaded support cost over tickets handled — numerator definition, channel splits, CSAT guardrail.

The share of tickets that end resolved — closed vs. cohort denominators, with reopens excluded.

Open tickets trended daily with age buckets, created-vs-solved flow, and burn-down capacity math.

Escalated over total tickets, split by escalation path and segmented by topic to find docs gaps.

Self-service resolutions over total demand, with explicit signals and a quality guardrail.

Opt-outs over delivered per campaign, read next to spam complaints to localize spikes.

Net subscriber growth with components broken out, plus cohort quality by acquisition source.

Annualized departures over average headcount, split voluntary/involuntary and by tenure.

Recruiting cost per hire by quarter and source, paired with 90-day retention as a quality gate.

Logo vs. gross renewal rate — renewals over contracts due, on-time vs. late, and the auto-renew split.

Expansion MRR over starting MRR — the growth component inside NRR, with contraction shown separately.

Active members over total members — the DAU/MAU idea for communities, plus the stickiness ratio.

The share of tickets resolved in a single touch — no reopens, no transfers.

The share of tickets that missed a first-response or resolution SLA target, per policy and priority.

Margin per won deal — revenue net of discounts minus the cost to deliver, from CRM and finance data.

The share of committed deals whose close date slips out of the period, from close-date snapshots.

Spend per conversion, blended and by channel, with warehouse conversions instead of platform claims.

The share of sessions with no engagement, GA4-style, segmented by landing page and traffic source.

Leads created minus duplicates and junk — honest top-of-funnel volume by month and source.

Unique opens over delivered — read alongside clicks, since Apple MPP inflates opens.

Passing tests over executed tests by suite, with quarantined tests excluded honestly.

How long work waits in kanban states and CI queues — often the biggest share of lead time.

Remaining scope against time toward a release date, with scope changes made visible.

Story points completed per sprint — a planning signal, not a productivity score.

Net new bugs by severity and open-bug age buckets — the trends behind the raw total.

The percentage of time a service is up — nines as downtime budgets, from synthetic checks.

Average operating time between failures — the reliability counterpart to MTTR.

Students who finish a course over those who have had time to finish it — with the maturity window that makes it honest.

The share of gross sales lost to platform commission, payment processing, and payout fees — derived, not quoted.

Billed label cost per order and as a share of order value, using post-adjustment carrier charges.

Deliveries arriving by the promised date, over shipments with tracking coverage and a known outcome.

Restaurant and retail average ticket, with voids excluded and comps, tax, and tips handled deliberately.

Product page views converted to first-time installs, and why Apple and Google define it differently.