A dashboard never fails from lack of numbers — always from excess. Twenty metrics going up and down with nobody knowing what to do about them: that is the fate of most dashboards. Here are the KPIs that earn their place, their reading traps, and the architecture that makes the whole usable.
Quick Answer: which KPIs for a website?
Six to ten indicators, in four families:
- Traffic — unique visitors and channel split: quality per source over raw volume.
- Engagement — duration, read rate, bounce qualified by time spent.
- Conversion — segmented global rate, funnel step-through rates, micro-conversions (basket, form started).
- Value — revenue, average order, and the king composite: value per visitor (revenue ÷ visitors), which folds traffic, conversion and basket into one number comparable over time.
Admission test for any KPI: what decision does it trigger if it deviates? No answer = vanity metric.
The 12 KPIs, family by family
Traffic — the raw material
- Unique visitors (trend, not absolute value): the gauge of your overall visibility.
- Channel split: SEO, direct, social, campaigns, referral, AI assistants. The structure states your dependency — 80% SEO is a strength and a risk.
- Quality per channel: engagement and conversion per source. The KPI that arbitrates budgets — a channel is judged by what it brings in, not what it brings (details here).
Engagement — does the traffic mean anything?
- Session duration and depth — crossed: long duration + one page = reading (fine for a blog); short + multi-page = wandering.
- Read/scroll rate of key pages: are your decisive arguments even seen? (the 50% line)
- Qualified bounce: bounce crossed with time spent — flash bounces signal a broken promise, long bounces a served content (reading grid).
Conversion — is the site doing its job?
- Conversion rate per goal and segment: never one global number — mobile/desktop and per channel at minimum (benchmarks).
- Funnel step-through rates: the diagnostic KPI — the one that names the work (method).
- Micro-conversions: baskets, forms started, pricing views. Leading indicators: they move before the macro conversion.
Value — how much, in the end?
- Revenue (or generated value: quotes, weighted leads) per period and channel.
- Average order value — always read with conversion: the two often offset (promo = conversion ↑, basket ↓).
- Value per visitor: revenue ÷ visitors. The best composite for overall health and period comparison — it absorbs traffic-mix variations without lying.
The reading traps that cost decisions
- The crushing average. Every global KPI hides diverging segments; the reflex: one click of segmentation before any conclusion.
- Daily variation mistaken for signal. Day-to-day noise is normal; the weekly trend decides. Keep real time for launches and campaigns.
- The KPI without its twin. Conversion without average order, traffic without quality, bounce without duration: every indicator has a partner that keeps it honest.
- The amputated measurement. If your analytics depends on a banner refused by 30–50% of visitors, ALL your KPIs are computed on a biased sample — and your trends mix real behaviour with refusal-rate variations (consent-exempt measurement removes that noise).
- Bots in the denominator. Unfiltered traffic inflates visitors and mechanically crushes conversion and engagement.
Building the dashboard: the three-storey architecture
Leadership storey (weekly/monthly): 4-6 business KPIs — visitors, value per visitor, revenue, global conversion, top channel movement. Readable in 60 seconds.
Operational storey (weekly): per domain. Acquisition: quality per channel, cost if campaigns. Conversion: funnel step by step, micro-conversions. Content: reading of strategic pages, SEO entries.
Investigation storey (on demand): everything else — heatmaps, replays, fine segments. You go down there when a KPI deviates, not by routine. That is where behavioural tools (replay, heatmaps) take over from the numbers.
This hierarchy answers a dashboard's real question: not "what happened?" but "must I act, and where?".
In practice with Mirage — The dashboard delivers the leadership storey without configuration: visitors, engagement, conversions, revenue and channels in real time, built-in period comparison, bots filtered by default. The next storeys are one click away: funnels, segments, replays. And every KPI is computed on 100% of traffic — no consent blind spot. Free 30-day trial.
FAQ
What are the essential KPIs for a website?
Four families: traffic (visitors, sources), engagement (duration, read rate, qualified bounce), conversion (global and per funnel step, micro-conversions) and value (revenue, average order, value per visitor). The rule: every displayed KPI must be able to trigger a decision — otherwise it's a vanity metric.
Which KPIs are specific to e-commerce?
Beyond the base: add-to-basket rate, cart abandonment per step, average order value, value per visitor (revenue ÷ visitors — the best composite indicator), returning-customer rate, and revenue per acquisition channel. The conversion × average order pair always reads together: a promo can improve one while destroying the other.
How many KPIs should you track?
A steering dashboard holds 6 to 10 indicators: beyond that, nobody reads it and signals drown. The right architecture: a leadership level (4-6 business KPIs), an operational level per domain (acquisition, conversion, content), and investigation metrics you only open when a KPI deviates.
How often should you check your KPIs?
According to the site's velocity: weekly for routine steering (daily variations are noise for most sites), monthly for trends and arbitration, real time only for critical moments (launch, campaign, spike). The daily full-dashboard reflex is more anxiety-inducing than useful.