Analytics becomes useful when a number changes a decision. Opening a dashboard each week and observing that traffic moved is a monitoring ritual, not analysis.
You do not need a data science degree to ask sound questions. You do need clear definitions, functioning measurement and restraint about what behavioural data can explain.
Begin with one decision
Write down the question before opening the tool. Examples:
- Did the new enquiry journey reduce failed submissions?
- Which content helps referred prospects verify a service?
- Is a fall in qualified enquiries caused by demand, the website or follow-up?
- Are people finding the current guidance after a regulatory change?
Then identify the population, period, comparison and action. A metric without an intended response tends to become decoration.
Check whether the website has several jobs. A recruitment article, client login and service page need different success evidence.
Read counts, rates and denominators together
A conversion rate is conversions divided by an eligible population under a stated definition. If visits rise from a broad campaign while qualified enquiries stay constant, the blended rate may fall even though the commercial journey did not worsen.
Always inspect:
- numerator and denominator
- absolute counts
- definition of the event
- relevant segment
- period and normal variation
- data missing through consent or technical failure
Avoid comparing tiny groups through dramatic percentages. One additional enquiry can create a large rate change when volume is low.
Define “conversion” carefully. A resource view, newsletter subscription, contact form and instructed client are different outcomes. Do not combine them into one reassuring total.
Segment only with a reason
Source, device, geography, audience, landing page and new or returning status can reveal different behaviour. Each extra split reduces sample size and can create accidental patterns.
Choose segments connected to the hypothesis. If mobile enquiries fell, test the form and inspect errors by relevant device. If referral traffic evaluates the firm differently from search traffic, compare their priority journeys.
Channel attribution is approximate. “Direct” can include untagged links and lost referral data. Consent, browser controls and cross-device use affect identity. Treat source as evidence with limitations, not a perfect record of origin.
Use engagement measures diagnostically
Traffic describes reach. It is useful when the objective is visibility among a defined audience or when it supplies the denominator for another measure.
Bounce or engagement rate can reveal an unexpected pattern on a specific page. It cannot be judged sensibly as one site-wide score. A visitor may read a complete answer and leave satisfied.
Time on page can mean attention, confusion, an idle tab or a measurement artefact. Combine it with task completion, scroll, navigation and qualitative research.
Multi-page content sessions may indicate exploration. They may also indicate that the first page failed to answer the question. Look at the sequence and user need.
There are few metrics to “actively ignore” in every context. The skill is to stop treating any one of them as a verdict.
Build a trustworthy enquiry measure
Test form events on relevant devices and browsers. Confirm that the analytics event fires only after a successful submission and does not double-count refreshes or retries.
Reconcile analytics submissions with the receiving system. Differences may expose consent limits, spam, event errors or lost enquiries. Record the accepted source of truth for operational reporting.
Classify enquiry quality and progression in the CRM or appropriate system. Website analytics cannot determine whether a conversation became suitable work unless the service connects those records lawfully and reliably.
Use both first-heard and decision-influence questions in intake. Self-report and digital attribution answer different parts of the journey.
Create a useful weekly check
A short weekly review should detect failures and unusual movement, not determine strategy from seven days of noise.
Check:
- Are critical journeys and tracking functioning?
- Did qualified demand or source mix move outside an expected range?
- Did a release, campaign or external event plausibly explain the movement?
- Is there a specific investigation or action?
Use a moving baseline suited to seasonality and volume. Annotate releases, outages, campaigns and public events. A sudden zero may be a measurement failure before it is a market signal.
Test forms and other high-value journeys directly. Analytics should complement operational monitoring, not replace it.
Use the monthly review for interpretation
At a longer interval, ask:
- Which priority journeys helped users succeed?
- Where did people encounter errors or abandon?
- Which content appeared in relevant evaluation journeys?
- What did clients, prospects and internal teams report?
- What changed after the previous intervention?
- Which decision should the evidence influence now?
Compare with the CRM, support logs, search data and user research. Analytics can locate a pattern; observation often explains it.
Keep an analysis log containing question, data, method, finding, limitations, decision and follow-up date. This stops the same ambiguous chart being interpreted differently each month.
Be careful with early-warning stories
A declining return-visitor rate does not reliably predict a pipeline fall several months later without firm-specific longitudinal evidence. An increasing time from first observed visit to enquiry may reflect cross-device gaps or a changed marketing mix.
Treat these as hypotheses. Examine cohorts, proposition, seasonality, campaigns and tracking changes. Validate with pipeline and client evidence before forecasting.
Search visibility deserves similar care. Separate branded, informational and high-intent queries, but do not assume every phrase has one intent. Review the actual results and landing experience.
Check data quality before drawing conclusions
Document:
- analytics configuration and consent effects
- bot and internal-traffic treatment
- event definitions and changes
- channel tagging conventions
- known cross-domain or cross-device gaps
- CRM field completion and ownership
- retention and access controls
When a definition changes, mark the break in the series. A cleaner event can make performance appear to fall because duplicate counts disappeared.
Protect privacy. Collect the minimum needed, control access and avoid identifying individuals merely because the tooling makes it possible. Use appropriate legal and data-protection advice.
Communicate what the data supports
State the finding, evidence, limitation and recommended action. For example:
Successful mobile submissions fell after the release. Direct testing found an error on one browser version. We fixed it and will monitor reconciliation with the enquiry system. We cannot estimate missed revenue reliably.
That is more useful than a traffic graph or an invented loss figure.
The companion article on measuring whether the website is doing its job defines the wider commercial framework. The article on reporting digital performance covers board translation.
If you want the three-number weekly check and the three-question monthly review as a pre-configured dashboard template, download it here. It's a one-page framework covering the weekly metrics, the monthly questions, and the early warning patterns - designed to hand straight to whoever manages your analytics platform and have something useful by the end of the week.
