A GA4 setup that survives scrutiny

Most analytics implementations are installed once, never specified, and then quoted in board meetings as though they were an accounting system.

Analytics does not fail loudly. It produces numbers whatever you do, and they look plausible until somebody compares them with the invoices. The fix is boring and worth it: decide what you are counting before you count it.

Start with a measurement plan, not a tag

Write down the handful of actions that indicate progress: enquiry submitted, qualified enquiry, purchase, key page reached, call started. For each, define the trigger precisely and the value if it has one. The document is a page long and it prevents the most common failure, which is eleven similar events nobody can tell apart six months later.

Events worth defining

• Lead events, separated by quality. A form submission and a qualified enquiry are different things and should not share a name.
• Contact intent. Email clicks, phone clicks, WhatsApp clicks. For many Indian businesses these matter more than form fills and are routinely untracked.
• Commercial milestones. Pricing page reached, calculator used, sample requested. These are the leading indicators inside the site.
• Outbound clicks where a partner or marketplace completes the transaction.

Consent, done properly

If you operate in or market to places with consent requirements, implement consent mode with denied defaults and update on the visitor’s choice, rather than firing everything and hoping. Expect a share of traffic to decline analytics storage; modelling fills part of that gap and the rest is a known, disclosed limitation. Under India’s data protection regime, being able to show what you collect and on what basis is becoming an operational requirement rather than a legal nicety.

The reports that mislead

• Default channel groupings. Useful until a campaign is tagged inconsistently, after which a chunk of paid traffic appears as direct. Enforce UTM conventions in writing.
• Attribution comparisons. GA4’s model and each ad platform’s model will disagree. Neither is wrong; they answer different questions, and neither is revenue truth.
• Engagement rate as a quality signal. It measures time and scroll, not usefulness.
• Any period spanning a tracking change without a note saying so.

Reconcile with finance, once a quarter

Take a month, pull analytics-reported conversions and revenue, and compare with the finance system. A gap is normal; an unexplained gap is a problem; an unexamined gap is how a marketing team ends up reporting numbers the CFO does not recognise. Write the reconciliation down, including the reasons for the variance, and reuse the explanation.

Where server-side helps

Server-side tagging improves data quality where browser restrictions and ad blockers bite, and gives you control over what is sent to third parties. It also adds cost and an engineering dependency. It earns its place when paid media is a large budget line or the data feeds decisions of consequence. It is rarely the right first step for a business that has not yet defined its events.

Once the numbers are trustworthy, the next problem is presentation. Most reporting fails not because the data is wrong but because nobody can see the decision in it: see the dashboard that gets read.