- ·GA4 + GSC + RUM covers 90% of real questions. Every extra tool has to justify itself.
- ·Ship a one-page dashboard with an action-items section or it won't get read.
- ·Server-side tagging pays off above ~50k sessions; below that, spend the budget elsewhere.
- ·Attribution isn't precise — combine last non-direct + assisted + branded-search lift.
Every year, someone declares GA4 dead. Every year, we ship another dozen GA4 implementations because they work, the data is defensible, and the alternative stacks add cost and complexity without answering meaningfully different questions.
This is the analytics stack we actually use in 2026. Not the aspirational one you see on a conference slide. The one that ships, works, survives budget reviews, and gets read weekly by the client.
What we standardise on
GA4 for behavioural data. Search Console for organic truth (impressions, clicks, position, query data). Ahrefs or Semrush for competitive data, chosen based on the client's vertical strength of each tool. A lightweight RUM library — web-vitals — for Core Web Vitals field data. Looker Studio for the client dashboard, because it's free, familiar, and shares cleanly.
That's it. Five tools, all of them well-understood, all of them survivable in a budget review. Every additional tool in the stack has to justify itself against these five, which is a higher bar than most add-on tools clear.
The reason we standardise: onboarding a new client team member is faster, cross-client benchmarking is possible, and internal quality control is possible. A bespoke stack per client sounds sophisticated and produces reports that are impossible to compare against anything.
What we skip
Heat-mapping tools that never get looked at after week two. We've evaluated the major ones repeatedly; the ones we keep are the ones a client actually opens in their weekly review, and that's a very short list.
Session recording, unless there's a specific UX question in scope. Recording is a diagnostic tool, not a monitoring tool. Running it continuously produces a mountain of data nobody reviews.
Any tool whose primary value proposition is 'AI-powered insights' delivered as generated commentary. In practice these tools generate paragraphs that sound like insights and aren't. We'd rather have the raw data and a human interpretation than a machine-generated narrative that obscures the underlying numbers.
The dashboard we ship
One page. Traffic (organic sessions, month-over-month and year-over-year), conversions (leads or transactions, by channel), top movers (up-and-to-the-right pages this month, and any pages losing traffic materially), portfolio KPIs (whatever three metrics the client has said they care about), and action items (what we're working on this week and what we're proposing next).
Anything more and clients stop opening it. This is the single most important lesson from a decade of shipping analytics: the dashboard that gets read weekly beats the dashboard that would be perfect if it were read.
The action items section is the most useful part and the most often missing from templated dashboards. It closes the loop between measurement and work, which is the entire point of the dashboard existing.
How we handle server-side
Server-side tagging via GTM for privacy compliance and INP wins. Worth it for higher-traffic clients where the third-party script weight of client-side tagging materially impacts Core Web Vitals. Overkill below ~50k monthly sessions, where the engineering time exceeds the returned lift.
The pragmatic rule: if client-side GTM is on your INP autopsy trace as a top-three offender, server-side is worth the investment. If it isn't, spend the budget elsewhere.
Server-side also opens the door to first-party data enrichment (matching visitor IDs to CRM data, for example) that the marketing team may want down the road. We don't build for that speculatively, but we don't foreclose it either.
Attribution honestly
The honest position on attribution in 2026 is that no model is right. GA4's data-driven attribution is directionally useful; it should not be treated as ground truth. Marketing-mix modelling is a better lens for higher-spend clients but is overkill below ~$50k/month in paid spend.
For most of our clients, the practical attribution stack is: last non-direct click as a working baseline, plus assisted-conversions view as a sanity check, plus branded-search lift as an upper-funnel proxy for AI-Overview and organic-content exposure. When those three tell a consistent story, we trust it. When they disagree, we investigate before we act.
Where clients push for perfect attribution, we push back gently and re-frame the conversation around directional confidence rather than false precision. It's a more mature conversation and it produces better decisions.
The measurement conversation with founders
The measurement conversation with a founder is usually not about data — it's about which decisions the data is supposed to inform. Once you land the conversation there, the reporting requirements simplify dramatically.
The three questions we ask at the start of every engagement: what decision are you making this quarter that better data would improve, what number would you change your behaviour based on if it moved, and what are you currently reporting to your board or investors that you're not confident in? The answers to those three questions shape the dashboard.
That's why the dashboard fits on one page. If it takes three pages to inform three decisions, we're either measuring the wrong things or overthinking the presentation.
We ship this exact playbook in engagements.
If what you just read maps to a problem in front of you, a 30-minute conversation is the fastest way to know if we can help.