How I Work

Most consultants are evaluated on claims you cannot check. This page is the alternative: the actual method, the actual scope boundaries, and the things I explicitly will not do.

I write both sides of the measurement layer — the application and the tracking — so conversion data doesn't break at the handoff. Everything below follows from that.


The Five Stages

Timings assume a mid-sized implementation. The order never changes, because every expensive measurement failure I have seen traces back to skipping one of the first two.

1

Stage 1 — Discovery & Loss Quantification

Week 1
  • Audit what is currently collected, and what is silently missing
  • Reconcile platform numbers against the CRM or backend truth
  • Write down the decisions the measurement is supposed to support
  • Establish the baseline that later work will be judged against

You end up with: A written diagnosis, including whether the project is worth doing.

2

Stage 2 — Measurement Plan

Week 1–2
  • Event catalogue: names, parameters, types, and where each fires
  • Key events and conversion definitions agreed before implementation
  • Custom dimensions budgeted against platform limits, not added ad hoc
  • Consent and privacy requirements settled at design time

You end up with: A specification anyone can implement — including someone who isn't me.

3

Stage 3 — Implementation

Week 2–4
  • dataLayer instrumented in the application source, not scraped via CSS selectors
  • Container built to the spec, with naming that survives a year of edits
  • Server-side infrastructure provisioned where the diagnosis called for it
  • Everything version-controlled so changes are reviewable

You end up with: Working collection, and the code that produces it.

4

Stage 4 — QA & Reconciliation

Week 3–5
  • Validated against real user journeys, not just preview mode
  • Cross-checked against the Stage 1 baseline to prove the delta
  • Edge cases exercised: consent denied, ad blockers, mobile, slow networks
  • Debug artifacts kept, so the QA is auditable after the fact

You end up with: Evidence the numbers are right, not an assurance that they are.

5

Stage 5 — Handover

Final week
  • Documentation written for whoever maintains it next
  • Container exports and specs handed over in full
  • Walkthrough session with the team who will own it
  • A named list of what will break if certain things change

You end up with: Infrastructure your team can operate without me. That is the goal, not a retention risk.


What I Don't Do

A scope list is more informative than a service list, because everyone's service list says the same things.

Statistical modelling and data science

Attribution modelling, propensity scoring, and forecasting are a different discipline. I build the pipelines that feed them and will say so rather than improvising — hire a data scientist for the modelling itself.

Ongoing campaign management at agency scale

One operator cannot run a large multi-account media desk and do the engineering well. Strategy, structure, and measurement — yes. Full-time day-to-day account management across many accounts — that is what an agency is for.

Rubber-stamping a setup I did not verify

If the diagnosis says the client-side container just needs fixing, that is what the proposal will say, even when the bigger project would have been the better invoice.

Consent workarounds

Server-side tagging moves where processing happens; it does not remove the need for a legal basis. Any request framed as getting around consent is one I turn down.


Capacity, Honestly

The most common objection to hiring an independent consultant is capacity, and it is a fair one. So, plainly:

  • A small number of concurrent engagements. The QA stage is what makes this work worth paying for, and it does not compress.
  • The person you talk to is the person doing the work. There is no bench, and nothing gets handed to a junior.
  • Availability is a single switch on this site. When I'm at capacity, every call-to-action says so instead of collecting enquiries I can't answer.
  • Handover is designed so your team can operate the result without me. Ongoing dependency is not the business model.

If your project needs guaranteed parallel throughput across many accounts, or 24/7 coverage, an agency is the correct answer and I will tell you so.


Things You Can Check Yourself

Without a reference to call, the next best thing is work you can inspect. Every tool below runs entirely in your browser — put in numbers whose answer you already know and see whether it agrees.

Also checkable: Google Ads and Google Analytics certifications tied to a real account, and a public GitHub history. None of it proves I'll be good to work with — but it is all verifiable, which a five-star review is not.


Frequently Asked Questions

Why are there no client case studies or testimonials?
Because I don't have publishable ones. Measurement work sits on top of a client's revenue data, conversion rates, and ad spend, and most of the engagements worth writing about are exactly the ones covered by confidentiality. Rather than publish vague, unverifiable outcome claims — 'increased conversions by 40%' with no way to check it — the proof on this site is work you can inspect directly: shipped products, published tools whose maths is stated on the page, and a documented method. That is a weaker signal than a named reference, and I would rather be straight about which one you're getting.
You're one person. What happens when you're at capacity?
You get told. Capacity is genuinely finite here — a small number of concurrent engagements, because the QA stage is what makes the work worth paying for and it does not compress. The site's contact buttons are wired to a single availability switch: when I'm not taking work they show that, with no contact form to submit into a void. There is no hidden queue and no junior taking the overflow, which is both the honest limitation and the point — the person you speak with is the person doing the work.
Isn't a full-stack developer who also does analytics a generalist?
It would be, if the two were unrelated. The specific failure this combination addresses is the handoff: marketing specifies tracking, developers implement something adjacent to it, the dataLayer drifts from the spec, and six months later nobody can say why the conversion numbers stopped reconciling. I write both sides of that boundary — the application and the tracking — which is narrower than it sounds, not broader. What it explicitly does not include is data science; see the scope list above.
How do I verify any of this without a reference to call?
Check what is machine-verifiable rather than what is asserted. The Google Ads and Google Analytics certifications are real credentials tied to an account. The GitHub history is public. The tools on this site run entirely in your browser — open one, put in numbers you already know the answer to, and see whether it agrees. The projects are described with what was actually built. None of that proves I will be good to work with, but it is all checkable in a way that a five-star review is not.
What does an engagement cost?
It depends on scope, and I would rather give you a real number after Stage 1 than a fake one before it. What I can do is remove the guesswork about the market: independent 2026 guides put a full server-side tagging implementation in the region of $8,000–$18,500, with hosting from roughly $50–400/month depending on traffic. A GA4 and GTM implementation without server-side infrastructure is a smaller project than that. If a proposal you receive — from me or anyone — will not model the cost before you commit, treat that as the warning sign.
What if the diagnosis says I don't need the project?
Then that is the deliverable, and Stage 1 is priced so it stands on its own. The most common finding in measurement work is not that a business needs sophisticated infrastructure; it is that the existing setup is misconfigured in two or three specific places. Telling you that is more useful than selling you a server container you will pay to run indefinitely.

Related