Marketing data & attribution consulting
Your campaigns probably aren't the problem. Your data is.
Get a free tracking health check
Give us your URL. We'll tell you what's working, what isn't, and what it means for your marketing data. No form, no commitment, no call required.
Attribution that doesn't add up
First-click, last-click, and platform-reported numbers all disagree — and nobody can say which one is right.
"New" customers who aren't new
The same customers get counted as new more than once, so your acquisition numbers are inflated, your real retention rate is hidden — and your conversion rate looks worse than it is, because you're dividing real conversions by an artificially large user count.
Spend decisions on guesswork
Budget moves between channels on numbers no one validated — because checking the data sits between the marketing team and the engineers, so it falls to neither.
Your organic numbers include paid influence nobody's measured
Paid media drives a portion of sessions that register as organic or direct. Without regression modelling that effect is invisible, unattributed, and quietly distorting your channel ROI.
How it works
Five phases, start wherever it makes sense
Most engagements start with the audit. What it finds usually determines what comes next — some clients fix one thing, others go deeper. Nothing is sold as a bundle up front, because the findings should drive the work, not the other way around.
-
00
Free tracking health check
We review your site's tracking configuration and funnel structure from the outside. No data access required. You receive a written report within three to five working days.
Free -
01
Data & Attribution Audit
A fixed-price diagnostic. I map your existing tracking, find the gaps and errors, and hand you a findings report with a prioritised roadmap — no commitment beyond this.
-
02
Identity & Tracking Foundation
Stitching users across GA4 user IDs, session IDs, canonical IDs, and event platforms like Snowplow — heuristic probabilistic matching by default, with graph-based resolution (Neo4j) where the identity graph is too complex for heuristics alone — so new-vs-returning classification is actually accurate.
-
03
Customer Behaviour Modelling
Conversion curves, cohort survival analysis, and funnel drop-off modelling — a real model of how customers move through your funnel, not a dashboard estimate.
-
04
Attribution Modelling
Shapley value attribution and regression-based modelling of paid impact on non-paid channels, benchmarked against your existing first/last-touch numbers.
-
05
Ongoing Retainer
Monthly attribution refresh, funnel health monitoring, and anomaly detection — so drift gets caught before it costs you a quarter of misallocated spend.
Case Study
What a client discovered when we looked underneath their campaigns
A marketplace spending £30k/month on paid social asked a straightforward question: is it working?
The answer was more complicated — and more valuable — than they expected.
Their attribution was lying to them
Last-touch reporting showed paid social contributing around 5% of conversions. Shapley value attribution told a different story: paid had the highest marginal contribution of any channel, meaning removing it would cause conversion rates to fall materially across the entire funnel — including channels that appeared to have nothing to do with paid.
Their organic numbers included paid influence they'd never measured
Regression modelling showed that ~5% of non-paid sessions would not have happened without influence from paid. Paid was quietly subsidising their organic and direct numbers. Nobody had quantified it before.
Their identity data was inflating acquisition numbers — and distorting every metric built on them
After applying a graph-based identity stitching algorithm across multiple user identifiers, the distinct user count fell by 25%. A quarter of "new users" weren't new.
More significantly, once the true new user count was established, first visit timestamps could be accurately assigned — previously, returning users misclassified as new were anchoring journey start dates incorrectly. This changed conversion curve shape, time-to-first-action metrics, and cohort survival analysis across the board. It also revealed that return visit rates were far higher than previously measured — users were coming back much faster than the data suggested. What had looked like CRM-driven re-engagement turned out to be an artefact of last-touch attribution: CRM was getting credit for closing journeys it hadn't started or sustained. Demand was already there. CRM was just the last door users walked through.
Their fastest-converting users were being measured with the wrong metric
Paid-acquired users converted to a first action in an average of 70 days — 50 days faster than organic. Shifting to a first-touch lens told a different story: Paid's contribution to conversions jumped from 5% to 15%. The channel wasn't underperforming. It was being measured from the wrong end of the journey.
What changed
Channel performance targets were realigned to funnel position rather than last-touch conversion. Paid was evaluated on first-touch contribution and incremental uplift rather than direct conversion share. Budget allocation decisions became defensible rather than intuitive.
The ROI on the analysis significantly exceeded the cost of it in the first month.
About
Marketing owns the numbers. Engineering owns the pipeline. I work in the gap between them.
I'm Damien Harley, a data and analytics engineer with 6+ years building end-to-end data platforms — pipelines, warehouses, orchestration, and the analytical products on top of them.
Across every business I've worked in — growth-stage to established marketplaces — the analytical work ended up in the same place: in front of a CMO or CEO, changing how budget got allocated or how performance got measured. And the analysis itself was usually sound. The problem was what it was built on.
That problem has a predictable shape. Half the knowledge needed to fix marketing measurement sits with the paid team, half sits with the engineers, and almost no one holds both. Marketers can read the numbers but can't see into the pipeline that produces them. Engineers can see the pipeline but aren't the ones being asked whether paid is working. So the data quietly stays wrong — attribution models that don't reflect how customers actually convert, identity data that inflates acquisition and drags the conversion curve earlier than reality (because GA4 user IDs, session IDs, and Snowplow event identifiers were never stitched into a consistent view of who a user is), paid judged on last-touch when its real job was top-of-funnel acceleration. Not because anyone failed — because fixing it needs one person standing in the gap.
That's the job I do.
Get in touch
Start with a straight answer.
Give us your URL. We'll review your tracking configuration, walk your funnel, and come back with a written report within three to five working days. If there's more to find, the audit is where we find it.