Attribution & Measurement: How to Actually Find What Drives Revenue in 2026

Attribution & Measurement: How to Actually Find What Drives Revenue in 2026

Attribution is the single most broken thing in modern marketing.

iOS 14 destroyed user-level tracking in 2021. Multi-touch attribution models output confidently wrong numbers. Last-click attribution overstates direct response by 40-60% and undercounts brand by the same. And every platform reports its own version of "the truth" that adds up to 280% of actual revenue when you sum them.

The brands that figured out measurement in 2026 aren't using better attribution tools. They're using a combination of incrementality testing, marketing mix modelling, and cohort analysis to triangulate what's actually working. The first half of this page walks through that combination. The second half indexes every deep-dive piece we've written.

Why platform attribution lies (and how badly)

Three structural problems make platform-reported attribution unreliable in 2026:

iOS 14 and Apple Intelligence broke user-level tracking. Since 2021, 70-80% of cross-domain tracking signal for iOS users is gone. Conversion API helps but doesn't fully close the gap. Platforms fill the gap with modelled conversions, which are educated guesses dressed up as data.

Every platform claims credit it doesn't deserve. Meta's attribution window includes 7-day view-through. Google's includes data-driven attribution that disproportionately weights its own touchpoints. TikTok counts a 2-second view as engagement. Sum every platform's reported revenue and you get 200-280% of actual revenue. Each platform thinks it drove 30-40% of revenue. Math says only 100% exists.

Last-click attribution overstates direct response by 40-60%. The last platform the user touched before converting gets all the credit. Brand campaigns, awareness video, organic content that warmed up the prospect get zero credit. This produces over-investment in last-click channels (typically Google search and retargeting) and under-investment in awareness channels that compound.

The aggregate result: platform-reported ROAS overstates reality by 30-60% in most accounts. Brands optimising against those numbers are pointed at the wrong targets. Their channel mix decisions are guesswork even when the dashboards look authoritative.

Layer 1: The honest top-line truth (blended CAC + MER)

Before any channel-level attribution, every marketing team needs two numbers that don't depend on attribution working at all:

Blended CAC = total marketing spend / total new customers acquired in the period. Includes all paid spend, agency fees, tooling, content production, even partially-attributed salaries. The honest number. Usually 30-50% higher than the sum of platform-reported CAC numbers because platforms over-attribute.

MER (Marketing Efficiency Ratio) = total revenue / total marketing spend. The headline metric that doesn't depend on tracking. Doesn't get gamed. Doesn't have iOS 14 problems. If MER is trending down quarter-over-quarter while platform-reported ROAS holds steady, something is wrong with the attribution and the truth is hiding in the blended numbers.

The brands that figure out measurement watch MER weekly. It's the canary. When it drops, they dig into why before scaling further. Brands that don't watch MER scale unprofitably for months before they notice the bank balance is off.

Layer 2: Channel-level signal (server-side tracking + MTA)

Once the top-line is honest, channel-level attribution gets less critical, but you still want directional signal for tactical decisions.

Server-side tracking with Conversions API. Pushes conversion events server-side so platforms get them even when client-side pixels fail. Required for healthy match rates (70%+ on Meta Events Manager). Brands with proper CAPI see 25-40% lower CAC than the same brand running client-side only. Foundation move before any multi-touch attribution work.

Multi-touch attribution as directional signal. Tools like Triple Whale, Northbeam, Polar try to reconcile platform-reported numbers into one view. Useful for understanding channel-level patterns. Less useful as ground truth because the underlying data is still incomplete. Treat MTA output as 30-50% directional, not absolute.

Customer Match audiences and lookalikes. Upload first-party customer data to platforms. Higher match rates, better optimisation, more accurate measurement of repeat customer behaviour. Brands without first-party data infrastructure pay 30-50% higher CACs in 2026 because they're flying blind.

Layer 3: The gold standard (geo holdout incrementality testing)

If you take one thing from this entire pillar, it should be this: run quarterly geo holdout tests on your biggest paid channel.

What it is. Pick a channel (e.g. Meta paid social). Pick two comparable regions (e.g. UK metros vs UK rest-of-country, or US East Coast vs US West Coast). Pause spend in one region for 4 weeks while continuing normal spend in the other. Compare conversion rates. The difference is the channel's true incremental contribution.

What it costs. 4 weeks of "lost" revenue in the holdout region (typically 10-15% of total revenue during the test period). Less than you think because some of those conversions happen anyway through organic + direct. The cost of the test is usually 2-4% of monthly revenue.

What you learn. Most brands we audit have never run one. The first time they do, they discover one of two things: a channel they thought was great is contributing 30-50% less than reported (over-investing), or a channel they thought was meh is genuinely driving incremental conversions (under-investing). Either way, the reallocation is usually worth a 15-25% blended MER improvement within 60-90 days.

Run one per quarter. By the end of year one you've validated your 4 biggest channels. By year two you have a measurement-validated channel mix that most competitors don't.

Layer 4: Cohort LTV by acquisition channel

Channel-level ROAS misses something critical: different channels acquire customers with different long-term value.

The math. Channel A has £30 CAC and customers buy 4 times in their first year at £45 gross profit per order. Year-1 LTV: £180. LTV:CAC of 6:1. Healthy. Channel B has £30 CAC and customers buy 1.2 times in year one at £45 gross profit. Year-1 LTV: £54. LTV:CAC of 1.8:1. Loss-making after marketing + ops overhead.

Same first-purchase ROAS. 3x different value. If you're scaling both channels equally, you're scaling unprofitable revenue alongside profitable revenue and dragging blended unit economics down.

How to track it. Tag every customer with their acquisition channel at first purchase (UTM in Shopify, custom field in CRM). Pull cohort gross profit at 90, 180, 365 days. Build a Looker Studio dashboard. Review quarterly. Reallocate budget toward channels with higher 365-day LTV, even if first-purchase ROAS looks similar across channels.

The compounding benefit. Brands that track cohort LTV by channel make completely different scaling decisions over 18-24 months than brands that don't. The differences in unit economics show up brutally in cash position by year two.

Below: the deep-dive cluster pieces, organised by layer.

The strategic frame

Before you fix your measurement stack, you need to understand what's actually broken and why platform-reported numbers can't be trusted.

The gold standard: incrementality testing

Geo holdout tests are the most underused weapon in marketing. They're slower than clicking around in dashboards, they require discipline, and they're the only way to know what your paid spend is actually doing.

  • Geo holdout test: step-by-step playbook. How we run incrementality tests for clients. Includes how to pick comparable regions, how long to run, how to interpret results, and how to defend the methodology when the CMO pushes back.

Cohort analysis: where the unit economics live

Channel-level ROAS lies. Customer cohort behaviour over 12-24 months is where the truth lives. Two channels with identical first-purchase ROAS can have wildly different LTV. The difference between a brand that scales profitably and one that hits a £10M plateau is usually whoever figured this out first.

The honest summary

Attribution in 2026 is not a tooling problem. It's a methodology problem.

The brands getting it right run a layered stack: platform data for fast signal, MTA for directional channel-level numbers, MMM for top-down validation, incrementality tests quarterly to ground-truth everything else, and cohort analysis to make sure the channels they're scaling actually produce profitable customers.

If you want help building this for your business, we offer a marketing measurement audit. We'll diagnose what your current attribution is actually telling you (vs what you think it is), and give you a roadmap to fix it.

Tools we built for this

Related pillars