Attribution is broken. Here's how to actually rebuild it.

If you ask ten marketing leaders what their attribution numbers say, you'll get ten different answers — and most of them will be wrong by 30 to 50 percent.

This isn't an indictment of any specific team. It's the state of marketing measurement in 2026. iOS 14 broke last-click attribution. Third-party cookie deprecation broke cross-site tracking. Ad blockers broke client-side conversion data. Privacy regulations broke certain user-level tracking entirely. And on top of all that, every advertising platform reports its own ROAS based on its own attribution model — none of which match each other or reality.

The result: most marketing decisions get made on data that's wrong by a wide margin. CMOs scale spend on channels that look profitable but aren't. They cut budget from channels that look weak but are actually working. They debate dashboards in leadership meetings without anyone in the room knowing which dashboard is right.

This is fixable. Not easily, but the work is well-defined. Here's what's actually broken in modern marketing attribution, what it takes to rebuild it correctly, and the order to do the work in.

What's actually broken (and why).

To rebuild attribution, you have to understand the specific failures. They're not all the same problem.

1. iOS 14 broke client-side conversion tracking on iOS.

App Tracking Transparency (ATT) introduced with iOS 14.5 means most iOS users opt out of cross-app tracking. Meta, Google, and other ad platforms can't reliably attribute iOS conversions to the ads that drove them. This is permanent, not a temporary issue. The fix is server-side conversion APIs — Meta's CAPI, Google's Enhanced Conversions, and equivalents from other platforms. Most brands have implemented some of these incompletely.

2. Third-party cookies are functionally dead.

Safari blocks them by default. Firefox blocks them. Chrome is in the long process of removing them. Even where third-party cookies still work, ad blockers and privacy extensions break them on a meaningful share of traffic. Cross-site tracking that depended on third-party cookies — most ad platform attribution, most retargeting attribution, most multi-step funnel tracking — is fundamentally less reliable than it was three years ago.

3. Ad platforms over-report ROAS.

Every ad platform's attribution model is tilted toward attributing conversions to itself. Meta claims credit for conversions that Google deserves. Google claims credit for conversions that paid social influenced. Both claim credit for conversions that organic, email, or word-of-mouth would have produced anyway. The over-reporting is consistent — usually 30 to 50 percent above actual incremental revenue contribution. Brands that scale spend based on platform-reported ROAS scale into channels that aren't actually profitable.

4. Last-click is wrong (and has been for years).

Most marketing analytics tools default to last-click attribution. The model gives 100% of conversion credit to the last ad or organic channel a buyer interacted with before converting. For B2B, where buying journeys span months and dozens of touchpoints, this is comically wrong. Even for DTC, where journeys are shorter, last-click systematically over-credits direct traffic and search retargeting.

5. Multi-touch attribution requires data most teams don't have.

Multi-touch attribution models — that distribute credit across multiple touchpoints — require clean event data, server-side tracking, and proper user identification across channels. Most marketing teams don't have this infrastructure. They have GA4 with default settings, half-implemented server-side tracking, broken cross-domain tracking, and no warehouse where the data could be modeled properly even if it were clean.

What rebuilding attribution actually looks like.

The good news: this is solvable. The infrastructure to do attribution correctly in 2026 exists. The challenge is implementing it in the right order, because skipping steps means optimizing on still-broken data.

The right order is roughly: server-side tracking → clean event taxonomy → multi-touch modeling → BI dashboards → CRO programs on top.

Step 1: Server-side tracking implementation.

This is the foundation. Server-side Google Tag Manager, Meta Conversion API, Google Enhanced Conversions, and equivalents for other platforms. The goal is to get conversion data from your server (where you have ground truth) directly to ad platforms (where you need it), bypassing the privacy-restricted client-side tracking that's increasingly broken.

What this looks like in practice: a server-side GTM container running on a subdomain you own. Conversion events fired from your backend (after a Stripe webhook, a CRM update, a product event) directly to a server-side tag, which then forwards them to Meta, Google, and other platforms with proper user identification (hashed email, phone, etc.) and event matching.

Implementation time: 4-8 weeks for most B2B sites, longer for complex e-commerce. Most teams have implemented part of this — usually CAPI for one event — but not the full event taxonomy.

Step 2: Clean event taxonomy.

Tracking the right events with the right properties. This sounds basic but is where most teams have technical debt. Inconsistent event names. Missing properties. Events firing twice. Events not firing at all on certain pages. Conversion events tracked at the wrong moment in the funnel.

A clean event taxonomy has named events for each meaningful funnel step (page_view, lead_form_submit, demo_booked, trial_started, paid_conversion), consistent property schemas across events (user_id, session_id, source, campaign), proper deduplication, and validation that events are actually firing across all surfaces.

Tools: Segment, RudderStack, or a custom event pipeline. Output: events flowing reliably into your warehouse with consistent structure.

Step 3: Data warehouse for marketing data.

Once events are flowing, they need somewhere to go. The right destination is a data warehouse — BigQuery, Snowflake, or equivalent — where event data, ad platform data (via Fivetran or Airbyte), CRM data, and product analytics can be joined and modeled together.

Without a warehouse, attribution analysis is locked to whatever model GA4 or Mixpanel provides. With a warehouse, you can model attribution however you want — and you can join marketing data to revenue data, which is where actual ROAS lives.

Implementation: BigQuery or Snowflake (both work, pick whichever your engineering team prefers). dbt for transformations. Fivetran or Airbyte for pulling data from ad platforms, CRM, and product tools.

Step 4: Multi-touch attribution modeling.

This is where the real work happens. With clean events in a warehouse, you can run actual multi-touch attribution models that reflect how buyers convert in your business — not the default last-click that comes with most tools.

The simplest useful model: time-decay attribution, where credit is distributed across touchpoints with more recent touches getting more credit. This corrects most of the worst issues with last-click without requiring the heavy machine-learning models more sophisticated approaches use.

For more sophistication: position-based attribution (40% to first touch, 40% to last touch, 20% distributed across the middle), or shapley-value attribution (which uses game theory to allocate credit), or fully custom models trained on your specific funnel.

The model matters less than the fact that you're using any model that distributes credit across touches. Anything is better than last-click for B2B.

Step 5: BI dashboards executives actually use.

The output of all the above is dashboards that show what's actually happening — refreshed daily, joined to revenue data, with attribution that reflects reality.

The dashboards that get used by executives have a few specific characteristics:

• They show revenue and pipeline contribution by channel, not just spend and clicks
• They use the team's modeled attribution, not platform-reported numbers
• They're built on the warehouse, so they don't require waiting for someone to export Excel files
• They have anomaly detection and trend explanations, not just static charts
• They're available to the people who need to see them, with proper access control

Tools: Looker, Metabase, Hex, Mode. Pick based on your team's preferences and budget. All of them work if implemented properly on top of a clean warehouse.

The ROI of doing this right.

Marketing leaders who haven't done this work usually ask: "Is this really worth the engineering investment?" The answer, in our experience, is yes — for any team spending more than $50K/month on paid media or running multi-channel marketing programs.

Here's why: marketing teams making decisions on accurate data outperform teams making decisions on inaccurate data. The size of the advantage depends on spend and complexity, but in the engagements we've worked through:

• One B2B SaaS team rebuilt attribution and reallocated $30K/month away from a channel that platform reports said was profitable but actually wasn't. Pipeline went up 20% on the same total spend.
• A DTC brand restored proper iOS conversion tracking via CAPI. Their reported ROAS came back in line with reality, which let them confidently scale spend that they'd been holding back from because the numbers looked uncertain.
• A fintech team built multi-touch attribution that finally captured the long sales cycle. Several channels that looked weak under last-click — content marketing, founder LinkedIn — turned out to be among the strongest pipeline drivers when measured properly.

The pattern: rebuilding attribution doesn't always reveal positive surprises. Sometimes it reveals that channels you thought were working aren't. But knowing the truth — even when it's painful — is always better than scaling spend on broken numbers.

The mistakes to avoid.

A few specific failure modes we see often:

Skipping server-side tracking and going straight to attribution modeling.

This is the most common mistake. Teams want sophisticated attribution but skip the foundation. The result is sophisticated models running on bad data. Garbage in, garbage out — but with more decimal places.

Trusting one tool's attribution as the source of truth.

Triple Whale, Northbeam, Polar Analytics, Mixpanel, Amplitude — all of them produce different numbers. None of them match exactly. The right approach is to use them as inputs to your own modeled view, not to pick one and treat its output as truth.

Running attribution and CRO in parallel before the foundation is fixed.

CRO is downstream of measurement. If your conversion tracking is broken, your A/B test results are unreliable. Fix tracking first, then run experiments.

Treating attribution as a one-time project.

Attribution infrastructure decays. Pixels break. Events stop firing. New ad platforms get added without proper tracking. Privacy regulations change. Treating attribution as "we did the project, now it's done" is how teams end up with broken numbers six months later. Ongoing maintenance and audits are essential.

Where to start.

If you're a marketing leader looking at your dashboards and not trusting the numbers, three concrete first steps:

1. Run a server-side tracking audit. Check whether you have CAPI implemented for Meta, Enhanced Conversions for Google, and server-side conversion firing for your other major platforms. Most teams find at least one major gap.
2. Compare platform-reported revenue to your actual revenue. Sum up "revenue attributed" across all your ad platforms over the last quarter. Compare to your actual closed revenue. If the platforms claim 1.4x or more of your actual revenue, your over-attribution problem is real and quantified.
3. Check whether you have a data warehouse. If your answer is "we use GA4," the answer is no. A real warehouse means BigQuery or Snowflake (or equivalent) with regular data ingestion from your tools. This is the foundation everything else gets built on.

Done honestly, these three audits will tell you whether your attribution is solid or broken — and if it's broken, what to fix first.

This is the work we do at Geo Solutions in our Analytics & CRO service. But the broader point: marketing leaders who don't trust their numbers can't make confident decisions. Rebuilding attribution is one of the highest-ROI engineering investments a marketing team can make. The teams that do this work outperform the teams that don't. The gap is real, and it's widening.