Multi-channel attribution from the bottom up — a practical 2026 guide

Top-down attribution (UTMs + analytics + Google Analytics 4) tells you what you already know. Bottom-up attribution (asking customers, modeled marketing mix, contact-graph reconstruction) tells you what you don't. Here's how to do both.

Why top-down attribution is broken in 2026

Three factors collapsed the old "UTM + cookies + GA" attribution model:

1. iOS 14 (and later) tracking permission: ~80% of iOS users decline app tracking. Meta and Google have less data on iOS users. Conversion data has gaps.
2. Third-party cookie deprecation: Chrome started phasing out 3PCs in 2024, completed in 2025. Cross-domain user tracking died for the same-session use case.
3. Dark social: Most discovery happens in private channels (Slack DMs, WhatsApp, Discord, podcast app on phone, etc.) that don't generate referrer headers. Direct traffic numbers are inflated; organic search numbers are deflated.

The result: standard analytics reports show 40-60% of conversions as "direct" or "none/(none)". That's not useful for budget allocation.

The bottom-up alternative

Bottom-up attribution starts at the customer (the "bottom" of the funnel) and works backwards. Three techniques:

1. Self-reported attribution at conversion

When a customer converts, ask them: "How did you first hear about us?" This is one of the most underrated attribution techniques. Add it to the post-checkout / post-signup flow.

The trick: open-text first, then pre-filled buckets. "A friend recommended you" is more useful than "Other".

What we've learned at Bamberg Digital across self-reported attribution data on 300+ leads:

• Self-reported attribution disagrees with last-touch analytics 60% of the time.
• The biggest disagreement: word-of-mouth + podcast mentions show up in self-reported but barely register in analytics.
• Self-reported "Google search" is more reliable than analytics organic search (analytics misses 50%+ of organic touches due to direct attribution).

2. Marketing mix modeling

Marketing mix modeling (MMM) is regression analysis on spend → outcome at the channel level. Weekly or monthly resolution. For each week, you have:

• Spend per channel (Meta Ads, Google Ads, content, podcast, etc.)
• Outcome (conversions, revenue, signups)

Regress outcome on spend with appropriate lags (Meta has 1-day lag, content has 3-12 month lag, podcast has 1-3 month lag). The coefficients tell you marginal revenue per dollar of spend per channel.

MMM doesn't require user-level data. It works on aggregate spend + outcomes. It's robust to iOS 14 / cookie loss. The trade: you need 6-12 months of weekly data to get reliable coefficients, and the model can't tell you about a specific customer's journey.

Conduit Analytics ships a basic MMM module that auto-fits a regression on connected ad accounts + revenue data. Good starting point; not a replacement for a dedicated MMM tool at enterprise scale.

3. Contact graph reconstruction

The most powerful bottom-up technique: reconstruct the customer's journey from the contact graph. Conduit's unified contact graph captures every touch across channels — when a customer's email matches an Instagram DM matches a form submission matches a calendar booking, you have a full timeline.

From that timeline, you can answer questions analytics can't:

• What was the first touch? (Not last touch)
• What was the assisting touch? (The thing in the middle that increased conversion probability)
• How long was the consideration period?
• Which channels appeared multiple times before conversion?

This is multi-touch attribution at the contact-graph level. It requires identity resolution across channels (email = same person as Instagram DM), which is exactly what Conduit's unified contact graph does.

The hybrid approach we use at Bamberg Digital

Combine all three:

1. Self-reported at conversion (asked at signup / checkout). Captures dark social and word-of-mouth.
2. Contact graph reconstruction for the captured touches. Captures multi-touch journeys.
3. MMM for channel-level marginal ROI. Captures the long-tail effects (content, podcast, brand) that don't map to specific contacts.

Each technique answers different questions. Together, they give a more accurate picture than any single method.

What to do with the data

Bottom-up attribution data should drive three decisions:

Budget allocation

Which channels have the highest marginal ROI per dollar of spend? Allocate more budget there.

Common surprises:

• Podcast sponsorships often outperform Meta Ads on ROI (when you measure correctly).
• Long-tail content has higher cumulative ROI than short-form social.
• Word-of-mouth referrals are often the #1 channel by revenue, despite being free.

Customer profile understanding

What does the typical customer journey look like? How many touches before conversion? What's the first touch usually? What's the closing touch?

This informs nurture sequence design. If most customers see 5-7 touches before converting, your nurture sequence should be 8+ touches.

Content strategy

What content shows up most often in pre-conversion journeys? Make more of it. What content shows up rarely? Sunset it or relaunch.

Tools and infrastructure

For agencies and SaaS at $1M-$10M ARR:

• Conduit (or HubSpot / similar) for contact graph + self-reported attribution capture
• GA4 for top-down baseline (still useful as a sanity check)
• Conduit Analytics MMM module for basic marketing mix modeling
• Custom dashboards in Looker / Metabase / Conduit Analytics for cross-method reconciliation

For enterprises at $10M+ ARR:

• Dedicated MMM tool (Recast, Lifesight, or in-house Bayesian model)
• Customer data platform (Segment, mParticle)
• Custom attribution engineering

Common mistakes

1. Trusting last-touch attribution. Last-touch is the worst attribution model. Use it only as the worst-case lower bound.
2. Skipping self-reported attribution. One question at checkout captures more truth than a year of analytics.
3. Over-optimizing on attributed channels. If Meta Ads gets 30% of attributed conversions, that doesn't mean 30% of growth depends on Meta. Most channels have synergistic effects.
4. Ignoring lag. Content posted 6 months ago is still influencing conversions today. MMM with appropriate lags captures this.
5. Believing the data is perfect. Attribution is always lossy. Use multiple methods and triangulate.

Bottom line

Top-down attribution is broken in 2026. Bottom-up attribution — self-reported + contact graph + MMM — gives a more honest picture of what's actually working. Combine all three for the best signal.

Conduit's contact graph + Analytics module is built for this. See the module or start a 14-day trial.