Reducing CAC by 35% Through Proper Marketing Attribution

A $20M D2C brand was burning ad budget on the wrong channels because last-click attribution told a misleading story. Multi-touch attribution revealed the truth — and saved them 35% on customer acquisition costs.

The Challenge: Spending More, Understanding Less

The marketing team was spending $4.2M annually on paid acquisition across Google Ads, Meta, TikTok, influencer partnerships, and affiliate programs. The budget was growing 30% year-over-year. But the CMO had a nagging suspicion: "We're probably wasting half our ad spend — we just don't know which half."

The root cause was a familiar one:

• Last-click attribution only. Google Analytics was the single source of truth, and it credited 100% of each conversion to the last touchpoint. This massively over-valued branded search and retargeting while under-valuing awareness channels that initiated the customer journey.
• Channel silos. Each ad platform reported its own numbers, and they didn't agree. Meta claimed 2,400 conversions in February; Google claimed 1,800; the actual number was 1,100. The double-counting problem made it impossible to calculate true ROAS.
• No cross-channel view. The team had to manually pull data from 6 different dashboards to build a weekly report. The process took a full-time analyst 8 hours every Monday — and the report was already stale by Tuesday.
• Over-investment in bottom-funnel. Because last-click attribution rewarded retargeting and branded search, 65% of the budget went to bottom-funnel tactics. The team was essentially paying to convert customers who would have converted anyway — while starving the top-of-funnel channels that actually drove new demand.
• Inability to prove ROI of content marketing. The content team produced high-quality educational content that drove significant organic traffic, but because these visitors rarely converted on first touch, the content program was perpetually under-funded and under-appreciated.

The board had approved a 40% marketing budget increase for the next fiscal year, but the CMO refused to spend more until she could prove where the current budget was actually working.

Our Approach: Build the Data Foundation, Then Reattribute

We designed a 6-week engagement to solve the attribution problem from the ground up — not by bolting another tool onto a broken data pipeline, but by building proper infrastructure.

Week 1: Data Audit & Source Mapping

We mapped every marketing data source — ad platforms, website analytics, CRM, email marketing, affiliate networks, and the e-commerce platform. We documented how data flowed (or didn't flow) between systems and identified every point where attribution broke down.

The audit revealed 23 distinct data quality issues, including UTM parameter inconsistencies, missing click IDs, timezone mismatches between platforms, and a JavaScript tracking error that had been silently dropping 12% of conversion events for three months.

Week 2–3: Marketing Data Warehouse

We built a centralized marketing data warehouse in BigQuery, pulling data from all sources through automated pipelines:

• Fivetran connectors for Google Ads, Meta Ads, TikTok Ads, Shopify, and Klaviyo
• Custom extraction scripts for the affiliate platform and influencer tracking system
• dbt transformation layer — standardized naming conventions, deduplication logic, and a unified customer journey model that stitched touchpoints across channels
• Identity resolution — matching anonymous website visitors to known customers using first-party cookies, email hashes, and device fingerprinting (privacy-compliant)

Week 3–4: Multi-Touch Attribution Model

We implemented a data-driven multi-touch attribution model using the Shapley value approach — a game-theory method that assigns credit to each touchpoint based on its marginal contribution to conversion probability.

Why Shapley over simpler models (linear, time-decay, position-based)? Because the client's customer journey was complex — averaging 7.3 touchpoints across 3.2 channels over 14 days — and simpler models introduce systematic biases that can be worse than last-click.

The model was trained on 18 months of historical data and validated against holdout sets. We also built a marketing mix model (MMM) as a complementary top-down view, useful for channels where user-level tracking is limited (podcasts, OOH, influencer).

Week 5: Unified Dashboard & Reporting

We built the marketing team's new command center in Looker:

• Channel Performance Dashboard — real-time spend, CPA, ROAS, and attributed revenue by channel with multi-touch attribution
• Customer Journey Explorer — visualize the most common paths to conversion, identify high-value journey patterns
• Budget Allocation Optimizer — scenario modeling tool showing expected revenue impact of shifting budget between channels
• Weekly Automated Report — replacing the 8-hour manual process with a scheduled email hitting inboxes every Monday at 7 AM

Week 6: Budget Reallocation & Optimization

The attribution model revealed a dramatic misallocation:

• Branded search was over-credited by 340%. It was capturing demand, not creating it. We recommended cutting branded search spend by 45%.
• Content marketing was under-credited by 280%. Blog posts and educational guides were the first touchpoint for 38% of eventual converters — but received zero credit under last-click.
• TikTok ads were the highest-ROI paid channel. Despite being dismissed as "awareness only," TikTok was initiating customer journeys that converted at 2.4x the rate of Google Display.
• Retargeting had severe diminishing returns. The optimal retargeting frequency was 4–6 impressions; the team was averaging 18 impressions per user, wasting 60% of retargeting spend.

We worked with the marketing team to implement phased budget reallocations, shifting spend from over-attributed bottom-funnel channels to under-attributed awareness and consideration channels.

Key Deliverables

• Marketing Data Warehouse — BigQuery + dbt, pulling from 8 data sources with daily automated refresh
• Multi-Touch Attribution Model — Shapley value model trained on 18 months of data, validated for accuracy
• Marketing Mix Model — complementary top-down model for channels without user-level tracking
• Unified Marketing Dashboard — real-time channel performance, customer journey visualization, budget optimizer
• Budget Reallocation Plan — specific, phased recommendations with expected impact modeling
• Automated Weekly Report — replacing 8 hours of manual work per week

Results

• 35% reduction in CAC — blended customer acquisition cost dropped from $47 to $31 within 90 days of budget reallocation
• 2.1x ROAS improvement — overall return on ad spend improved from 1.8x to 3.8x on the reallocated channels
• $2M annual ad spend optimized — budget reallocated from under-performing to high-performing channels without increasing total spend
• 8 hours/week saved — automated reporting eliminated the manual Monday data pull entirely
• Content marketing budget doubled — with proper attribution proving content's role in the conversion path, the team secured additional investment
• TikTok budget 3x increase — the attribution model proved TikTok's contribution, unlocking budget from skeptical leadership

"We thought we understood our marketing performance. We were wrong. The attribution model showed us that our 'best' channel was actually our most wasteful, and the channel we were about to cut was our most efficient. This engagement literally saved us from making a $1M mistake in budget allocation."

— CMO, D2C Health & Wellness Brand

Why This Approach Works

• Infrastructure first, insights second. Bolting an attribution tool onto messy data gives you confident-looking wrong answers. We fixed the data pipeline before building the attribution model.
• Multiple attribution methods. No single attribution model is perfect. Combining bottom-up (Shapley) with top-down (MMM) provides triangulated confidence in budget decisions.
• Actionable output. The budget allocation optimizer doesn't just tell you what happened — it models what would happen if you shift spend, so the team can make decisions with confidence.
• Phased reallocation. We didn't recommend cutting 45% of branded search overnight. Budget shifts were implemented in 4 phases over 8 weeks, with measurement gates at each stage to validate the model's predictions.