ROAS consultant: pilot at margin, data-driven attribution, dashboards

ROAS (Return on Ad Spend) as displayed by default in Google Ads and Meta is a gross revenue ROAS, which doesn’t account for product margin or return rate. A ROAS of 4 at 60% margin is far more profitable than a ROAS of 6 at 15% margin — yet the ad algo will prefer the second, because it doesn’t know margins differ. My role as a ROAS consultant is to fix the measurement: push real margin to ad platforms, cross Ads spend with returns and refunds, build multi-source dashboards that actually inform budget decisions. ROAS dashboard fee: €3,600 for ~6 consulting days.

Why gross ROAS lies

Three structural biases:

No view on product margin. Gross ROAS = tracked revenue / ad spend. If your product categories have margins from 15% to 65%, the algo optimizes blindly. A push on a low-margin promotion category can show high gross ROAS but destroy real profitability.

No view on returns. On fashion/luxury sites, return rates can hit 25-40%. A €200 order returned 60% leaves €80 net revenue. Gross ROAS counts the €200. The algo keeps driving traffic to heavily-returned product pages.

No view on incremental margin. A sale made without paid media (existing customers, organic SEO) also counts in attributed Ads ROAS due to attribution. Incremental ROAS (the sale that ads actually generated) is typically 30-60% of gross ROAS on Google Ads, 40-70% on Meta.

Standard missions

1. Product margin via Google Ads OCI. Push real margin (not revenue) as Conversion Value via Offline Conversion Import. Algo learns to target profitable products. See offline conversions for pipeline.

2. Net ROAS dashboard (Looker Studio + BigQuery). Multi-source dashboard crossing: ad spend per channel/campaign, GA4 tracked revenue, back-office or ERP margin, returns and refunds, computed gross + net + estimated incremental ROAS.

3. Activated data-driven attribution. GA4 DDA activation, BigQuery exports, custom attribution models if DDA insufficient (Markov, Shapley), Looker Studio dashboards. See GA4 + BigQuery.

4. ROAS per customer cohort (LTV). For sites with strong retention (subscription, repeat buy), 90-day ROAS can be far below 365-day cohort ROAS. Build LTV model that pushes expected cohort value to Google Ads.

Typical dashboard architecture

Sources                    BigQuery                Looker Studio
 │                           │                        │
 GA4 events ───────────────▶│                        │
 Google Ads API ───────────▶│  views                 │
 Meta Ads API ─────────────▶│  ▶ roas_per_channel    ├──▶ Dashboard
 Back-office (margin/returns)▶│ ▶ roas_per_product   │
 CRM (LTV cohorts) ────────▶│  ▶ roas_cohort         │
                             │  ▶ margin_per_category │

Sources connected via native Looker Studio extensions (Google Ads, Search Console), Fivetran/Airbyte (Meta, LinkedIn, TikTok), custom Cloud Functions (back-office, CRM).

Concrete case

Fashion/lifestyle retailer ~€10M tracked revenue, displayed gross Google Ads ROAS of 4.5. Mission: audit to identify gross vs net gap.

Work done: product margin extraction from back-office per category, cross-reference with attributed Google Ads sales over 90 days, factor in returns, push margin as Conversion Value via OCI.

Analysis result: real net ROAS = 2.1 (vs 4.5 displayed). The “premium promos” segment showed gross ROAS 6 but net ROAS 0.4 (loss). Budget reallocation to segments with margin above 40% in 6 weeks. At constant budget, total Google Ads margin +38% at 90 days.

Frequently asked questions

Time to deploy?

Standard ROAS dashboard: 6 consulting days, 3-4 calendar weeks. Multi-source complex (4+ ad platforms, custom back-office): 8-10 days, 6 weeks.

No access to product margin?

For sites where margin is confidential or not systematically tracked, proxy possible: average category margin (internal estimate), target margin defined by management. Imperfect but better than gross revenue.

Dataviz tools used?

Looker Studio default (free, native BigQuery). Power BI for Microsoft shops. Metabase for open-source. Tableau rarer.