Multitouch marketing attribution in LATAM: how 12 brands took ROAS from 1.5x to 4.2x

One-minute summary

Last-click died a while ago. Most LATAM PYME performance teams still pay for the “win” of a channel that just stole credit from five others. Lifting a 12-brand beauty portfolio from a 1.5x ROAS to 4.2x in two quarters is what proves multitouch is no longer optional.

For the busy reader, the points that matter right now:

• Multitouch marketing attribution shares the credit for a conversion across every customer touchpoint — Google, Meta, email, retargeting, organic, WhatsApp — and not just the last click.
• In 2026 the rule-based models (first-click, linear, time-decay, position-based) are deprecated in Google Ads. Only last-click and data-driven attribution (DDA) remain. DDA is also the GA4 default.
• LATAM specifics: 70% of traffic is mobile, iOS share runs 18 to 25% in PE, CO, and MX, ATT opt-in is below 30%, and WhatsApp closes 40 to 60% of B2C deals. Standard cookie tracking captures at most 60 to 75% of the real funnel.
• The cookieless stack — server-side, Enhanced Conversions, Conversions API — is no longer optional. Without it, Meta and Google see 25 to 40% fewer conversions than the ones that actually happened.
• MMM revival: Meta open-sourced Robyn and Google released LightweightMMM. For brands above 10 thousand dollars a month in media spend, MMM is a real complement to DDA.
• Documented case: portfolio of 12 beauty brands, server-side stack, unified attribution model, and MMM calibration. ROAS went from 1.5x to 4.2x in two quarters without raising media spend.

Five years that broke attribution

Multichannel attribution is not a fad — it is the answer to five years of regulatory and technical shifts that hollowed out ad platform reports. A quick recap is the only way to see why a 2019 stack no longer holds.

#1. 2021–2022: ATT and the first crack

In April 2021 Apple turned on App Tracking Transparency (ATT) — a mandatory opt-in for cross-app tracking. Global opt-in settled near 25%. In LATAM, public benchmarks from Mexico, Argentina, and Chile place it between 20 and 28%. Meta lost roughly 10 billion dollars of revenue in 2022 alone, on its own estimate. For advertisers that meant up to 75% of iOS conversions stopped attributing correctly in Facebook Ads Manager without a server-side stack.

LATAM PYMEs felt the change as a paradox. Dashboard ROAS climbed (because some sales fell off the books) while real revenue dropped. Several teams cut spend on “underperforming” channels that were actually working — their contribution simply went invisible.

#2. 2023–2024: GA4 and the death of rule-based models

On July 1, 2023 Universal Analytics stopped collecting data. GA4 launched event-based and data-driven attribution became the default. In September 2023 Google Ads deprecated four rule-based models: first click, linear, time decay, and position based. Only DDA (default) and last click remain.

For most PYMEs the shift happened in the background. Agencies kept reporting last-click ROAS without noticing that Google Ads and Meta were optimizing bids on completely different signals. Reporting and reality drifted apart in silence.

#3. 2024–2026: cookie saga and MMM revival

Google postponed third-party cookie deprecation in Chrome three times. In July 2024 the Privacy Sandbox team announced cookies would not be fully removed — users will choose via a prompt. By 2026 the infrastructure is in place, but fingerprinting and cross-site tracking remain technically possible.

In parallel came the MMM revival. Meta open-sourced Robyn (R-based MMM with a Python wrapper) and Google released LightweightMMM. Brands above 10 thousand dollars a month in media spend started using MMM to measure incrementality as a complement to user-level attribution — not a replacement. This is not a return to 2008: these are modern open-source tools built on the classic Marketing Mix Modeling foundations.

#4. What sets the LATAM market apart in 2026

• Privacy regulation. Brazil has lived with LGPD since 2020 (the local GDPR equivalent). Chile passed Law 21.719 on data protection (in force since 2024). Peru applies Law 29733; Colombia Law 1581; Mexico the LFPDPPP. All require explicit consent for tracking. The soft-compliance window is closed.
• Longer payment funnel. PIX in Brazil, Yape and Plin in Peru, MercadoPago in Argentina and Mexico, Nequi in Colombia, Webpay in Chile. The gap between add-to-cart and confirmed payment runs from 5 minutes to 48 hours — the standard 1-day-click window loses up to 30% of real conversions.
• WhatsApp as a sales channel. In Peru, Colombia, and Mexico more than 60% of B2C deals close inside WhatsApp Business. This is dark traffic: no standard attribution model sees it without a deliberate setup.

Stack and attribution models in 2026

Any attribution model is guesswork if the underlying stack does not capture data correctly. Pipeline first, model second.

#1. The minimum data stack

For a PYME spending under 5 thousand dollars a month, GA4 + GTM client-side + Meta Pixel is enough. Above 10 thousand dollars a month, server-side tracking and Enhanced Conversions are mandatory. Above 50 thousand dollars a month, add an MMM baseline at least once per quarter.

#2. Attribution models: practical cheat sheet

• Last-click — fits direct-response channels (brand paid search, retargeting). Not a full-funnel model.
• First-click — useful for brand and awareness metrics (TV, YouTube, OOH). Deprecated in Google Ads, available in GA4 exploration reports.
• Linear — splits credit evenly. Simple and blind to relative channel strength.
• Time-decay — more credit to touches closer to the conversion. Works for long B2B cycles.
• Position-based (U-shaped) — 40% first touch, 40% last, 20% middle. Useful for product-led funnels.
• Data-driven attribution (DDA) — Shapley value on top of ML. Default in Google Ads and GA4. Minimum: 300 conversions and 3,000 ad interactions in 30 days per conversion action. Details in the official GA4 documentation.
• Markov chain (custom) — via Python or the R package ChannelAttribution. A real option for brands with a proper CDP.
• MMM — channel-level incrementality. Not a replacement for attribution: a complement.

#3. Google Ads and Meta configuration — 2026 reality

Checklist to verify today:

1. DDA enabled in Google Ads (Tools → Measurement → Attribution → Conversion action settings).
2. Enhanced Conversions active — hashed email and phone pass through a server-side tag, not the web pixel.
3. Meta CAPI configured for every priority event. Without it, 25 to 40% of iOS conversions are invisible.
4. Same model across every platform. If GA4 runs DDA and Google Ads still uses last-click, numbers will drift 20 to 40% and the team will argue about which KPI is “real.”
5. Attribution windows tuned to LATAM. For PIX, Yape, and MercadoPago payments, minimum 7-day-click. For high-ticket (above 500 dollars), 28-day-click.

#4. Odoo as the single source of truth

Ad platforms lie — each one keeps credit for itself. The only source of truth is the ERP, where real revenue is recorded. In Odoo the minimum pipeline looks like this:

1. UTM parameters arrive from the landing and are written to utm_source / medium / campaign / term / content on the lead form.
2. When the Sales Order is created, UTMs are copied to the custom fields x_utm_*.
3. Daily ETL: Odoo → BigQuery (or ClickHouse) → JOIN with GA4 events → attribution model on real revenue, not proxies.
4. A confirmed transaction (status paid on the Account Move) triggers the conversion event back to Meta CAPI and Google Enhanced Conversions — closing the loop.

If the ERP is not Odoo, the principle is the same: UTM ↔ Sales Order ↔ Payment confirmed, with feedback flowing back to the platforms. Without that chain, the model is built on sand.

When multitouch works and when it doesn't

Multitouch is not a fit for every business. Forcing it where it does not belong burns hours and demoralizes the team. The five patterns I see most often:

Works: e-commerce with more than 100 conversions per month

Beauty, retail, fashion, fintech onboarding, online education. Long funnel (3 to 7 visits to convert), several channels, enough conversions for DDA to be statistically meaningful. Here multitouch yields 15 to 30% efficiency upside without raising spend — purely through reallocation.

Works with caveats: B2B SaaS with a long cycle

If the average sales cycle exceeds 60 days, GA4 attribution windows (90-day max) cover the funnel, but DDA precision drops because of low conversion volume. Best approach: linear or time-decay combined with manual qualitative review of the journey in Salesforce, HubSpot, or Odoo.

Does not work: low-volume B2B (under 30 deals per month)

With fewer than 30 monthly closes, DDA falls back to rule-based — effectively last-click. Paying for Adobe Analytics or Mixpanel does not pencil out: the spend will not earn itself back. Stick to manual UTM tracking + Excel cohorts and focus on accumulating data.

Does not work without a fix: WhatsApp-heavy LATAM funnel

If more than 40% of deals close in WhatsApp (typical in Peru, Colombia, and Ecuador), the user leaves the website for chat and the attribution chain breaks. Fix: WhatsApp Business API + click-to-WhatsApp tracking + a custom tag in the CRM. Only with that link does WhatsApp re-enter the multitouch chain.

Does not work: offline-first business

Neighborhood restaurants, craft workshops, clinics with no digital funnel. Multitouch is overkill here. Use MMM (geo experiments + holdout) or simply A/B test channels week by week.

5 common LATAM PYME mistakes

#1. Confusing attribution with incrementality

Attribution answers “who gets the credit?” Incrementality answers “what would have happened without this channel?” These are different questions. A 5x DDA ROAS does not mean the channel drives incremental sales — it may simply be capturing users who would have purchased anyway.

Fix: holdout tests (geo-split or Meta Ads lift studies) and an MMM calibration at least once per quarter. Robyn and LightweightMMM already ship validation modules tied to lift experiments.

#2. Not configuring server-side tracking

Without Meta CAPI and Google Enhanced Conversions, in LATAM 2026 between 25 and 40% of conversions go missing: iOS + Safari ITP + Brave + ATT. The campaign optimizes on 60 to 75% of the real funnel — the rest of the budget drains.

Fix: GTM Server (Cloud Run or Stape.io, 15 to 30 dollars a month), Meta CAPI Gateway, and Enhanced Conversions with hashed email and phone. Setup: 30 to 60 hours for a specialist.

#3. Mixing models in reporting

Performance reads last-click in Meta Ads Manager. Leadership reads DDA in GA4. Finance reads UTM in the CRM. Three different truths, endless debates. When DDA is selected in GA4, Google Ads must also run DDA, or the figures will diverge 20 to 40%.

Fix: set the attribution model at the organization level, document it, sync it across platforms. The other models stay as a sanity check in exploration reports, never as KPIs.

#4. Ignoring LATAM payment delays

PIX, Yape, MercadoPago, Webpay — confirmation runs between 5 minutes and 48 hours. The default 1-day-click window closes the lead as “not converted.” If the campaign optimizes on conversions, it is learning the wrong signal.

Fix: for LATAM, minimum 7-day-click. For high-ticket (above 500 dollars), 28-day-click. Send the payment confirmed event via Conversions API, not the web pixel — the pixel can fire before the actual payment.

#5. Skipping MMM at media spend above 10 thousand dollars a month

Once a brand runs 8 or more channels and spends above 10 thousand dollars a month, user-level attribution physically cannot cover offline channels (TV, OOH, sponsorships, influencers without promo codes). MMM closes that gap. Robyn is free, and a data scientist can stand up a baseline in 40 hours. LATAM brands report 10 to 25% media-efficiency gains after the first MMM cycle.

Fix: if the budget allows it, hire a data scientist per project, not full-time. MMM gets refreshed once per quarter, not more often.

Case: 12 beauty brands, ROAS 1.5x → 4.2x in two quarters

Context. A group of 12 beauty brands (a multibrand portfolio, regionally comparable to a beauty conglomerate like The Estée Lauder Companies). Each brand ran its own campaign structure in Meta Ads, Google Ads, and TikTok Ads. Aggregated media spend around 180 thousand dollars a month. Performance team: 4 people for all 12 brands.

Problem. Aggregated ROAS at 1.5x — close to break-even after COGS and logistics. Meanwhile each platform reported a last-click ROAS between 3x and 5x. The discrepancy between the tools and the ERP revenue swung between 60 and 70%. Symptoms: brands that “should be profitable” showed negative gross margin in P&L marketing and finance argued once a week.

What was done across two quarters:

1. Tracking audit. Finding: 11 of the 12 brands ran without Meta CAPI, and 8 had no Enhanced Conversions in Google. Roughly 32% of iOS conversions were invisible.
2. Server-side stack. GTM Server on Cloud Run (45 dollars a month total for all 12 brands via a shared container), Meta CAPI Gateway, Enhanced Conversions for all 12 brands. Setup time: 3 weeks.
3. Unified attribution model. Migration of every account to DDA in GA4 and Google Ads. Meta: 7-day click + 1-day view. Team adoption took 4 weeks (debates, demos, and retraining on how to read the reports).
4. CRM truth pipeline. UTM ↔ Sales Order ↔ Payment confirmed in the ERP. Daily ETL into ClickHouse. JOIN of web events to real revenue. Every dashboard (Looker Studio) rewritten against the ERP source.
5. MMM baseline. Robyn on 18 months of historical data. Finding: Meta Awareness campaigns — written off as “no ROAS” — delivered an 18% incremental lift to retargeting and brand search. Seasonality calibrated (Mother's Day, LATAM Black Friday, Fiestas Patrias).
6. Budget reallocation. 25% cut on overlapping retargeting, 30% increase on upper-funnel video, 15% increase on influencer programmatic with UTM codes. ROAS targets rewritten from last-click to DDA-incremental with MMM correction.

Result after two quarters:

• Aggregated ROAS moved from 1.5x to 4.2x — with no increase in media spend.
• iOS conversion visibility rose from 60% to 92%.
• The platform-vs-ERP revenue gap closed from 65% to 8%.
• 3 of the 12 brands flipped from loss-making to profitable in performance channels.
• The weekly reporting meeting shrank from 6 hours to 90 minutes (single source of truth).

The case is not a magic pill. Of the six steps, four were infrastructure and organizational alignment — not secret models. Multitouch attributes correctly only when the data pipeline below is clean and the methodology is unified. Similar cases exist across LATAM beauty and retail: typical ROAS lift sits between 25 and 180%, depending on how broken the starting stack was.

What to do this week

If the diagnosis matches what you are seeing, three concrete moves to start with:

1. Tracking audit this week: check that Meta CAPI and Google Enhanced Conversions are active for every priority event. It is free and usually recovers 20 to 30% of “lost” conversions within the first week.
2. Sync the attribution model across Google Ads, Meta, GA4, and the CRM. No mixed reports: one model, one dashboard, one truth.
3. Wire the ERP / Odoo as source of truth. Platform ROAS is always a biased estimator. Real money lives in Sales Orders with confirmed payments.

For PYMEs already running Odoo, we offer an Odoo + marketing pipeline audit — a 30-minute call, stack review, written roadmap on output. If the operation scales across brands or countries, see the retail pricing intelligence pillar and the Odoo retail pricing pillar, where we cover multi-brand pricing and dynamic offers. For the beauty vertical specifically: Odoo for beauty and cosmetics. For cross-border campaigns, the country-by-country pillars: Odoo in Peru, Odoo in Mexico, Odoo in Colombia, Odoo in Argentina, Odoo in Chile.

Frequently asked questions