Telecom attribution in LATAM: MTA, MMM, and MTS + Philip Morris cases

Why telecom attribution in LATAM is its own discipline

In Lima, a postpaid subscriber crosses 7–11 touchpoints over 30–60 days before switching operators. In Mexico City, an IQOS user shows up in zero Meta pixels — the product is regulated out of digital ads entirely. If your marketing dashboard reports last-click ROAS, you are paying for conversions that would have closed anyway, and you are not paying for what actually moves gross adds.

LATAM telco operators and regulated-CPG manufacturers share one pain: classic attribution models (last-click in Google Ads, last-touch in Meta Ads Manager) break down on long-cycle products and in regulated mode. Per the GSMA Mobile Economy Latin America, the region has ~420 million unique mobile subscribers, average ARPU of USD 7–14, and in 8 out of 10 countries ≥ 30% of gross adds come through an offline channel. Performance teams cut TV and brand YouTube formats — and two quarters later watch gross adds drop 15–25%.

This pillar covers the why and the how: why MTA (multi-touch attribution) is not enough, how MMM (Marketing Mix Modeling) fits in, which geo-experiments work in LATAM, and how Odoo can act as a customer data hub without the USD 2M Adobe + Salesforce stack. With two case studies: MTS-profile (Eastern Europe, 60M+ subscribers) and Philip Morris-profile (regulated CPG in Mexico and Colombia).

One-minute summary

• Telecom attribution needs a hybrid stack: MTA + MMM + incrementality testing. Pure last-click destroys unit economics on long-cycle postpaid products.
• LATAM privacy enforcement is tightening from 2024: ANPD (Brazil) is fining under LGPD, INAI (Mexico) is being restructured into the Secretaría Anticorrupción, and Peru updated Ley 29.733. A cookies-only stack no longer works.
• Regulated CPG (tobacco, premium alcohol) is cut off from ≥ 70% of digital channels. Attribution gets built through geo-experiments and retailer sell-out data.
• Odoo as customer data hub gives telcos and CPG brands a unified "lead → activation → ARPU → churn" schema without a USD 500k CDP stack.
• MTS-profile: MMM on PyMC, +14% gross adds at the same budget, CAC −19%. PMI-profile: geo-experiments + sell-out attribution, +27% sell-out volume YoY.

The market: 4 holdings, 60–80% of sales offline

The LATAM mobile market has 4 dominant holdings: América Móvil (Claro), Telefónica (Movistar), Millicom (Tigo), and local operators (Entel, Personal, AT&T MX). Their digital marketing budgets run USD 50–200M per year in large countries. But 60–80% of the sales flow ends offline: dealer network, branded stores, partner retail, call center. This is not e-commerce, where attribution boils down to configuring GA4.

Issuing a postpaid invoice takes 7–30 days from first touchpoint. In that window the subscriber moves through:

• Search queries ("plan claro 5G ilimitado", "best postpaid plan peru")
• Comparison sites (rankia.pe, rebajatuscuentas.cl, comparayahorra.com)
• TikTok/Reels with UGC reviews (50%+ of young LATAM users treat TikTok as a search engine, Kantar IBOPE 2024)
• WhatsApp consultation with a dealer rep
• Call center call for checkout or portability

Each touchpoint claims the conversion. Last-click Google Ads gives 100% credit to brand search, first-touch Meta Pixel to Reels, GA4 data-driven attribution spreads it around — but without an offline signal, all of this is putting together a puzzle with three pieces out of ten.

Regulated CPG is even harsher. Philip Morris cannot advertise IQOS on Meta Ads — it violates Facebook Ads Policy 4.4 (Tobacco). Google Ads Restricted Products blocks all tobacco creatives, including heat-not-burn and nicotine pouches. TikTok Branded Content Policy shuts down creator content for tobacco. The real touchpoint network for PMI, BAT, and JTI:

• Point-of-sale in tobacconists and convenience stores (Oxxo, Tambo+, Walmart cigar lounges, D1)
• Age-gated direct site (iqos.com)
• B2B channels for retail partners and distributor incentives
• Personal selling in LAU (Limited Adult User) zones, premium experiential events

Marketing attribution for regulated CPG is built through geo-incrementality and sell-out data from retailers, not clicks. This is a completely different methodology. PMI IQOS case in Mexico City: USD 30M annual budget, zero dollars in performance digital ads — and still the team has to defend ROI to HQ in Lausanne. The default Marketing Cloud + GA4 dashboards are useless here.

What changes in LATAM 2024–2026

iOS 14.5+ ATT (since 2021): Meta and Google pixels lost 30–60% of conversion signal in LATAM. iOS market share is 28% in Chile, 20% in Mexico, 18% in Argentina (StatCounter LATAM, April 2026). Telcos felt the hit first — performance teams see ROAS dropping while real sales hold steady. That is a signal to rebuild measurement, not to cut budget.

Google Privacy Sandbox / cookie deprecation in Chrome: third-party cookies are fully deprecated as of Q1 2025 for 100% of users. Cross-site tracking via third-party cookies in LATAM is dead for practical purposes — Chrome holds 73–80% of regional browser share. Audience extension through Audience Network, AdSense, and programmatic display all lost match precision.

Privacy enforcement by country:

• Brazil: ANPD issued its first large LGPD fines (Lei 13.709/2018) in 2024–2025. Telcos and financial services landed in the crosshairs — fines up to 2% of revenue, capped at R$ 50M per violation.
• Mexico: LFPDPPP reform 2025 — the private sector moves from INAI to the Secretaría Anticorrupción y Buen Gobierno via constitutional reform transitorios. A 6–12 month enforcement vacuum is possible in 2025–2026, but sanctions issued before the transition still hold.
• Colombia: the SIC actively fines under Ley 1581 de 2012 — typical fine of COP 100–500M for mishandled consent.
• Peru: the Autoridad Nacional de Protección de Datos Personales (ANPDP, under Minjus) issued an updated regulamento in 2024, with fines up to 100 UIT (≈ S/ 535,000 in 2026).
• Chile: the new Ley de Datos Personales (Boletín 11.144-07) was approved by Congress and takes effect in 2026, harmonizing the regime with GDPR.

Regulated advertising across LATAM:

• Tobacco: the WHO Framework Convention (FCTC) is ratified by every LATAM country except Argentina. Full cross-media ad bans: Colombia (Ley 1335/2009, Art. 16), Mexico (Ley General para el Control del Tabaco 2008, Art. 23), Peru (Ley 28705/2006), Chile (Ley 19.419/1995 rev. 20.660/2013), Uruguay (Ley 18.256/2008 — the strictest plain packaging in the region).
• Alcohol: in Chile, Ley 19.925 governs hours and context for alcohol > 20%; in Colombia, location targeting is restricted (300 m from schools).
• Finance / crypto / gambling: Argentina's CNV tightened advertising rules for crypto businesses in 2024; Mexico's CNBV issued new fintech advertising rules in 2025.

For the marketing stack this means: first-party data only, server-side tracking, and MMM for regulated and long-cycle.

How to build a telecom attribution stack

#1. First-party data layer

• CMP (Consent Management Platform) with granular consent: Cookiebot, OneTrust, or open-source Klaro. ANPD- and LGPD-compliant configuration — separate consent per purpose (analytics, personalization, advertising), opt-in by default, audit trail.
• CDP or customer data hub: in Odoo this is the combination of crm.lead + res.partner + marketing.activity + a custom attribution.touchpoint model. It unifies web events (GA4), call-center logs (Asterisk/Aircall), retail POS (Odoo POS), and reseller channel (Odoo Sales Partner Manager).
• Identity resolution: hashed email + phone (SHA-256). In LATAM, phone-as-identity beats email-as-identity — 90%+ coverage vs 50–60% for email. WhatsApp-driven consent flows boost match rate +25% vs email-only.

#2. Server-side tracking

• sGTM (Google Tag Manager Server-side) on Google Cloud Run or AWS Lambda — ~USD 5–15/mo on average traffic. Intercepts events on the server, enriches them with first-party data, ships to GA4, Meta CAPI, TikTok Events API.
• Meta Conversions API (CAPI): mandatory since 2022 in the post-ATT world. Without CAPI, Meta loses 40–60% of match rate in LATAM; integration runs through sGTM with payload deduplication.
• Google Enhanced Conversions: hashed user data in Google Ads — +5–15% conversion lift (documented on Think with Google).

#3. Hybrid MTA + MMM attribution

Multi-touch attribution (MTA) — for online-heavy products (prepaid recharges, MVNO activations, in-app upgrades). Tools:

• Data-driven attribution in GA4 (requires a minimum of 600 conversions per channel per 30 days)
• Markov chain attribution via R package ChannelAttribution or Python pychattr
• Shapley value attribution for top-of-funnel valuation

Marketing Mix Modeling (MMM) — for long-cycle, offline-heavy, regulated channels. Production-ready open-source tools:

• Google LightweightMMM — Bayesian MMM on JAX. Production-ready for a team of 1 data scientist + 1 marketing analyst. Supports carryover effects, saturation curves, geo-level modeling.
• Meta Robyn — R-based, automated hyperparameter tuning via Nevergrad. Good fit for small/mid telco without an in-house DS team.
• PyMC-Marketing — Bayesian Python, actively developed since 2024. The most flexible model — custom priors, hierarchical models for multi-country MMM.

Minimum MMM feed: 2 years of weekly data per marketing channel (spend + impressions) + outcome variable (gross adds, sell-out units, brand search volume) + control vars (price, distribution, weather, seasonality, holidays, FX rate).

#4. Incrementality testing

Both MTA and MMM need incrementality calibration. Without it, MMM shows correlation, not causation.

• Geo-experiments: split-test territories (e.g., 16 Chilean regiones × 2 groups). Run the campaign only in treatment — measure the lift in gross adds.
• Google GeoLift (open-source R package) — automated geo-experiment design + power analysis + post-hoc evaluation.
• Holdout cells: 5–10% of the audience gets no digital ad → measure the organic baseline. In LATAM, ATT-induced data loss makes this especially valuable.

#5. Odoo as the coordinating hub

On the classic stack (Salesforce Marketing Cloud + Adobe Experience Platform + Tealium + Tableau), a telco pays USD 500k–2M/year in licenses. Odoo replaces 70% of that:

• marketing.campaign — holds every campaign with budget allocation per channel
• crm.lead + custom UTM fields — lead entry point with full UTM context
• res.partner — unified customer view (web + retail + reseller + call center)
• attribution.touchpoint (custom model) — every touchpoint with timestamp, channel, value attribution
• BI layer: Odoo dashboards + ClickHouse underneath for the MMM-feed export. ClickHouse handles 10B+ events/month on a USD 200–500/month cluster.

When it works and when it does not

#1. Works: postpaid B2C with digital share ≥ 30%

If ≥ 30% of gross adds come through online (web checkout, in-app upgrade, e-commerce), the MTA + MMM hybrid gives 80–90% attribution accuracy. Profile: MVNO in Brazil (Vivo Easy, Surf Telecom) with 50% online activations — Robyn MMM shows ROI per channel with ±15% confidence interval.

#2. Partial fit: B2B enterprise telecom (Movistar Empresas, Claro Negocios)

Long sales cycle of 60–180 days, account-based, sales rep does 80% of the work. An MTA stack is useless. What works: account-based attribution through CRM + sales-touch logging + intent data (6sense, Bombora) + LinkedIn Insight Tag. Do not try to apply GA4 conversion modeling — write a custom Odoo crm.lead.scoring with source weights from historical close-rate.

#3. Partial fit: regulated CPG (tobacco IQOS, premium alcohol)

Digital ads are banned — Meta, Google, TikTok are closed. What works:

• Geo-incrementality on the retail partner network (Oxxo, 7-Eleven, Walmart, D1)
• Sell-out data from the distributor + matching against brand search lift (branded keywords ≠ tobacco advertising under Google Ads interpretation)
• Owned channels: brand website with age gate, CRM email/SMS to verified LAUs
• Out-of-home + experiential: measured through geo-experiments + sell-out delta

MMM for regulated CPG is different: control variables dominate (regulatory news, plain packaging, excise tax shifts), and treatment channels number 3–5 instead of 17.

#4. Does not work: pure offline retail telecom

If 95% of sales are walk-ins at a dealer point with no digital pre-shopping (rural Peru, northern Argentina, indigenous markets in Bolivia), attribution collapses to brand tracker + share of voice mapping. That is not attribution — it is brand measurement (Kantar Millward Brown, Ipsos). Do not build MMM on 100 weekly observations: it will overfit.

#5. Does not work: short-cycle prepaid with street-vendor recharges

In Colombia and Peru, 40–60% of prepaid recharges run through street vendors with no trackable digital touchpoint. Attribution for this cohort is impossible without RFID/GPS instrumentation of the points. Here, aggregate mix-modeling + brand health KPIs work, but channel-level ROI does not.

Typical mistakes in LATAM telecom attribution

#1. Last-click for long-cycle postpaid

70% of LATAM telcos in 2024–2025 (based on my audits) still use last-click as the primary ROI metric in Google Ads / Meta Ads Manager. On long-cycle (30–60 days), brand search captures 60–80% of credit, while top-of-funnel (TikTok, YouTube) pays without recognition. Performance cuts YouTube → next quarter gross adds drop. Death spiral.

#2. Ignoring the call center as a touchpoint

In Peru and Colombia, 30–50% of postpaid activations close in the call center (inbound and outbound). If the analyst does not log "call → activation" with a UTM source (Asterisk integration, Aircall webhooks, Genesys Cloud), MMM will credit the channel that sent the lead, not the one that closed the deal.

#3. Mixing brand and performance metrics in one model

A brand awareness YouTube campaign is measured by lift in unaided recall (Google Brand Lift Study). A performance campaign is measured by CAC. Cram both into one MMM without instrumental variables and the model becomes uninterpretable. Split MMM into a brand layer (6–18 month decay) + a performance layer (1–4 week decay).

#4. Failing to net out dealer commission distortion

In Mexico, Claro and Telcel pay tiered commissions to dealers per gross add. The dealer is motivated to claim organic traffic as their own — to take the full commission. If you trust the dealer CSV without cross-validation against network signal (IMSI registration timestamp, HLR data from IFT-mandated reports), the whole attribution lies by 15–30%.

#5. Privacy compliance as bolt-on, not by-design

The marketing team spends 6 months building an attribution stack, then the DPO arrives with an LGPD checklist — and 4 out of 12 integrations need to be redone without re-consent. Re-collecting consent across a 5M base is another 3 months at a 40% opt-in rate. Privacy review happens at solution design, not at deployment. In Brazil, ANPD already issued ≥ 30 fines in 2024–2025 for this pattern.

Case #1: MTS-profile — 60M+ subscribers, Python MMM stack

A large Eastern European operator (MTS / Megafon / Beeline scale profile, USD 7B+ revenue, 60M+ subscribers) built its marketing measurement stack this way.

Situation (2022): USD 80M/year marketing budget across 17 channels (TV, OOH, programmatic, search, social, influencer, sponsored content on Telegram, retail marketing). Attribution = Adobe Analytics + last-click Google Ads. The marketing team could not defend TV budget to the CFO — "no visible ROI."

What they did:

• Migrated all events to a ClickHouse warehouse (10B+ events/month)
• Built a Bayesian MMM in PyMC-Marketing on 156 weekly observations
• Geo-experiments: 12 regional clusters × A/B treatment with a YouTube campaign; lift +8.3% in treatment vs control (p < 0.05)
• Integrated MMM output into a CFO dashboard: ROI per channel with 90% confidence interval

Outcome: TV down 22%, programmatic display down 18%, +18% in YouTube/TikTok creator content, +12% in Google performance. After 8 months: gross adds +14% at the same budget, CAC down 19%. The CFO signed off on year-on-year marketing investment for the first time without cutting TV.

"Management compared MMM output against the actual lift from the geo-experiment. They matched within a 12% margin — so the model got signed off. If they had been more than 30% apart, the whole model would have gone in the trash."

Case #2: Philip Morris-profile — regulated CPG in LATAM (IQOS context)

A large international tobacco manufacturer (PMI / BAT / JTI profile, USD 80B+ global revenue) launched a heat-not-burn product in Mexico and Colombia.

Situation: digital advertising is fully banned by regulators (Ley General para el Control del Tabaco MX, Ley 1335 CO). USD 30M annual marketing budget had to be allocated across: trade marketing (B2B incentives to retailers), in-store sampling, brand experience events for LAUs, CRM to verified users. Effectiveness per channel — unknown.

What they did:

• Built a sell-out data pipeline from 8 retail partners (Oxxo, 7-Eleven, Walmart, D1, Justo & Bueno) — weekly SKU-level units in/out
• Geo-experiments: experiential events in 3 Mexican cities, hold-out in 3 cities → measured 6-week sell-out delta
• Brand search lift via Google Trends API + branded search volume in Google Ads (branded keywords are allowed)
• Odoo + ClickHouse layer for the unified view: trade incentives + sell-out + brand search + LAU CRM activity

Outcome: the team identified that events delivered 2.1× ROI vs trade marketing in the first 6 weeks, but trade marketing is critical for distribution availability (numeric distribution +18% after the Q1 push). Reallocation: +35% events budget, −20% trade promotions, −15% generic OOH. Sell-out volume +27% YoY on a match-store basis.

Download the checklist: telecom attribution stack readiness

13-point checklist for CMO / Head of Marketing at a telco or regulated CPG: privacy compliance (LGPD / LFPDPPP / Ley 29.733), data stack maturity, MMM readiness, MTA setup, geo-experiment infrastructure. Leave your email — we send a PDF + Excel template to self-audit your attribution stack.

FAQ