Telkomsel has the data, the subscribers, and the ad platform. Metatech builds the advertising intelligence layer — the technology that turns these assets into $20-50M/yr in new revenue.
Three revenue pillars where custom engineering creates the moat. Metatech delivers the technology layer that transforms Telkomsel's data assets into advertiser-ready products.
Help AppsFlyer, Adjust & Branch recover lost attribution via CGNAT resolution
MMPs lose 65-80% of iOS attribution post-ATT. Telkomsel's CGNAT logs can resolve shared IPs to specific subscribers. Metatech builds the secure query API, clean room schema design, and partner integration layer — enabling Telkomsel to monetize this data asset immediately.
Push MSISDN-to-Device-ID match rate from current levels to 75-85%
The core enabler for all advertising. MyTelkomsel login matching is already live. Metatech builds: CGNAT correlation engine, multi-source graph fusion (Cassandra/Neo4j), confidence scoring with decay, real-time resolution API (<50ms for RTB), and clean room integration with super-app partners.
Prove that TADEX ad exposure drives real physical store visits
Telkomsel owns both sides (TADEX for ad exposure + network data for store visits), but has no platform to connect them. Needs: cell-tower-to-location processing engine, POI geofencing system, TADEX↔location attribution engine, ghost-ad incrementality measurement, and an advertiser analytics dashboard. This is a greenfield engineering project.
Combined revenue potential across three technology pillars Metatech builds for Telkomsel
For each opportunity: if Telkomsel already has it, they don't need us. If the data doesn't exist, we can't build it. Our value is the intelligence layer that turns their raw assets into ad products.
⚡ Metatech's approach: Build the application intelligence layer on top of existing infrastructure. We complement — not compete with — current data engineering vendors by focusing on the ad-tech product layer.
⚡ Metatech delivers: Secure CGNAT query API, AWS Clean Rooms schema design, per-MMP integration adapters, and real-time attribution enrichment. Combined with Telkomsel's DigiAds partnerships, this becomes the fastest path to revenue — production-ready in 3-4 months.
⚠ Note: The highest-value matching method (MyTelkomsel login) is already live without us. Metatech's value is adding the remaining 6 methods and fusing them into a unified graph. If Telkomsel is satisfied with login-only matching (~50M users), they may not need the full graph build.
✓ This is where Metatech adds the most value. Telkomsel has the raw data on both sides (ad impressions + location) but no platform connecting them. There is no existing vendor doing this. It's a full greenfield build — and the resulting product doesn't exist anywhere in Indonesia.
| Pillar | Metatech Builds | Timeline | Revenue Impact |
|---|---|---|---|
| MMP Attribution | CGNAT query API, clean room schemas, per-MMP integration adapters | 3-4 months | $5-15M/yr |
| Identity Graph | CGNAT correlation engine, graph fusion (Cassandra/Neo4j), confidence scoring, real-time resolution API | 12-18 months | $10-20M/yr |
| Foot Traffic | Geofencing engine, attribution pipeline, visit inference model, ghost-ad incrementality, advertiser dashboard | 6-9 months | $10-30M/yr |
| Data Clean Room | AWS Clean Rooms query logic, partner onboarding tooling, cross-region orchestration, PDP compliance layer | 4-6 months | Enabler |
Metatech's value proposition: "We build the advertising intelligence layer — the algorithms, attribution logic, and advertiser products — on top of Telkomsel's data assets and cloud infrastructure. We don't replace data engineering vendors or cloud providers. We build the thing that turns data into ad revenue."
A data clean room is a secure computation environment where two parties can match and analyze their data without either side seeing the other's raw data. This is the foundation for all three opportunities.
Partner uploads their data — encrypted and hashed before entry. They never see Telkomsel subscriber data.
Encrypted matching environment — neither party can extract the other's raw data
Telkomsel uploads network data — subscriber IDs are pre-hashed. Raw MSISDN never leaves Telkomsel.
Campaign X drove Y installs • Z store visits • W% lift vs control
No personally identifiable information is exposed to either party
The core use case: An MMP like AppsFlyer records an ad click and an app install from the same public IP, but can't confirm it's the same user because of CGNAT. Telkomsel resolves this.
User clicks ad
MMP records IP + port + timestamp
Join on IP:port:ts
CGNAT resolves to subscriber
Same subscriber opens app
MMP sends install event
Same subscriber_token
= confirmed attribution
| IP matching accuracy | 30-60% |
| iOS post-ATT attribution | 20-35% |
| CGNAT resolution | Impossible |
| Cross-device linking | Probabilistic |
| IP matching accuracy | 90%+ |
| Telkomsel subscriber coverage | ~50% of Indonesia |
| CGNAT resolution | Deterministic |
| Cross-device linking | Subscriber graph |
| Gap | Problem | Telkomsel Solution | Impact |
|---|---|---|---|
| IP Resolution | CGNAT shares IPs across users | CGNAT log resolves IP:port:ts to subscriber | Highest |
| Cross-Device | No shared ad ID across devices | IMEI-to-MSISDN device graph | High |
| Session Continuity | Browser→store→app chain breaks | DNS/DPI tracks full journey | Medium |
| Offline-to-Online | No digital click from billboards | Cell tower proximity + install correlation | Medium |
| Web-to-App | 3P cookies dying | Subscriber ID persists regardless of cookies | High |
For the Identity Graph: Telkomsel matches its MSISDN subscriber data with partners who have both phone numbers (via login) and device advertising IDs. Clean rooms enable this without exposing raw data.
E.g. Gojek knows: phone number + GAID + user behavior. They hash phone before upload.
Match on hashed phone number
Telkomsel knows: MSISDN + IMEI + network behavior. They hash MSISDN before upload.
subscriber_token ↔ advertising_id mapping at 0.90 confidence • Both parties benefit from enriched targeting • No raw PII exchanged
Two leading platforms validated for Indonesia. Recommendation: start with AWS Clean Rooms (Telkomsel is already a preferred AWS customer), add Snowflake for partners on that ecosystem.
Building the full advertising data platform requires strategic partnerships across four categories. The value exchange must be compelling for each partner type.
Where Metatech plugs in as the engineering partner. Custom builds that create the moat — not off-the-shelf integrations.
CGNAT correlation engine, multi-source graph fusion (Cassandra/Neo4j), confidence scoring with decay. Real-time resolution API (<50ms for RTB). Deduplication and conflict resolution across 7 matching methods.
AWS Clean Rooms schema design, query API, and partner onboarding tooling. Standardized data contracts for super-app partners (TikTok, GoTo, Shopee). Cross-region compute orchestration.
Segment builder with telco-enriched demographics. Lookalike modeling from seed audiences. Real-time activation API for TADEX campaign targeting. Privacy-safe audience export for clean rooms.
MMP query adapter for AppsFlyer/Adjust/Branch integration. OpenRTB bid enrichment with identity signals. UID2 token service for open-web interoperability. CPID-equivalent carrier-persistent token generation.
PDP-compliant consent management integrated into MyTelkomsel. Granular opt-in/opt-out per partner and use case. Differential privacy for aggregate outputs. Full audit logging for regulatory compliance.
Advertiser-facing reporting with attribution visualization and ROI measurement. Identity match rate monitoring. Partner data quality scoring. Real-time campaign performance with identity-powered insights.
🛠 Estimated build: 8-12 engineers, 12-18 months for full identity platform. POC with MyTelkomsel login + 1 clean room partner in 3-4 months.
Seven methods to match MSISDN to device advertising IDs. MyTelkomsel login-based matching is already live — the highest-impact, most privacy-compliant method. Here's the full stack.
Each match source has a different confidence level and decay rate. The identity graph uses weighted composite scoring to determine the most reliable current mapping.
| Source | Confidence | Score | Decay Rate |
|---|---|---|---|
| MyTelkomsel Login | 0.99 | 30-day half-life | |
| SDK + SIM + GAID | 0.95 | 14-day half-life | |
| Partner Clean Room | 0.90 | 14-day half-life | |
| CGNAT with Port | 0.85 | 1-day half-life | |
| Wi-Fi Hotspot | 0.80 | 1-day half-life | |
| CGNAT without Port | 0.60 | Hours | |
| Device Fingerprint | 0.50 | 7-day half-life |
| Timeline | Match Rate | How |
|---|---|---|
| 12 Months | 65-75% | MyTelkomsel optimization (already live) + first clean room partnerships |
| 24 Months | 75-85% | + CGNAT matching + SDK partners + CPID-equivalent + UID2 integration |
| 36+ Months | 80-85% | Post-device-ID transition: flexible graph with GAID + UID2 + carrier-persistent tokens (Google Privacy Sandbox retired Oct 2025) |
Can we prove someone saw our ad and then walked into a store? Yes — and Telkomsel can do it better than anyone because we own both the ad platform and the location data.
Today, brands spend billions on digital ads. But for businesses with physical stores (retail, restaurants, automotive, banking), the biggest question is: "Did my online ad actually drive someone to walk into my store?"
Foot traffic attribution answers that question. It connects ad impressions (who saw the ad) with store visits (who physically showed up) — giving advertisers proof that their money is working.
Because it turns digital ads from a "we think it works" into "we can prove it works." Brands will pay 2-5x higher CPMs for campaigns that include foot traffic measurement — because they finally get accountability for offline results.
"Device ABC saw the ad"
"Device XYZ visited the store"
Different IDs. Trying to match them loses 40-70% of data. Plus, only people with the app installed are tracked.
Result: Most of the data is lost. Tiny measurable audience.
"Subscriber 123 saw the ad"
"Subscriber 123 was near the store"
Result: 100% ID match. No identity resolution loss. Accuracy depends on location data source (see tiers below).
The real advantage is zero ID loss when matching. But location accuracy varies by data source — be honest about what each tier can do.
GPS from MyTelkomsel app users with location permission. Accuracy: 3-10 meters.
Can confirm: "Subscriber entered this specific store."
Coverage: ~30-50M app users (est.)
Comparable to what Google/Meta offer — but with telco demographics on top.
4G/5G E-CID + Wi-Fi hotspot auth. Accuracy: 15-150 meters.
Can confirm: "Subscriber was at this mall or shopping complex."
Coverage: Most urban subscribers with active data sessions
Good enough for mall tenants, F&B chains, big-box retail.
Cell-ID + Timing Advance for all subscribers. Accuracy: 50-500 meters.
Can confirm: "Subscriber was in this neighborhood / district."
Coverage: All 170M+ subscribers, no app needed
Good for brand lift studies and catchment area analysis, not individual store visits.
Same subscriber ID for ad exposure and location. Zero data loss on the matching step — the part where competitors lose 40-70%.
SIM registration gives real age, gender, home location. Better control groups than any app-based panel.
30-50M at store-level (GPS). 100M+ at mall-level (network). 170M+ at zone-level. No one else in Indonesia has this range.
Combine GPS users (high accuracy) with network data (high coverage) for statistically modeled attribution across all tiers.
Foot traffic attribution is a proven, revenue-generating product at the world's biggest ad platforms.
Google uses phone location history to tell advertisers: "Your Search/YouTube ad drove X thousand store visits this month." Available to any Google Ads customer. The #1 reason retail brands increase Google ad spend.
Facebook/Instagram offers a dedicated "Store Traffic" campaign objective. Uses phone GPS from the app to measure who visited after seeing an ad. Used by McDonald's, Starbucks, automotive brands.
Closest telco model. SKT's T-ad platform uses subscriber data for targeted advertising with location signals. Korea's dense cell network (~52M pop) gives better cell-tower accuracy than most markets. Telkomsel has 3x the subscribers.
The pioneer of location attribution. Panel of 25M opted-in devices. Sells measurement to Coca-Cola, Samsung, BMW. Limitation: small panel size means sampling bias.
Grab launched ads in 2022, using ride/delivery location data to attribute store visits. Growing fast in SEA. Limitation: only covers Grab app users.
Telco data monetization. Singtel's data subsidiary sells mobility analytics (foot traffic patterns, catchment analysis) to retailers and urban planners. Uses network data from 4M subscribers. More analytics than ad attribution.
Telkomsel's real edge:
Google and Meta have GPS accuracy but only for their app users with Location Services on — and they can't link to telco demographics. Foursquare and Grab are limited to their own small panels. Telkomsel's advantage is combining zero-loss ID matching + SIM-verified demographics + tiered location data from store-level (GPS) to zone-level (network) across 170M+ subscribers. Nobody else in Indonesia can offer this stack.
What Telkomsel already has vs. what needs to be built:
Already exists: TADEX ad serving, DigiAds DMP with subscriber segments, network location data, MyTelkomsel GPS, SMS location-based campaigns.
Does NOT exist yet (our build): Attribution pipeline connecting TADEX impressions → location events. Geofence database of advertiser store locations. Stream processing for real-time matching. Ghost ad methodology. Incrementality reporting. Advertiser dashboard showing "your ad drove X store visits."
This is where Metatech builds the engineering moat. The entire "Does NOT exist yet" list above is custom software — not off-the-shelf. Here's what Metatech delivers:
H3/S2 spatial indexing for 100K+ store locations. Real-time point-in-polygon matching against cell tower, Wi-Fi, and GPS signals. Sub-50ms geofence resolution at 200K events/sec.
Kafka + Apache Flink stream processing connecting TADEX ad impressions to location visit events. Configurable attribution windows (1-30 days). Deduplication and dwell-time filtering (≥5 min threshold).
ML model that fuses multi-tier location signals (GPS + cell tower + Wi-Fi) into probabilistic visit scores. Handles accuracy gaps between Tier 1 (store-level) and Tier 3 (zone-level) gracefully.
PSA (public service ad) holdout methodology integrated with TADEX ad server. Statistical matching for control groups using telco demographics. Automated lift calculation and significance testing.
Self-serve reporting UI for campaign managers. Metrics: attributed visits, visit rate, incremental lift, CPIV, dwell time. Real-time and batch reporting. Store-level heatmaps and catchment analysis.
ClickHouse / TimescaleDB storage for 850M-3.4B daily location events. Optimized for time-series queries and spatial aggregation. Data retention and PDP-compliant anonymization pipeline.
🛠 Estimated build: 6-8 engineers, 6-9 months to production POC with 3 pilot advertisers.
For store visit attribution, <50m accuracy is ideal. Hybrid approach required: GPS for opted-in app users, Wi-Fi for hotspot areas, cell tower for universal coverage.
| Source | Accuracy | Coverage | Store Attribution |
|---|---|---|---|
| GPS (MyTelkomsel app) | 3-10m | App users with permission | Excellent |
| Wi-Fi Hotspot Auth | 15-50m | Hotspot users only | Good |
| 4G/5G E-CID | 50-150m | Active data sessions | Mall-level |
| Cell-ID + TA | 50-200m urban | 100% of subscribers | Area-level |
| Signaling (Diameter) | 100-500m | All LTE subscribers | Zone only |
The gold standard for proving ad causation (not just correlation). This is what Google, Meta, and Foursquare use. Telkomsel can implement natively because it controls TADEX ad serving.
Subscribers who saw the advertiser's ad on TADEX.
Measure: visit rate to advertiser's stores
Statistically matched subscribers who were eligible but shown a filler ad instead.
Measure: visit rate to same stores (baseline)
The visits caused by the ad, not just correlated. This is the metric advertisers trust and will pay premium for.
Now 70M+ users across 6 markets, 30+ telco partners. Consent-based network identity tokens. Expanding to CTV and Americas. The strongest telco ad-tech reference globally.
Network-level rotating pseudonymous ID powering Yahoo ad platform. Carrier data + programmatic. Sold to Apollo 2021. Legacy but proven the model works.
450M+ subscribers powering India's largest telco ad platform. Deterministic mobile identity at massive scale. Most comparable to Telkomsel's 170M base. Watch closely.
AI-powered ad platform using telco subscriber data. Pivoted to AI infrastructure (data centers, GPU). Ad tech deprioritized. Lesson: dedicated engineering partner (like Metatech) keeps execution focused while telco focuses on core business.
Sold Amobee for $239M in 2022. Tried to build everything in-house. Key takeaway: telco ad-tech succeeds with a focused technology partner model — exactly the approach Metatech proposes for Telkomsel.
Launching T-Advertising platform leveraging subscriber data across European markets. Part of Utiq consortium. Dedicated ad-tech unit with clean room infrastructure.
Indonesia's PDP Law (UU PDP No. 27/2022) — full enforcement since October 2024. Modeled on EU GDPR. Penalties up to 2% of annual revenue.
Existing consent covers telecom services. Does NOT automatically cover advertising use of subscriber data. Must be clearly separated.
NEW required layer via MyTelkomsel app. Clear explanation of what data is used, for what purpose, with whom it's shared. Granular opt-in/opt-out.
Disclose specific partners and purposes before any data enters clean rooms. Users can opt out of individual partnerships. Right to erasure enforced.
Once built, these three capabilities create a defensible position that no other player in Indonesia can replicate.
No ad tech company has network-level data for this scale of users. Structural monopoly.
No app developer has passive coverage without requiring installs. Even feature phones visible.
No MMP can resolve CGNAT to subscriber without carrier cooperation. Telkomsel is the gatekeeper.
No location company has passive data for ALL mobile users. Always-on, no SDK needed.
No competitor can match ad exposure, location data, and identity resolution in one system.
| Team | Size | Focus |
|---|---|---|
| Identity Graph | 10-15 | Data eng, ML, backend |
| MMP Integration | 5-8 | API, clean room, ops |
| Foot Traffic | 8-15 | Data eng, ML, geospatial, frontend |
| Legal/Compliance | 2-3 | PDP specialists |
| Total | 25-40 | Dedicated engineers + legal + partnerships |
| Revenue Stream | Model |
|---|---|
| MMP Attribution | Per-query fee: $0.001-$0.01/match 100M queries/mo × $0.005 = $6M/yr |
| Foot Traffic | CPV model: Rp 5,000-25,000/visit Tiered location intelligence suite |
| Identity Graph | TADEX premium pricing uplift Better targeting = higher CPMs |