Skip to main content

CDP Capability: Own vs. Rent

This table separates strategic assets from infrastructure. Identity graph, survivorship logic, and consent state are strategic assets. We own those. Activation, orchestration, and UI tooling are infrastructure. We rent those and stay switchable.

The distinction matters because conflating the two leads to renting back our own crown jewels years from now. This table sets that line before vendor conversations begin.

CapabilityOwn/RentWhyBuild Options (if Own)Vendor Options (if Rent)
Identity graph & resolution logicOwnCompounding strategic asset; core IPBigQuery tables + dbt models for nodes/edgesMatching signal only: LiveRamp
Survivorship rules / golden record engineOwnEncodes business logic, not a product featuredbt models with versioned rule logic
Consent/preference stateOwnLegal liability + trust assetBigQuery table as state of recordCollection UI only: OneTrust
Event schema / semantic layerOwnDefines business vocabularydbt Semantic Layer (native, since already on dbt)
Historical event data / raw behavioral storeOwnPortability, replay, ML training dataBigQuery native storage
Real-time identity resolution serviceOwn (logic) / Rent (infra)Latency-sensitive but logic must stay yoursCustom service (Cloud Run/GKE) reading resolved graphInfra: Bigtable
Probabilistic/fuzzy matchingRentHard to replicate well in-houseSplink (OSS, Spark-based) if buildingLiveRamp
Streaming/event ingestion pipesRentCommodity plumbingPub/Sub + Dataflow (GCP-native, no vendor needed)
Reverse ETL / activation orchestrationRentCommoditized, low switching costNot recommended to buildHightouch
Segment/audience builder UIRentMarketing-facing interface, replaceableNot recommended to buildHightouch
Ad platform / ESP connectorsRentPure commodity integrationNot recommended to buildHightouch
Consent management UIRentInterface, not the state itselfNot recommended to buildOneTrust
Personalization/decisioning engineRent (evaluate case-by-case)Can commoditize; revisit if models become IPBigQuery ML if models become differentiated IPAdobe Target (only if already in Adobe stack; otherwise defer)
Fraud/risk identity signalsRent (signals) / Own (decisioning)Signal sourcing is commodityDecisioning logic: dbt/BigQuery MLSocure
Data quality monitoring/observabilityRentTooling, not strategicGreat Expectations (OSS) if buildingMonte Carlo
Composable orchestration layerRent (thin layer)Should sit over owned assets, not be oneNot recommended to buildHightouch

Glossary

CDP (Customer Data Platform) — Software that collects customer data from many sources and combines it into one profile per person.

Identity graph — A map that links all the different IDs (email, phone, device, account) that belong to the same customer.

Identity resolution — The process of figuring out which records belong to the same person and linking them together.

Deterministic matching — Linking records using exact matches, like the same email address or customer ID.

Probabilistic matching — Linking records using best guesses, like similar browsing patterns or device info, when there's no exact match.

Survivorship rules — Rules that decide which piece of data wins when two systems disagree (example: which address is correct).

Golden record — The single, trusted version of a customer's data after all sources have been combined and conflicts resolved.

Consent management — Tracking what each customer has agreed to, like whether they opted in to marketing emails.

Composable CDP — A CDP built from separate tools on top of your own data warehouse, instead of one all-in-one product.

Packaged CDP — An all-in-one CDP product from a single vendor (example: Segment, Adobe, mParticle).

Reverse ETL — Sending data from your warehouse out to other tools, like ad platforms or email systems.

CDC (Change Data Capture) — A way to track and copy changes from a source system as soon as they happen.

MTU (Monthly Tracked Users) — A pricing model where cost is based on how many unique users are tracked each month.

Data portability — The ability to take your data out of a vendor's system in a usable format if you switch tools.

System of record — The one place considered the official, trusted source for a piece of data.