A strong number to start with: according to industry research estimates for 2024–2025, up to 30–40% of marketing spend goes to “noise” due to fragmented customer profiles, duplicated identifiers and faulty attribution. I see this in projects every month: there are leads, CRM is full, yet ROI is shrinking. The real cause is most often not the creative or the auction bids, but the foundation: data are not collected into a single customer data platform and cannot be linked into end-to-end customer journeys.

Ready to argue with yourselves and your processes? What will deliver greater growth in 90 days: adding another channel, or properly unifying marketing data into a single CDP platform with real-time activation?

I’m convinced: the latter wins. My colleagues at BUSINESS SITE have repeatedly achieved double-digit LTV uplift and a drop in CAC precisely after putting the data in order.

In this article I’ll explain how a Customer Data Platform for marketing works, which architectures to choose (cloud, hybrid, on-prem), how to set up identity resolution, integrate CRM/ERP and adtech without losing attribution, calculate ROI and TCO, and launch the first use cases within 90 days. I recommend reading to the end — this is not theory but selected practices that have already been proven on e-commerce, retail, B2B and mobile products in Ukraine.

CDP: What it is and why marketing needs it

cdp chto eto i zachem marketingu h2 img 1 CDP platforms how to consolidate marketing data

Customer Data Platform (CDP) – это системный слой, который собирает first-party события из всех источников (web, мобильное приложение, POS, кол-центр, CRM), автоматически стыкует их в единый профиль клиента (unified customer profile, UCP) и активирует сегменты в каналах: email, push, SMS, web-персонализация, programmatic и DSP.

I consider a CDP to be the “brain” of personalization and omnichannel marketing.

Business benefits for entrepreneurs and managers are clear and measurable: real-time offer personalization increases conversion and ARPU, a unified customer data platform reduces CAC through precise audiences, and coordinated cross-channel communication raises LTV. In BUSINESS SITE e-commerce projects, clients saw a 15–25% uplift in repeat purchases after launching cart triggers and recommendations based on the UCP.

How does a CDP differ from a DMP and a CRM? A CRM manages relationships and deals, a DMP works with cookie-based segments for media, and a CDP, using identified first-party data, builds the golden record and orchestrates the customer journey.

When is CRM/MDM enough? If there are few communications, two to three channels, and personalization is static. When is a CDP needed? When you need omnichannel, mobile+web, POS, marketplaces (Rozetka, Prom.ua), integrations with ‘Nova Poshta’, payments (LiqPay/PrivatBank, Monobank, Fondy, WayForPay) and real-time capabilities.
Expectations for time to value (TTV): the first quick wins are possible in 4–8 weeks: abandoned cart, price-drop, back-in-stock, win-back. In our experience, full-scale orchestration with micro-segmentation and ML takes 3–6 months, which aligns with practices of global CDP vendors.

Marketing CDP features

The ingestion and streaming ETL functions include collecting web and mobile events via SDK and server-side tracking, integration of POS and call center, as well as streaming processing (Kafka, Kinesis, Pub/Sub). Such an event bus provides event reliability and low latency for real-time scenarios.

Then the CDP performs identity stitching and de-duplication: it links identifiers (email, phone number, device ID, cookie, user_id from CRM) into an identity graph, creating a unified customer profile. I recommend starting with deterministic matching (deterministic), and then carefully adding probabilistic (probabilistic) rules for cross-device.

Real-time segmentation and orchestration provide instant activation: reverse ETL into channels (email/SMS/push, adtech/DSP), webhooks and APIs for the website and app. This is critical for ‘warm’ windows: 30–60 seconds after a signal often determine the fate of a conversion. Support for ML and a feature store enables scaling personalization: micro-segmentation, propensity scores, next-best-action. We at BUSINESS SITE are piloting MLOps processes to update models reliably and predictably.

Security and access controls are the foundation of trust: RBAC, audit logging, encryption in transit/at rest, SSO/SAML/OAuth2, KMS keys and environment segmentation (dev/stage/prod). I recommend embedding privacy-by-design principles into configuration and pipelines.

How CDP unifies marketing data

kak cdp ob ediniaet dannye marketinga h2 img 3 CDP platforms how to consolidate marketing data

The pipeline looks like this: источник → ingestion → schema mapping → identity matching → unified customer profile → активация. Sources: website, mobile app, marketplaces and payments, POS, CRM/ERP, «Nova Poshta», call center. Ingestion normalizes events and metadata, adds timestamps and the source of truth (system of record).

An important choice: data schema: schema-on-write provides strict validity and faster queries, schema-on-read gives flexibility for analytics. In the customer data model (CDM) the following entities usually exist: Customer, Account (for B2B), Order, Product, Event, Consent. I recommend establishing data contracts ahead of time and versioning events.

Identity graph combines deterministic and probabilistic matching. Deterministic keys (email/phone with verification), the core, probabilistic rules add connectivity for cross-device and offline→online. Pseudonymization and tokenization of PII protect identifiers, and the golden record aggregates confirmed customer attributes.

Data quality is measured by metrics: completeness, accuracy, freshness. We set up monitoring dashboards: event frequency, share of “broken” payloads, average latency, share of identified users.

CDP architecture: cloud, hybrid, on-premises

arkhitektura cdp cloud hybrid on prem h2 img 4 CDP platforms how to consolidate marketing data
Cloud CDP provides scalability, high SLAs, rapid integration and lower TCO in the medium term. Hybrid is suitable if PII is stored locally, while anonymized features and activation are in the cloud. On-prem is chosen by companies with special data requirements, aware of the responsibility for high availability, DR and infrastructure costs.

CDP as a Service (CDPaaS) and an API-first architecture accelerate time to value: pre-built connectors, webhooks, CI/CD for configurations, schema versioning, test environments. Event-driven CDP provides low latency for personalization; batch-oriented: optimal for nightly loads and reporting. In practice, a hybrid is often appropriate: events in real-time, heavy transformations in batches.

Integrations with the data stack are standard: data lakehouse/warehouse (Snowflake, BigQuery, Redshift), MDM as the “master” of reference data. We build two-way synchronization: the CDP reads attributes from the DWH and sends back features via reverse ETL.

Attribution and CDP Integration Control

atributsiia i kontrol integratsii cdp h2 img 5 CDP platforms how to consolidate marketing data
Typical integrations include CRM (for example, Salesforce, HubSpot), ERP for orders and inventory, analytics platforms, communication channels (email, push, SMS), programmatic and DSP. It’s important to maintain consistent identity across systems: hashed identifiers, stable external_id, and aligned consent rules.

Reverse ETL activates segments in adtech/DSP, maintaining control over frequency, exclusion of existing customers, and lookalike. We use technical patterns: webhooks for triggers, SDK for the mobile app, server-side APIs for high reliability, and batch connectors for reference data and histories.

For offline→online schemes (POS, call center) we use loyalty codes, QR and phone number as the identification key, and then link the receipt to the UCP. In mobile products we augment SDK events with on-device models for edge personalization, which reduces latency and load.

We test integrations with schema contracts (data contracts), mock services, and SLA monitoring. Based on BUSINESS SITE’s experience, a separate «integration runbook» reduces incident resolution time and maintains the service level.

Privacy and consent: compliance with GDPR and CCPA

privacy consent sootvetstvie gdpr ccpa h2 img 6 CDP platforms how to consolidate marketing data
The principles of privacy-by-design mean that privacy is designed at the architecture level: least privilege access, data segmentation, encryption, access logging, and separate storage of PII. Technologies such as PETs, pseudonymization, hashing and PII tokenization reduce the risk of leaks.

Integration of the CDP with a CMP (consent management platform) ensures proper operation of consent flows: collection of consents, dynamic application to activations and auditing of actions. In our projects any segments in the CDP are checked for the presence of the corresponding consent before being sent to a channel.

Data retention and deletion policies, legal hold and auditability are documented and technically implemented via automated retention rules. GDPR, CCPA/CPRA and ePrivacy requirements affect the choice of data location, access processes and reporting. The roles of data governance and data stewardship assign responsibility: marketing — for business logic, IT — for the platform, legal — for compliance.

CDP implementation: stages and migration risks

Working implementation template: preparation → pilot → scaling → optimization. During preparation we clarify goals (KPI, ROMI, LTV uplift), data sources, the CDM schema, and agree on data contracts. We run the pilot on 1–2 priority use cases (for example, cart-related and win-back), then scale channels and segments.

Migration strategies: phased (sequential by source), parallel (old and new systems running in tandem) and big-bang (rarely justified). I more often choose phased+parallel for critical systems: lower risk and easier to control data quality.

Risk management includes rollback plans, A/B validation of segments, monitoring of completeness/accuracy/freshness. Change management is a cross-functional team (marketing+IT+legal), assigned data stewards, and user training. To minimize TTV, we identify quick wins for the first 90 days and document them in the roadmap.

How to Choose a CDP for Large Enterprises

In the RFP for selecting a CDP we include requirements: functionality: ingestion/identity/activation, integrations (CRM/ERP/adtech/DSP), SLA and SSO/SAML/OAuth2, security (KMS, RBAC, audit), compliance (GDPR/CCPA), latency/throughput targets, data portability, 3-year TCO. It’s important to ensure API-first, reverse ETL, and real-time support.

We assess the risk of vendor lock-in by indicators: closed schemas, no raw data export to the DWH, reliance on a proprietary identifier, exit penalties. Open event standards, a dedicated data lakehouse, a portability contract and migration pilots reduce the risk.

Integrations and vendors: Segment, Tealium, mParticle, Treasure Data, RudderStack: each is appropriate in its own circumstances. Based on BUSINESS SITE experience, Segment and mParticle are strong in mobile scenarios, Tealium — in tag management and enterprise compliance, Treasure Data for data-heavy cases, RudderStack for API-oriented teams.

Prepare a list of questions for the technical interview: latency under load X, pricing model (events vs profiles), SLA/DR, scalability, support for schema evolution, reverse ETL limitations, feature roadmap, experience in Ukraine (marketplaces, “Nova Poshta”, local PSPs).

ROI, TCO and KPIs after CDP implementation

Metrics for C-level: LTV, CAC, ARPU, ROMI, incremental revenue/lift, payback period and NPV. I recommend calculating ROI at the use-case level: for example, abandoned cart – conversion uplift from X% to Y%, average order value Z, margin M, frequency F, cost per 1000 events and channel send costs.

Methodology: establish a baseline over 4–8 weeks, run randomized segments (control/target), measure incrementality and calculate revenue uplift minus incremental costs. For cross-channel we use multi-touch attribution and complement it with incrementality testing to avoid overvaluing the “last click”.

TCO expenses: license/events, integration and customization, support and SRE, infrastructure (if hybrid/on-prem), training and change management, as well as opportunity cost. KPI roadmap: by 3 months – share of identified users and initial retargeting uplifts; by 6 – reduced CAC and increased ARPU; by 12 – stable ROMI and LTV uplift by cohorts.

Contents of the TCO model and 3-year budget template
Include CAPEX (implementation, integrations, migration, training) and OPEX (license, channels, support, infrastructure, A/B platform). Model event/profile growth of 20–50% per year and the impact on price. TCO optimization is achieved by prioritizing use-cases with fast payback, eliminating duplicate systems in the stack, and reducing latency to the threshold that actually affects conversion.

throughput, SLA, latency and security

Set target SLO/SLA: for real-time activations, end-to-end latency of 1–3 s, availability 99.9%+, throughput: ability to handle peak N events/s without degradation. For batch, predictable windows and load deadlines.

Infrastructure requirements: horizontal scaling, DR with recovery time RTO/RPO, regular backup and continuous monitoring. Security: encryption at rest/in transit, KMS, SSO/SAML/OAuth2, RBAC, audit logging and network segmentation. In BUSINESS SITE projects we automate delivery of CDP configurations via CI/CD and add schema tests and pipeline regressions.

Observability metrics: percentage of delivered events, mean/95th percentile latency, connector errors, segment build speed, percentage of activatable profiles, data quality by completeness/accuracy/freshness. This is not just control – it’s daily risk management.

CDP use cases for e-commerce and B2B

E-commerce: real-time offer personalization on product cards and in the cart, remarketing taking into account marketplaces (Rozetka, Prom.ua) and delivery statuses of “Nova Poshta”. For one online store, after implementing CDP and back-in-stock scenarios we achieved a 12% increase in repeat purchases and a 9% LTV uplift in a quarter; lesson: the freshness of availability and price data matters.

Retail: unifying POS and online via phone/loyalty card, offline→online identification and omnichannel attribution. In the pharma segment we activated cross-promo based on receipts and doctor recommendations, while paying attention to consent and medical ethics; key takeaway – clear separation of attributes and access reduces risks and speeds up approvals.

B2B: account-based approach, where the CDP builds profiles of companies and contacts, linking the website, webinars and CRM. In the construction sector we reduced CAC by 18% thanks to micro-segmentation and prioritization of leads for sales enablement; lesson – aligning the attribution model between marketing and sales increases trust in the metrics.

Mobile scenarios: SDK event collection, on-device models for churn predictions, personalization of push and in-app banners. In a travel product, win-back segments based on a drop in order frequency delivered +14% to ARPU over 60 days; emphasis: correct frequency and creatives by segment.

Quick pilots: churn prediction, win-back, lookalike audiences, A/B and incrementality tests. The BUSINESS SITE practice confirms: small, clearly measurable experiments build executive trust and form a roadmap for scaling.

Optimize the marketing stack with a CDP

The strategy for using first-party data through a CDP begins with prioritizing sources: website and app, CRM, payments (PrivatBank/Monobank), delivery (“Nova Poshta”), marketplaces. Next we build martech/adtech integrations and determine which systems remain “leading” and which we switch to data consumer mode via reverse ETL.

The role of MDM vs CDP: MDM: the master of reference data and golden records for products/counterparties, CDP – behavioral events and activations. A coordinated architecture removes duplication and reduces TCO. Orchestration and journey orchestration in the CDP enable building omnichannel campaigns, while customer journey analytics and attribution dashboards provide a clear picture of ROMI.

Linking the CDP to DSPs and programmatic without losing control is achieved through hash identifiers, frequency caps, and two-way exchange of impression/click/conversion statuses.

Optimization methods: a feature store for reusing features, personalization of offers using propensity models, and automation of triggers taking consent and recency into account.

The first 90 days after launching the CDP

  • Clarify goals and core KPIs, record the baseline.
  • Set up connectors for web/mobile, CRM, payments and delivery; describe data contracts.
  • Enable governance: roles (data stewardship), access (RBAC), consent and retention rules.

Month 2:

  • Launch real-time segmentation for cart, price-drop, win-back.
  • Integrate email/push/SMS and one programmatic channel via reverse ETL.
  • Enable first ML features: churn score, probability of category purchase.

Month 3:

  • Run A/B and incrementality testing, update attribution (multi-touch).
  • Optimize latency and throughput, expand segments and channels.
  • Prepare a C-level report on ROI/TCO and a scaling plan for 6–12 months.
Control checkpoints: share of identified profiles, data freshness in the UCP, GDPR/CCPA compliance, security (SSO/SAML/OAuth2, audit logs), latency SLO.

Checklist quick wins: abandoned cart, back-in-stock, repeat purchase, win-back “90 days without an order”.

FAQ: questions from management and marketers

  • How to quickly prove ROI of CDP implementation to leadership? Run 2–3 pilot use-cases with control groups: abandoned cart, win-back, price-drop. Record the baseline, apply incrementality testing and present revenue uplift minus incremental costs. Payback is often achieved within 2–4 months.
  • Which CDP architecture is better for scalable company growth? In most cases cloud or CDPaaS with an API-first and event-driven core; hybrid — if local storage of PII is required. Criteria: latency/throughput, SLA, integrations, data portability, TCO over 3 years.
  • How to assess vendor lock-in risk when choosing a CDP provider? Check raw data export, independence of the UCP from proprietary keys, support for open-schema and reverse ETL, and the exit terms in the contract. Request a POC to migrate part of the data to your DWH.
  • Which KPIs and metrics to use over 3–12 months? 3 months: identification, open/click/conversion by triggers, initial uplifts. 6 months: reduction in CAC, growth in ARPU and ROMI, share of omnichannel customers. 12 months: LTV uplift by cohorts, NPV and payback, SLA stability.
  • How to organize governance and allocate responsibility for data? Appoint data owners by domain, data stewards in marketing and IT, approve data contracts and schema change processes, involve legal for consent and retention, and document a RACI matrix.

Conclusion: key recommendations and CTA

CDP is not another tool, but a strategic layer that connects channels, events and profiles into a single decision-making system. When a unified customer data platform and orchestration operate in real time, marketing stops guessing and begins to grow measurably: lower CAC, higher LTV and ROMI, transparent attribution and manageable TCO.

Top 5 practical steps to get started:
  1. Prioritize sources and use-cases with quick payback: cart, win-back, back-in-stock.
  2. Document the customer data model and data contracts, set up server-side tracking.
  3. Launch event-driven activation in 2–3 channels and connect reverse ETL.
  4. Enable governance and security: RBAC, SSO/SAML/OAuth2, KMS, CMP integration.
  5. Prepare an RFP checklist with requirements for latency/SLA, integrations, data portability and TCO over 3 years.

If it would be useful to receive a structured starter pack, I have prepared for blog readers: an RFP template for selecting a CDP and a 90-day implementation checklist with KPIs and checkpoints. The BUSINESS SITE team regularly supports such launches, from architecture and integrations to incrementality testing and attribution dashboards. In our experience, a sound strategy for using first-party data through a CDP turns channel chaos into a controllable growth machine.