Build vs Buy: Identity Resolution on Databricks vs a Packaged CDP

Scott Brinker, dubbed the “Godfather of Martech” by Ad Age, has spent nearly two decades mapping the martech landscape. In his March 2026 research report The New Martech “Stack” for the AI Age, co-published with Databricks, he gives a name to what’s happening to enterprise marketing architecture right now: The Great Unstacking.

The premise is simple but consequential. For two decades, enterprise marketing technology was organized as a vertical stack: data warehouse at the bottom, systems of record in the middle (CRM, CDP), systems of engagement above that (MAP, DXP), specialist apps at the top, integrations bolting it all together. It was, as Brinker puts it, a “Tetris arrangement of boxes”, each box a self-contained bundle of data, logic, and interface.

That arrangement is buckling. AI agents need context from everywhere. Data platforms are pushing upward, arguing that since they already house the data, they’re the natural platform for software and AI operating on that data. And the integration tax — the constant effort of keeping all those boxes synchronized — has become, in Brinker’s words, “marketing’s most expensive invisible tax.”

What emerges from this pressure is what Brinker calls the composable canvas: a fluid architecture where data is the shared substrate, and capabilities assemble around it rather than sitting in rigid layers on top of it. The center of gravity in this new architecture is not the CDP. It’s the data platform.

“We’ve been trying to modify the same martech stack we’ve had since the internet started interneting. Folks, we’re going to have to build a new one.”
— Bryce Peake, former VP of Marketing Decision Sciences, Domino’s
— from The New Martech “Stack” for the AI Age

This shift has direct implications for identity resolution — the foundational problem of knowing who your customer actually is across all your data sources. Brinker’s report argues that CDPs, which emerged in the 2nd Age of Martech as integration hubs, are no longer the right center of gravity for this work. The data platform is. But before we explain the solution, let’s be precise about what’s broken.

The Real Problems with Packaged CDPs

1. They Are Not Fast to Deploy

The 60–90 day figure in CDP vendor marketing refers to getting a basic pipeline flowing, not a reliable identity graph you can trust for production decisions. Enterprise CDP implementations — Salesforce Data Cloud, Adobe Real-Time CDP, Twilio Segment, mParticle — consistently take 12 to 24 months from contract to production-quality identity resolution. Data modeling alone — mapping your source systems to the CDP’s opinionated schema — is a multi-month project. Add schema alignment, data quality remediation, connector configuration, matching model tuning, and validation: you’re well into year two before you have something you’d stake a campaign on.

2. They Are Not Cheap

Enterprise CDP contracts range from $300,000 to over $2 million annually before implementation costs. Systems integrators add $500,000–$1.5 million in year-one professional services. Brinker’s Martech for 2026 survey (Fall 2025) found integration remained a Top 3 challenge for the majority of martech respondents — despite CDPs being architected explicitly to solve it. The promise is simpler integration. The reality is integration complexity at higher cost, with a vendor in the middle.

3. Your Data Has to Leave Your Lakehouse

Every packaged CDP requires your customer data to be replicated to the vendor’s cloud involving extraction pipelines out of Databricks, synchronization overhead as schemas evolve, bi-directional sync of resolved identities back, and a new data residency surface in every compliance conversation. The CDP becomes a perpetual derivative of your real data, always fighting to stay synchronized with Delta Lake.

4. Matching Is a Black Box

CDP vendors protect their matching algorithms as proprietary. You accept their confidence thresholds and merging logic. When a match is wrong — and some percentage will always be wrong — you cannot audit the decision, retrain the model, or implement domain-specific rules. For KYC/AML, patient identity, and regulated B2B environments, this opacity is a compliance liability.

5. Per-Profile Pricing Penalizes Your Growth

CDP pricing scales directly with profiles, events, and destinations. As your customer base grows, your CDP bill grows proportionally, often non-linearly. Compute-based Databricks pricing does not scale the same way.

6. They Create Lock-In at Your Most Critical Layer

“We want to own the core construct of the data and infrastructure. If we change agencies, all we need to do is flip the activation layer at the top. The foundation stays with us.”
— Kumar Ram, VP/Global Head of Marketing Data Sciences, HP
— from The New Martech “Stack” for the AI Age

Your customer identity graph is your most strategic data asset. Once it lives in a CDP’s data model — once your merge history, match confidence scores, and identity linkages are in their infrastructure — switching means re-resolving all your identities, remapping all downstream systems, and losing historical linkages that often cannot be reconstructed.

7. They Were Built for B2C — B2B Identity Is a Different Problem

Traditional CDPs were architected for resolving individual consumers across digital channels. B2B identity resolution requires resolving accounts (companies), contacts within those accounts, and hierarchical relationships between subsidiaries, parent entities, and buying groups. A single enterprise account appears across CRM, ERP, ZoomInfo, Bombora intent data, and product usage — under different names, domain variations, and DUNS numbers. Most B2B organizations running CDPs maintain separate account matching processes outside the CDP because the CDP simply cannot handle it.

Zingg on Databricks: The Solution

Zingg on Databricks closes every one of these gaps. It is an open-source, ML-based identity resolution engine that runs natively on Databricks Spark, reads from Delta Lake, and writes a ZINGG_ID — a persistent, unified identity key — back to your Delta tables. No extraction. No vendor cloud. No black box. This is precisely the composable CDP architecture Brinker’s report describes: identity resolution operating directly on the data platform.

“The traditional layered structure can be better understood as a graph. It’s the foundation that consists of data and semantics, upon which a more fluid network of bi-directional communication between agents is built.”
— Tasso Argyros, VP Engineering, Databricks
— from The New Martech “Stack” for the AI Age

Your Entire Enterprise Data Estate, In One Place

Because Zingg runs natively on Databricks, it can access all enterprise data: finance records, product usage telemetry, supply chain data, HR org structures, ERP customer records — every table in your Delta Lake. CDPs ingest marketing data. Zingg on Databricks resolves identities across your entire enterprise data estate, producing a richer ZINGG_ID than any marketing-only CDP could create.

The Platform Provides Data Quality — Zingg Provides Identity

Databricks’ native infrastructure — Delta Live Tables expectations, Unity Catalog lineage, and your existing integrations with dbt, Monte Carlo, or Soda — ensures records are clean and governed before Zingg ever processes them. Zingg operates on that foundation, adding the ML-based probabilistic matching layer on top.

“Once all of our marketing data was centralized in the lakehouse, we revisited models that had never fully leveraged that richness. The rebuilt lead scoring model converted at 4X the previous rate.”
— Elizabeth Dobbs, AVP of Marketing Technology, Data & Growth, Databricks
— from The New Martech “Stack” for the AI Age

B2B Account Resolution, Natively

Zingg resolves B2B accounts natively — account hierarchies, subsidiary-to-parent matching, contact deduplication across CRM and MAP, buying group identification — powering ABM workflows in Adobe Marketo Engage, Salesforce Account Engagement, and 6sense.

The Production Architecture

Zingg reads from Delta Lake, runs its blocking and ML matching pipeline on Spark, and writes the ZINGG_ID back to your Delta tables — immediately joinable with every table in your lakehouse: behavioral events from Braze or Iterable, transaction history, product telemetry, and ML feature stores. Activation to Salesforce Marketing Cloud, HubSpot, Klaviyo, Google Ads, and Meta flows through Hightouch or Census.

Conclusion

The Great Unstacking is already underway. The organizations ahead of it recognized early that the data platform, not the CDP, is the center of gravity for customer intelligence in the AI age. Databricks gives you the lakehouse. Zingg gives you the identity resolution engine for B2C customers, B2B accounts, and every enterprise entity that matters. The ZINGG_ID lives in your Delta Lake, governed by Unity Catalog, enriched by your entire data estate, owned by your organization.

Resolve where your data lives. Own the result.

🔗 Zingg for Databricks  |  Talk to the Team  |  Deployment Guides

📄 The New Martech “Stack” for the AI Age — Scott Brinker & Databricks (March 2026)
📄 Martech for 2026 — Scott Brinker & Frans Riemersma (December 2025)