Stop Buying CDPs. Build the Decision Layer for Humans & Agents.
Warehouse-first, composable decision intelligence with a multi-graph knowledge layer and agentic orchestration turning customer data into compounding outcomes.
When I wrote about the death of CDPs, it stirred up a reaction. Let me clarify: “CDP as a vendor category” the VC-inflated, boxed product may be on its way out. “CDP as a framework” for unifying, analyzing, and activating customer data is more relevant than ever.
The category got crowded. Features converged. Some vendors faced pressure or exits; others pivoted or re-labeled to chase budgets from CIOs instead of CMOs. Meanwhile, enterprises are doubling down on what I call Internal CDP 3.0: a warehouse-first, composable, Agentic AI ready that keeps what works in existing tech stack, avoids rip-and-replace, and stays agile with focus on outcomes.
TL;DR
The CDP vendor product category is consolidating; the CDP framework (unify → analyze → activate) endures.
Winners are moving to warehouse-first, composable stacks with three graphs—ID (who), Intent (when), Interest (why)—feeding an always-on decision engine.
The outcome isn’t “more data” but faster, cheaper, better decisions via continuous experimentation and reinforcement learning.
Why “CDP the product” is fading (and why the framework survives)
Buyers shifted budgets to the data warehouse and modern data platforms. Enterprises don’t want another silo—or another team. They want decision velocity on top of Snowflake, Databricks, BigQuery (and, in some cases, an MDM backbone like Reltio). So yes, boxed CDPs decline. But the CDP framework—collect → unify → analyze → activate—becomes the substrate for decision intelligence.
The three buyer camps (be honest about where you are)
AI-Native Builders: Warehouse-first, plugging a pre-built decision layer (e.g., iCustomer) on top. Shipping experiments weekly.
DIY Rebooters: Ripping/morphing rigid CDPs, rebuilding on a data mesh/warehouse with systems integrators. High burn, variable ROI and speed.
Cloud-Suite Campers: MA/CEP stack + light “AI.” Easy to start; hard to scale or escape. Channel-first instead of decision-first.
A 5-Stage CDP → Decision Maturity Model
All-in-One Campaigner (CEP/MAP-heavy)
Use: Batch email/multichannel from siloed tools
Trap: Pretty UIs, ugly data; low decision IQ
Unified Profiles + Segments
Use: 1P unification, self-serve audience sync
Trap: Thin real-time, weak cross-channel learning
ID Graph + Enrichment
Use: Stable identity across devices/channels
Trap: Knows who, not when/why
Composable DW + Multi-Graph (ID + Intent + Interest)
Use: Structured + unstructured context; hybrid martech/adtech orchestration
Value: Decisions consider who/when/why together
Always-On Decisioning + RL
Use: Continuous testing, policy-aware personalization, budget reallocation
Value: Decision velocity → revenue velocity
The Multi-Graph You Actually Need
ID Graph (Who): Resolve identities into durable profiles.
Intent Graph (When): Short-horizon signals (browses, price checks, repeats, abandon).
Interest Graph (Why): Durable passions/preferences (topics, values, communities).
Why it matters: Personalization without why is spam; without when it’s late; without who it’s waste.
The Loop That Wins: Graphs → Decisions → Orchestration → Learning
Centralize & cleanse in your DW; model once, reuse everywhere.
Establish ID/Intent/Interest as first-class data products.
Decide: Next-best-action/offer/experiment with policy guardrails.
Orchestrate across ads, email, SMS, app, web, agents.
Learn: Close the loop with outcomes; let RL re-weight offers, timing, and channels.
Fast case study
Context: Cloud collaboration brand targeting SMBs/prosumers active in niche communities.
Who: Unified handles + reviews + community roles → precise SMB/prosumer profiles.
When: Live threads (“need scalable storage”), hiring/funding triggers → outreach windows.
Why: Motivations (price, simplicity, collaboration) mined from UGC/reviews.
Activation: Timed offers, micro-influencers, AMAs in relevant communities.
Results: ~2.8× sign-ups, ~40% lift trial→paid among engaged cohorts, stronger WOM on G2/Twitter. (Attribute conservatively; link if public.)
What to do in the next 90 days
Assign owners for ID/Intent/Interest data products; define schemas + SLAs.
Stand up real-time signal capture (events + UGC/content processing).
Ship 3 decisioning experiments (offer, timing, channel) with holdouts; report decision latency and lift weekly.
Add policy & governance (PII boundaries, consent, safety rails) before scale.
Kill one low-impact dashboard to fund one high-impact decision loop.
The iCustomer stance
iCustomer implements the multi-graph → decision engine → learning loop directly in your warehouse, with policy guardrails and agentic workflows. The goal isn’t another “single view.” It’s compounding decision lift—measured in conversion, CAC payback, and media-waste reduction.
Close
If your “CDP” hasn’t moved your decision speed or conversion curves, you don’t need more segments—you need a decision layer. Keep the framework. Upgrade the engine.
Reply with your current stage (1–5) and the slowest decision—timing, offer, or channel. I’ll suggest one experiment you can run this week.
Thanks for reading! If you have any questions on building a composable, graph-driven Agentic CDP framework, feel free to reach out.