A decision intelligence platform is a system that makes autonomous, real-time decisions about which action to take, for which customer, through which channel, at which moment — governed by a single, explicit objective function. In B2C, that objective function is predicted customer lifetime value.
This is not a dashboard. It is not a reporting layer. It is not a CDP that unifies data without acting on it, and it is not a marketing automation platform that fires rules without knowing their economic consequence. A decision intelligence platform occupies a specific architectural position: it sits between your data and your execution tools, consuming context from every system in your stack and producing coordinated decisions that every system acts on.
In January 2026, Gartner published its inaugural Magic Quadrant for Decision Intelligence Platforms [1]. In March 2026, Scott Brinker — dubbed the “godfather of martech” by Ad Age — published a 33-page framework with Databricks describing a composable canvas architecture for the AI age, with decisioning as Ring 4 of five concentric capabilities [2]. In April 2026, Brinker and Frans Riemersma published the 2026 State of Marketing Attribution report, declaring Attribution 1.0 dead and positioning attribution as a decisioning discipline rather than a reporting function [3].
Three independent publications. One convergent conclusion: decisioning is now a recognized, named, and budgeted enterprise capability. The category exists. The architecture has been mapped. The question is no longer whether you need a decisioning layer — it is what that layer should optimize for.
I hold a PhD in Physics and an Executive MBA from IE Business School. I taught machine learning and AI at Sofia University for fifteen years. Four years ago I founded Releva, a platform that now processes data for 250+ B2C brands across 15 countries. This article is my analysis of where the industry frameworks agree, where they leave gaps, and what a decision intelligence platform must do to close those gaps in practice.
The ad tech / martech gap is not a technology problem
For over a decade, the industry has discussed the convergence of ad tech and martech. The diagnosis is consistent: ad tech handles paid media and customer acquisition. Martech handles owned channels and customer retention. The two systems operate on different data, different metrics, different teams, and different timelines [4].
The proposed solutions have been consistent too: CDPs to unify the data. Clean rooms to bridge the identity. Marketing clouds to consolidate the tools. According to Forrester, over 90% of enterprises are planning or executing strategies to align their ad tech and martech infrastructures [5].
And yet the gap persists. Why?
Because the problem was never data unification. CDPs solve data unification. The problem is that ad tech and martech optimize for fundamentally different things — and no amount of data sharing changes that.
Ad tech optimizes for the cheapest next acquisition. It operates on 7-day attribution windows, last-click models, and cost-per-acquisition targets. It systematically finds fast converters because those are the conversions that register within the measurement window [6]. Meta’s CPMs increased 20% year-over-year in 2025 across every industry, with no sector spared [7]. Google Shopping CPCs jumped 33.72% [8]. The algorithm gets more expensive and more efficient at finding the same low-value customers.
Martech optimizes for engagement within owned channels. It operates on open rates, click-through rates, send volumes, and automation triggers. Klaviyo optimizes email opens. Braze optimizes in-app engagement. Neither optimizes for the economic outcome of the customer relationship.
The ad tech / martech gap is not a plumbing problem. It is an objective function problem. Both systems make decisions. Neither system makes decisions governed by the same economic goal. The bridge between them is not a data layer — it is a decisioning layer that tells both systems what to optimize for.
What Gartner named — and what it missed
Gartner’s inaugural Magic Quadrant for Decision Intelligence Platforms (January 2026) validates that decisioning is an enterprise capability worthy of its own category [1]. The MQ populated the category with vendors whose center of gravity is governance-oriented, single-event decisioning for regulated industries: SAS brings statistical optimization and rule-based decisioning for banking. FICO brings credit scoring and compliance. Aera Technology brings supply chain decisioning.
These are serious platforms solving real problems. But none of them solve continuous, autonomous, value-trajectory optimization for consumer relationships. The distinction matters:
- Banking decisioning: Should we approve this loan? (Single event, binary outcome, regulatory constraint.)
- B2C decisioning: What is the optimal next action for this customer, across all channels, to maximize their predicted lifetime value? (Continuous, multi-channel, value-trajectory optimization.)
The Gartner DIP MQ named the category correctly. It populated the category with the wrong vendors for B2C. This creates a window: the enterprise buyer now has a name for what they need (“decision intelligence”), budget allocated for it, and no vendor in the MQ that does what they actually require for consumer relationships.
Meanwhile, the Gartner CDP MQ (also January 2026) documented a bifurcation thesis: CDPs are splitting into platformization (becoming broader suites) versus agentification (embedding AI agents) [9]. CDPs moving toward agentification need an objective function to govern what their agents optimize toward. Without one, agentic CDPs are automation without direction — agents that act without knowing whether their actions increase or decrease customer value.
What Brinker mapped — and what he left underdeveloped
Scott Brinker’s March 2026 report with Databricks — “The New Martech Stack for the AI Age” — retired the stack metaphor and introduced the composable canvas: a five-ring architecture where a unified data foundation replaces integration plumbing [2].
The five rings:
- Data Core — unified data foundation (customer, company, content, code, control data)
- Semantic Layer — shared definitions, taxonomies, context graphs
- Context-as-a-Service (CaaS) — platforms that serve enriched context to applications
- Decisioning — AI decisioning engines, reinforcement learning, orchestration
- Apps & Agents — the outer ring of commercial, custom, and ephemeral applications
The framework is architecturally sound. MarTech Square’s independent review calls it “the most serious attempt in years” to move beyond the rigid stack model [10]. Brinker himself notes that decisioning “could be more powerful as an independent service — a consumer of context, rather than a provider” and warns that when “decisioning logic lives inside individual channels and agents, it fragments — replacing integration debt with decisioning debt” [2].
But across 33 pages, Ring 4 — Decisioning — receives approximately two paragraphs of substantive treatment [10]. This is not a criticism of the report’s scope. It is an observation about where the industry’s thinking stands: the architecture has been mapped, but the decisioning discipline within it remains underdeveloped.
MarTech Square’s review names three gaps that need filling regardless of which platform occupies the data core [10]:
- Decisioning governance, not just data governance. Data governance asks: is this data accurate? Decisioning governance asks: is this decision correct, explainable, and aligned with the economic objective?
- Decision tracing as a first-class capability. Brinker’s “context graph” — a record of why decisions were made, not just what happened — deserves to be a headline architectural principle, not a footnote.
- Decisioning as a standalone service. When decisioning is embedded inside individual apps and agents, it fragments. The composable canvas needs a composable decisioning core.
The gap is not in Brinker’s vision. The gap is in implementation: who provides the decisioning layer, and what does it optimize for?
What Attribution 2.0 reveals — and where it stops
Brinker and Riemersma’s 2026 State of Marketing Attribution report completes the picture [3]. Their central thesis: Attribution 1.0 — the era of last-touch scorekeeping, dashboard-driven reporting, and credit allocation — is dead. Attribution 2.0 replaces it with a decisioning discipline:
- Conversion tracking → Revenue tracking. Stop counting events. Start measuring value.
- Campaign-centric → Customer journey-centric. Stop optimizing campaigns. Start optimizing customer trajectories.
- Reporting → Decision support. Stop building dashboards. Start making decisions.
- Siloed truths → Shared language. No shared language across teams means no boardroom impact.
The report introduces several concepts that map directly to what a decision intelligence platform must do:
Revenue forensics. Who are your most profitable customers? What do they buy? Where are the margins? Nine out of ten marketers cannot answer these questions — and that gap is the entire problem.
The control tower metaphor. Different teams with different dashboards, aligned on shared navigational signals. This is not a reporting system. It is a coordination system — precisely what a decisioning layer provides.
Signal quality over signal volume. The report argues for “data depth over data debt” — accepting that data is never complete, but ensuring that the data you do have is architecturally sound. This validates server-side tracking (100% capture) over pixel-based collection (60-70% capture).
3-5 decisive moments. Fewer triggers matter than expected in profitable journeys. The lowest-cost action to advance each customer’s behavioral state is the decisioning question — not “which campaign gets credit.”
Where the report stops: it describes decision support. It does not describe decision autonomy. Attribution 2.0 is framed as informing human decisions. A decision intelligence platform goes further — it makes the decision, executes it, measures the outcome, and learns from the result. The human governs the objective function and the constraints. The platform handles the continuous optimization.
Loyalty as context-as-a-service: the missing signal
In March 2026, Brinker published a separate analysis asking whether loyalty systems could be “killer context-as-a-service (CaaS) platforms” [11]. His argument: loyalty programs sit on a unique intersection of behavioral data, transactional data, and declared preferences that no other system in the stack possesses. This data should flow outward to every other system — ad platforms, personalization engines, communication tools — as enriched context.
The insight is correct but incomplete. Loyalty data as context is valuable. But context without decisioning is passive. The question is not “can my loyalty data enrich my ad targeting?” — the answer is obviously yes. The question is: “does my system know that this specific loyalty member has a predicted CLV of €3,200, is 15 days past their predicted repurchase window, responds best to push notifications at 7pm, and should receive a category-expansion recommendation rather than a discount?”
That is a decisioning question, not a context question. The loyalty data provides the signal. The predictive CLV model provides the objective function. The decisioning layer determines the action. The loyalty engine calibrates the reward. Each component is necessary. None is sufficient alone.
What a B2C decision intelligence platform actually does
Based on the convergence across Gartner, Brinker, and the Attribution 2.0 framework — and based on four years of building and deploying one — here is what a B2C decision intelligence platform must do:
1. Define and enforce a single objective function. Every system in the stack must optimize toward the same economic goal: predicted customer lifetime value. Not opens. Not clicks. Not 7-day ROAS. Not rules fired. One number, accessible to every system in real time, governing every decision.
2. Bridge ad tech and martech architecturally. The platform sends enriched signals to ad platforms (PredictedValue events via server-side CAPI) so acquisition optimizes for lifetime value, not transaction value. It simultaneously orchestrates owned channels (email, SMS, push, on-site personalization, product recommendations) so retention compounds the same objective. One objective function. Both sides of the stack.
3. Operate as a standalone decisioning service, not embedded logic. Following Brinker’s own warning: when decisioning fragments across apps and agents, it creates decisioning debt. The platform consumes context from every system and produces coordinated decisions that every system acts on. It is a consumer of context, not a provider — exactly the architectural role Brinker describes for Ring 4.
4. Capture 100% of signals — not the 60-70% your pixel sees. ATT opt-in rates average 35% globally [12]. Ad blockers block 25-30% of pixel fires [13]. Meta attribution has deteriorated 40-60% [13]. A decisioning platform built on incomplete data makes incomplete decisions. Server-side tracking at the integration layer is not optional — it is the foundation.
5. Support autonomous operation with human governance. The autonomy ladder: at the bottom, the human makes every decision manually. At the top, the system operates autonomously within governed constraints. Most enterprises start at constrained autonomy — the system recommends, the human approves — and graduate to self-optimizing as trust builds. The objective function and the constraints are always human-defined. The continuous optimization is always machine-executed.
6. Transfer intelligence across markets and verticals. A bidirectional ontology maps brand-specific data onto abstract behavioral patterns that transfer across geographies and product categories. New market deployments take days, not months. The behavioral patterns — purchase frequency, category expansion, price sensitivity, channel responsiveness — are structurally similar whether you are selling car accessories in Malaysia or fashion in Romania.
The five structural blind spots this architecture closes
When you deploy a decision intelligence platform with a shared objective function, five structural failures in the stack resolve simultaneously:
- The invisible segment — server-side tracking recovers the 30-40% of paid traffic your pixel misses.
- The untrackable journey — three-tier identity resolution connects anonymous browsing sessions to identified purchases across devices and time.
- The unexplained drop-off — diagnostic layer maps churn to specific causes (invisible, wrong timing, wrong channel, over-communicated) instead of treating it as one aggregate number.
- The unpredictable value — predicted CLV replaces backward-looking ROAS as the governing metric for every decision.
- The unmodelable growth — ontology transfer enables new markets in days rather than months.
For the full diagnostic — including self-tests you can run on your own data — see our mid-market five blind spots guide (for brands using Klaviyo and standard ecommerce stacks) or our enterprise structural failures analysis (for brands running MoEngage, Braze, SAS, Segment, or Dynamic Yield).
The economics
The average enterprise B2C stack — CDP, engagement platform, personalization engine, recommendation system, analytics team, data engineering, agency fees — costs $930K to $3.4M per year [14]. Customer acquisition costs have risen 222% over eight years [15], with a further 18.4% increase in 2025 alone. Ecommerce brands lose an average of $29 on every new customer they acquire [16]. Meta CPMs increased 20% year-over-year in 2025 across all industries [7]. 42% of marketing budgets are wasted on customer acquisition [17].
A decision intelligence platform deployed as a layer on top of the existing stack costs $100-200K per year for enterprise. Deployed as a full replacement for engagement, personalization, recommendation, and orchestration tools, the same $100-200K replaces $730K-3.2M in fragmented tooling.
The proof: Carsome, Southeast Asia’s largest integrated car marketplace ($1.7B valuation), deployed Releva alongside their existing MoEngage + Segment + Dynamic Yield stack. Within 90 days, Releva-attributed revenue reached MYR 36.8M per month — 82× the annual platform cost. Email open rates went from 1.2% to 18%. Click rates from 6.1% to 36%. The 45 workflows previously managed in MoEngage migrated in one week. See more case studies.
Book a diagnostic to see what the five blind spots look like in your own data.
FAQ
What is a decision intelligence platform? A decision intelligence platform is a system that makes autonomous, real-time decisions about which action to take, for which customer, through which channel, at which moment — governed by an explicit objective function. In B2C, that objective function is predicted customer lifetime value. It sits between your data and your execution tools, consuming context and producing coordinated decisions.
What is the difference between a decision intelligence platform and a CDP? A CDP unifies customer data into profiles. It answers “who is this customer?” A decision intelligence platform takes that unified data and makes autonomous decisions: what to do next, for whom, through which channel, toward which outcome. A CDP is a data layer. A decision intelligence platform is a decision layer.
What is the difference between a decision intelligence platform and a marketing automation platform? Marketing automation platforms fire rules: if customer did X, then do Y. A decision intelligence platform optimizes for predicted outcomes: this customer’s predicted CLV is €3,200, the lowest-cost action to increase it is Z, delivered via push at 7pm. Rules are backward-looking. Prediction is forward-looking. The difference is autonomous optimization versus manual orchestration.
How does a decision intelligence platform bridge ad tech and martech? By sending predicted customer lifetime value as a signal to ad platforms (via server-side CAPI) so acquisition optimizes for value, not transactions — while simultaneously orchestrating owned channels (email, SMS, push, on-site) so retention compounds the same objective. One objective function governs both sides of the stack.
Where does a decision intelligence platform sit in Brinker’s composable canvas? Ring 4 — Decisioning. Brinker describes this as “AI decisioning engines, reinforcement learning, orchestration.” He notes it “could be more powerful as an independent service — a consumer of context, rather than a provider.” A decision intelligence platform is precisely this: a standalone service that consumes context from the semantic layer and CaaS platforms, and produces coordinated decisions that apps and agents execute.
Does Gartner have a Magic Quadrant for decision intelligence? Yes. Gartner published its inaugural Magic Quadrant for Decision Intelligence Platforms in January 2026. The current MQ is populated primarily with vendors focused on regulated-industry decisioning (banking, insurance, supply chain). The B2C consumer-relationship decisioning category is nascent — which is both the gap and the opportunity.
How long does deployment take? 14 days for integration alongside existing tools. Day 14: diagnostic on your own data. Day 30: first measurable results. Day 90: full ROI quantified. Existing tools keep running throughout. No migration required to start.
References
[1] Gartner (2026). Magic Quadrant for Decision Intelligence Platforms. Pidsley, Idoine, Herschel, Quinn, Carlsson. January 26, 2026.
[2] Brinker, S. (2026). The New Martech “Stack” for the AI Age. Produced with Databricks. March 2026. https://www.databricks.com/resources/ebook/new-martech-stack-ai-age
[3] Brinker, S. & Riemersma, F. (2026). 2026 State of Marketing Attribution Report. “Attribution 1.0 is dead. Attribution 2.0 is about direction, not credit.”
[4] MarTech.org (2026). “Martech, Adtech, and Sales Tech: Are They Converging?” https://martech.org/martech-adtech-and-sales-tech-are-they-converging-and-should-they/
[5] Forrester (2025). “Over 90% of enterprises are planning or executing strategies to align their MarTech and AdTech infrastructures.”
[6] DOJO AI (2026). Meta Ads Attribution in 2026. “Attribution accuracy deteriorated 40-60% over 18 months.” https://www.dojoai.com/blog/meta-ads-attribution-2026-changes-fixes
[7] Triple Whale (2026). Facebook Ads Benchmarks 2025. “Meta CPMs increased 20% YoY across all industries. 35,000+ ad accounts analyzed.” https://www.triplewhale.com/blog/facebook-ads-benchmarks
[8] WordStream (2025). “Google Ads CPC +12.88% YoY. Shopping ads CPC +33.72%. Overall ROAS declined 10.03%.” https://www.wordstream.com/blog/facebook-ads-benchmarks-2025
[9] Gartner (2026). Magic Quadrant for Customer Data Platforms. Foo Kune, Dooley, White, Bloom, Brosnan. January 26, 2026.
[10] MarTech Square (2026). “The MarTech Canvas: New Composable Architecture Standard.” March 22, 2026. https://martechsquare.substack.com/p/the-martech-canvas-new-composable
[11] Brinker, S. (2026). “Could loyalty systems be killer context-as-a-service (CaaS) platforms?” Chiefmartec Newsletter. March 6, 2026. https://newsletter.chiefmartec.com/p/could-loyalty-systems-be-killer-context-as-a-service-caas-platforms
[12] Adjust (2025). ATT Opt-In Rates: 2025 Data & Benchmarks. “Industry-wide average opt-in rate: 35% as of Q2 2025.” https://www.adjust.com/blog/att-opt-in-rates-2025/
[13] DOJO AI (2026). “25-30% of web users run ad blockers blocking Meta Pixel. Meta attribution deteriorated 40-60% over 18 months.”
[14] Releva Enterprise Meeting 2 Deck. Enterprise stack costs: SAS/Pega $200K-1M+, CDP $50-200K, Engagement $50-300K, Recommendation $30-100K, Analytics team $200-600K, Data engineering $300-800K, Agency $100-400K.
[15] Profitwell (2026). Customer Acquisition Cost Benchmarking Report. “222% eight-year CAC increase. 18.4% YoY rise in 2025.” https://www.paddle.com/
[16] GrowSurf / SimplicityDX. “Ecommerce brands lose an average of $29 on every new customer acquired.”
[17] Deliberate Directions (2026). “42% of marketing budgets are wasted on customer acquisition.” https://deliberatedirections.com/
[18] Brinker, S. (2026). “Stacks on a Plane: Reshaping martech on a universal data layer.” Chiefmartec Newsletter. March 2026. https://newsletter.chiefmartec.com/p/stacks-on-a-plane-reshaping-martech-on-a-universal-data-layer
[19] Avenga (2026). “MarTech and AdTech trends to watch in 2026.” “More companies are choosing to build and control the essential layers — data, decisioning, delivery, and measurement.” https://www.avenga.com/magazine/adtech-trends-to-watch/
[20] Bain & Company / Reichheld, F. “A 5% increase in customer retention can increase profits by 25-95%.” https://hbr.org/2014/10/the-value-of-keeping-the-right-customers
[21] Stouse, M. (2026). “The Data Refinery Problem: Why Data-First Architectures Are Failing on Their Own Terms.” “Gartner predicts that through 2026, organizations will abandon 60% of AI projects for lack of AI-ready data.” https://markstouse.substack.com/p/the-data-refinery-problem-why-data
[22] Chan, D. (2026). “Re: The New Martech Stack for the AI Age.” “What’s being framed as a new paradigm may simply be a more flexible expression of existing architecture.” https://medium.com/@iamdavidchan/re-the-new-martech-stack-for-the-ai-age-b0e62a906b5a
